Designing A/D Converters for the James Webb Space Telescope

The James Webb Space Telescope is a solar-orbiting infrared observatory that will complement and extend the discoveries of the earth-orbiting Hubble Space Telescope.

The James Webb Space Telescope (JWST) has four infrared cameras to view the stars. One of the most critical components in the signal path that converts starlight infrared analog signals to digital signals, for subsequent image processing, is an analog-to-digital (A/D) converter. Three of the four IR cameras use arrays of A/D converters designed by a single individual, Dr. Lanny Lewyn. Those three cameras are the Near Infrared Camera (NIRCam), Near Infrared Spectrograph (NIRSpec), and the Fine Guidance Sensors (FGS). The A/D performance in the NIRCam is mission-critical, because the NIRCam is used to perform the precision alignment of the 18 primary mirror segments.

Figure 1. Loose, Lewyn, et. al. ‘SIDECAR Low-power control ASIC for focal plane arrays including A/D conversion and bias generation.’ Proc. SPIE, 2003, pp. 782 – 794

The infrared light gathered by the large primary mirrors is reflected by a secondary mirror into an array of photodetectors in each IR camera. The faint light from these images causes packets of electron currents to be generated, which are then converted into very small voltage signals within each photodetector. These signals are pre-amplified and routed to an array of A/D converters. They must produce a corresponding digital output having a noise level corresponding to less than 5 rms electrons at the photo detectors.

The A/D outputs are formatted on-chip in the SIDECAR™ Teledyne application-specific integrated circuit (ASIC), then routed to the Integrated Science Instrument Module (ISIM) data system. The ISIM then combines the data from all cameras and sends them to Earth using a digital radio link, which operates like the communication between cell towers and your smart phone. A/D conversion is required for every one of the millions of pixels in the near-infrared cameras. All of the JWST’s 28.6 GB of science data will be transmitted from the JWST high-gain antenna twice a day.

The JWST Mission

Figure 2. The fate of the universe is predicted from different cosmological theories as ending in an ultimate collapse, or ‘Big Crunch’ (yellow curve) where the attraction force of dark matter(Ωm) greatly exceeds the expansion force of the dark energy (Ωv). Where the converse is true (red curve), a ‘Big Rip’ occurs and acceleration continues until the fundamental particles of matter are ripped apart. If the force of both dark matter and dark energy are weak, the universe gradually drifts slowly apart and the end is the ‘Big Chill’ (blue and green curves) where objects continue to drift apart endlessly (Source: NASA GSFC).

The James Webb Space Telescope is a solar-orbiting infrared observatory that will complement and extend the discoveries of the earth-orbiting Hubble Space Telescope, with longer wavelength coverage and greatly improved sensitivity. The much larger mirror (6.5 m vs. 2.4 m) and longer wavelengths enable Webb to look significantly closer to the beginning of time (a.k.a. the Big Bang), and to hunt for the formation of yet-unobserved first suns and galaxies over 10 billion years ago. Those suns were huge, 30 – 300 times larger than our sun, 1 and millions of times brighter. As time in the universe goes, they were short-lived, existing for only a few million years before exploding into Super Novae. The hope is that these images will help us better understand the physics of this evolutionary process.

Understanding more about how our universe began may also offer significant clues to how it will end. By looking back through extreme distances (and time), we will increase our understanding of dark energy and dark matter. Dark energy is what pushes the universe to expand. We can’t ‘see’ dark matter, but whatever it is, it accounts for an unseen mass within our galaxy. We know it is there, because the gravitational pull from that extra mass is necessary to keep the outer star systems in our galaxy from flying away, given their apparent angular velocity. Figure 2 shows how the universe might end; in a ‘Big Crunch,’ or a ‘Big Chill,’ depending on the balance between dark matter and dark energy.

The design of JWST’s electronics was well under way by 2005 and the cameras were delivered by the prime contractors (Lockheed Martin and the University of Arizona) in 2012. The total cost of the JWST is over $10 billion, involved three space agencies, and over 1000 people. One of them was Dr. Lanny Lewyn.

When I was President and co-founder of Snowbush Microelectronics, I was approached by Len Kozlowski from Rockwell who asked us to design an A/D converter. After reviewing the specifications and schedule, I told Len I couldn’t do it, at least not successfully. This was difficult to say, as designing A/D converters is what most analog designers prefer. I told Len his only possibility for meeting his specifications and schedule was to subcontract the required A/D converter design to Dr. Lanny Lewyn.

Figure 3. The A/D differential nonlinearity (DNL) requirement of +/- 0.3 LSB required use of a new algorithm for matching MSB and LSB DAC voltage boundaries, while the integral nonlinearity (INL) requirement (without onboard calibration) of +/- 2 LSB required a combination of two centroiding algorithms and extreme lithographic uniformity. (Source: Loose, et. al. in “The SIDECAR ASIC” SPIE 2005)

Lanny and I became friends shortly after I became a Professor at UCLA (1980 – 1991). As electronics professors, we were encouraged to work actively one day per week with industry. Many times, I worked directly with Dr. Lewyn on a wide variety of projects. When Professor David Johns and I co-founded Snowbush Electronics in 1998, we often subcontracted design work to Dr. Lewyn’s company (LCI), especially for A/D converter projects, so we knew well what his capabilities were.

Design Challenges for The Nircam A/D

Both the schedule and the specifications required for the NIRCam SIDECAR ASIC were demanding. An array of 36 A/ Ds was required to fit within a specific IC width limit. The power limit was 1.5 mW at 100 Kb/s. They also must have 16-bit +/- 2.5 LSB INL accuracy and satisfy a special imaging requirement of +/- 0.3 LSB DNL. A 10× speed mode was required, but at a lower resolution (still way better than anything else available given the power increase limits).

The ASIC design program lead and system architect was Marcus Loose at Rockwell. Marcus and Dr. Lewyn worked directly together on a single-chip solution. For the A/Ds and imager readout control circuits, Dr. Lewyn did both the A/D transistor-level circuit design and the analog layout as well; in a typical development, this amount of work might be done by a team of four/five engineers.

To meet the demanding schedule, he used the dimensionless design methodology (Gamma rules) that he developed for the Pair-Gain 14b HDSL A/D. Dimensionless layout design rules can be ported to multiple technology scales and foundries. It was a significant improvement over an earlier dimensionless circuit design and layout approach (Lambda rules) created by his advisor at Caltech, Carver Mead. 2

A successive approximation register (SAR) was devised, which used a precision capacitor array for the most-significant-bits (MSBs) and a precision resistive divider for the least-significant-bits (LSBs). To comply with the severe differential nonlinearity (DNL) requirements (usually +/-0.5 LSB), an algorithmic method was used for perfectly matching the voltage switching boundaries of the MSB capacitors to those of the lower LSB resistors. For clarity, the +/- 0.3 LSB DNL requirement specifies a random error of only one part in 218,453. 3

No autocalibration methods were permitted to correct for small A/D linearity errors because of the low power requirement and the non-random nature of the signals. The INL requirements had to be satisfied by using a combination of centroiding algorithms that had been previously used by Dr. Lewyn to supply Toshiba with an 18 b audio DAC with an INL of +/- 1 LSB. One of those algorithms, a perfect centroiding algorithm, involved duplication and rotation of elements rather than the side-by-side transistor method, invented several years earlier by Kiyoshi Kanekawa of Toshiba.

The Second Lagrange Point (L2)

Figure 4. The orbit of the JWST around the Second Lagrange Point (L2) is quasi-stable, requiring a small amount of fuel to maintain position. The orbit around L2 permits the JWST batteries to recharge by the sun, when out of the earth’s shadow. As a result of a highly accurate orbital insertion maneuver, the amount of fuel remaining may sustain observational accuracy for as many as 20 years, not the initial 10-year plan. (Source: NASA GSFC)

So where is the JWST now? It is orbiting at the Second Lagrange Point, or L2. Since the photodetectors and camera electronics, operating below –200° C on the cold side of the telescope system, must be shielded from the sun, the L2 solar orbit (Figure 4) was chosen, rather than a near-Earth orbit like the Hubble Space Telescope (HST), because the heat shield must always be interposed between the sun and the telescope when receiving images. In a low orbit, the Earth would block many observations for about half a day.

The distance of L2 is approximately a million miles from Earth, and 93 million miles from the sun. For comparison, the moon is only 238,855 miles from the earth.

The NIRCam

Figure 5. A single NIRCam sensor head is mounted on the Dewar for cryogenic temperature testing at 37 K, which is -236 °C. In the NIRCam a SIDECAR ASIC is mounted directly behind each of the 4 Mpix photodetectors. (Source: U. Arizona)
Figure 6. NIRCam imaging covers the two adjacent fields of total area 9.7 arcmin. The short wavelengths use four detectors (blue area) in each of the modules while long wavelengths are covered by two detectors (red area). (Source: Space Telescope Science Institute (STScI)).

The Near-Infrared Camera (NIRCam) is JWST’s primary imager in the wavelength range from 0.6 μm to 5 μm. It consists of two, nearly identical, fully redundant modules (Figures 5 and 6), which point to adjacent fields of view in the sky and can be used simultaneously. Each module uses a dichroic to also observe simultaneously in both the short wavelength channel (0.6 μm–2.3 μm) and long wavelength channel (2.4 μm–5.0 μm).

The first images from the James Webb Space Telescope released by NASA show the single sun-like star HD 84406 in the Ursa-Major Galaxy as seen from each of Webb’s 18 primary mirror segments. At the time, those segments were not in alignment with each other, so the image effectively shows a single star from 18 different angles. The challenge was to align each mirror segment so that the star it sees coincides with the stars seen from the other mirrors to become a perfectly overlapping, single-star image.

The A/D and the Hubble Space Telescope (Hst)

Must we wait for the JWST telescope to be calibrated before we know if Dr. Lewyn’s A/Ds work? No! While construction of the JWST was underway, a failure occurred in the electronics system for the Hubble Advanced Camera for Surveys (ACS) on January 27, 2007. The SIDECAR ASIC was selected to replace the original image processing system on the ACS, using Space Shuttle STS-125 HST Servicing Mission SM-4 in May 2009. This mission was successful, and now the x36 A/D array in SIDECAR currently processes all the image content from the 2 working channels of the ACS imager with lower noise and higher image fidelity than the previous electronics. The SIDECAR A/D array was also selected for the European Space Agency Euclid mission that is being launched in 2022 to map the geometry of the Dark Universe.

The A/D and Very Large (6-10m) Ground-Based Telescopes

As a result of the excellent overall performance of the SIDECAR signal processing system, it is currently deployed, or in the planning stage, for use with a new class of several large (6 – 10m) ground-based telescopes. In each camera application, it is paired with Teledyne HAWAII-2RG 4 MP detectors. Several use an array of separate, guidable optical elements to reduce the effects of atmospheric distortion in order to begin the direct exploration of exoplanets. Most of them are in Chile, including the Gemini South 8.1m telescope on Cerro Pachon, the University of Tokyo Atacama Observatory (TAO) 6.5 m telescope on Chajantor, and the Magellan 6.5 m telescope pair (Baade and Clay) on Cerro Las Campanas. Also, two other large telescopes are in Mauna Kea, Hawaii: the Subaru 8.2 m telescope and the Keck 1 10 m telescope.

The Next-Generation A/D

When Dr. Lewyn was approached by NASA to provide an A/D array for the next generation Nancy Grace Roman Space Telescope (‘The Roman’), he realized that his current architecture could not provide a 10x increase in speed while maintaining the same 1.5 mW power level as the JWST SAR-architecture A/D. He needed to start again.

Figure 7. The first published image taken by the James Webb Space Telescope shows a mosaic image created by a single sun-like star (HD84406) in the Ursa Major Galaxy, about 160 million miles away. It was created over 25 hours beginning on Feb. 2, 2022. It appears as many separate images because the 18 JWST mirrors were just beginning the alignment process. (Source: NASA)

Moving past the Roman requirement, he began a multi-year effort to design a next-generation A/D (Nex-Gen ADC) that was based on a pipeline architecture. A compelling reason for choosing a pipelined architecture is that the power efficiency of the high-gain op amp required for a SAR architecture is poor. While op amp power efficiency is optimum at a gain of 3, a pipeline A/D speed can be fastest when choosing a gain of 2, rather than 4. On the other hand, to achieve low INL and DNL, more innovative architecture, matching, and centroiding concepts were required for the capacitor weighting network arrays.

To take advantage of much higher speeds using advanced planar technology nodes, Dr. Lewyn updated his dimensionless layout design rule set. Improving lithographic fidelity, device matching, and eliminating odd-cycle errors using two-color decomposition were some of the objectives of the rule-set changes. A new basic device footprint and additional ‘correct-by- construction features’ were added to minimize systematic errors that are undetectable by simulation. Techniques similar to those used in sensitive linear systems to provide substrate coupling isolation improved suitability for DoD applications that require radiation-resistant (rad-hard) technology.

The Nex-Gen 16b A/D is currently being ported into a multi-project-wafer (MPW) test chip system that will verify the design parameters of interest in two different basic versions. On this project, Dr. Lewyn is working co-operatively with IQ-Analog (a San Diego Microelectronics Company) using an advanced planar CMOS technology. This technology is ten-times more advanced than what was available at design time for the JWST. The new A/D is also highly programmable so it can be optimized for a wide variety of applications without redesign. The narrow form-factor will permit its incorporation in a x32 array for multi-channel imaging, or alternatively, in multi-slice ultra-high-speed applications.

This article was written by Kenneth Martin President, Granite SemiCom Inc. (Toronto, Canada). For more information, visit here .

References
  1. The sun is over 300,000 times heavier than the earth and contains 99.8 percent of the total mass of the solar system.
  2. Professor Mead’s text book , Introduction to VLSI Systems, co-authored with Lynn Conway, is widely credited for starting the VLSI Revolution. (Source: Wikipedia).
  3. A 16-bit A/D converter outputs digital signals corresponding to 65,536 different levels of signal strength.