3.1.

Automated Literature Search

An automated search of the astronomical literature was done to assess the direct impact of ATI awards. Because investigators are required to acknowledge their grants in publications that follow from supported research, it is possible to connect specific research products to particular awards using publicly available literature repositories.

The sample for the automated search was selected from publicly available data for the ATI program that was available on or about October 4, 2017. Public records for each award consisted of 25 fields of data with information such as NSF award number, managing organization (NSF Division), PI, title, awarded amount to date, start and end dates, program element code(s), award instrument (standard grant, continuing grant, cooperative agreement, or interagency agreement), and abstract as well as other data. The nsf.gov award search page was used to download all public information from awards with program element code 1218 from January 1, 1987, through December 31, 2016. This search yielded results for 582 awards. Many awards receive funding from multiple programs within an NSF Division and sometimes across NSF Divisions. 23 awards were removed from the sample due to their being managed by another division outside of the Division of Astronomical Sciences (denoted as AST in the NSF Organization field). To select principally ATI awards, awards were removed from the sample that had other program element codes listed first in the case of multiple program element codes or that were obviously not ATI projects based upon the title, abstract, or other information (e.g., title included MRI designation for Major Research Instrumentation program). The final sample for automated analysis consisted of 496 awards.

Each award was searched in the astronomical literature using the Astrophysics Data System (ADS). Code was written in Python to retrieve records from the database for use as input ADS search criteria. A publicly available interface was used to interact with ADS.⁵ For each award record, ADS was searched for papers with the following criteria: (1) the award ID in the full text of the paper, (2) the PI name as an author, and (3) publication date range specified by the award effective date and 2017.

The search was restricted to peer-reviewed publications. The decision to exclude conference presentations, even though many instrument papers are published in nonrefereed proceedings, was chosen to make the results of the automated search conservative, but more robust. Although some conference proceedings include published manuscripts that allow authors to acknowledge awards, some (notably including American Astronomical Society meetings) include only abstracts that do not typically contain acknowledgments. Thus leaving out conference proceedings in the automated analysis allowed for a straightforward apples-to-apples comparison between instrumentation and pure science papers. However, as presented extensively in Sec. 4, many important instrumentation results were presented in conference proceedings. A logical extension of this work would be to judiciously include certain conference proceedings that publish complete manuscripts. Such inclusion would undoubtedly favor instrumentation results, which are frequently published in the Proceedings of the Society of Photo-optical Instrumentation Engineers (SPIE). The number of acknowledging papers as well as their number of citations of each acknowledging paper was saved for each award. These search results were stored in a PostgreSQL database for analysis.

3.2.

Award Acknowledgments

Table 1 illustrates the most widely acknowledged awards and the most highly cited paper among those acknowledging awards. The most widely acknowledged ATI award, 0906060/Baranec supported deployment of a micromachined deformable mirror for the AO system on the Robo-AO robotic observatory. This award was acknowledged 31 times in the peer-reviewed literature. The most highly cited paper from this award was cited 79 times.⁶ The other two most widely acknowledged awards 0705139/Ge and 1006676/Mahadevan supported technology for studying exoplanets. These ATI awards have been broadly aligned with the objectives of the most recent astronomy and astrophysics decadal review, for which “seeking nearby, habitable planets” was one of three main science objectives.⁷

Table 1

Three most widely acknowledged ATI awards. Column (1) is the NSF award ID. Column (2) is the PI of the award. Column (3) is the number of acknowledgments found in the automated search. Column (4) is the ADS bib code for the most highly cited paper among the acknowledging publications for this award. Column (5) is the number of citations that the most cited publication has in ADS.

Figure 5 shows the distribution of acknowledgments for the ATI and PLA programs in the peer-reviewed literature. 44% (216/496) of ATI awards are acknowledged at least once in the peer-reviewed literature, compared with 31% (138/445) in PLA. Multiple acknowledgments were found with less frequency, and a handful of awards from each program were acknowledged more than 20 times. Figure 5 shows that ATI and PLA awards are acknowledged with approximately the same frequency in the peer-reviewed astronomical literature despite the fact that ATI awards emphasize technology development, whereas PLA awards are for “pure science.”

Fig. 5

Distribution of acknowledgments of ATI (blue) and PLA (orange) awards. Only awards with at least one acknowledgment are included for clarity. ATI has more acknowledgments than PLA overall, although the difference is not statistically significant (KS test, $p$-value 0.76).

Figure 6 shows award amounts over the study period for both programs, where the symbol size corresponds to the number of acknowledgments of each award. This figure shows that widely acknowledged awards (large symbols in the figures) have occurred in each program throughout the past 30 years, although there was a notable deficit of acknowledgments in PLA in the decade 2000 to 2010. Figure 6 also illustrates that wide acknowledgment can be found among both large and small awards. There is no strong trend between award amount and acknowledgment.

Fig. 6

Award amount (in thousands of dollars) versus start date for (a) ATI (blue symbols) and (b) PLA (orange symbols). Symbol size corresponds to the number of ADS acknowledgments.

3.3.

Literature Citations

Table 2 shows the five most highly cited papers that acknowledge their awards from the last 30 years of the ATI and PLA programs. The most highly cited ATI paper presents first results of the degree angular scale interferometer (DASI) measurement of the cosmic microwave background angular power spectrum.⁸ The receivers, the dewars, the local oscillator distribution scheme, and the correlator were all designed and developed with support from 9413935/Readhead in building the Cosmic Background Imager (CBI) of which DASI was a copy.

Table 2

The most highly cited papers that acknowledge ATI and PLA awards. Columns (1) and (2) indicate the NSF award ID and PI last name, respectively. Column (3) is the ADS identifier for the paper. Column (4) is the year of publication of the paper. Column (5) is the number of citations. Column (6) is the median number of citations for the year of publication. Column (7) is the impact, defined as citations/median.

In addition to the number of citations and the year of publication for each paper, Table 2 also shows the median number of citations for that year of a paper published in The Astronomical Journal. This median value was obtained from a search of ADS for all papers in The Astronomical Journal for the year in question. Notably, median citations in this journal are in the 20 to 30 range throughout the 1990s and 2000s, but they decline significantly after 2009. Therefore, after 2009 median citations in this journal should not be taken as representative of the entire field (applies to one paper in the table, entry marked with an asterisk).

An impact factor is defined as the number of citations divided by the median number of citations for that year, and it is tabulated in the last column of Table 2. The tabulated impact factor after 2009 is considered unreliable due to uncertainty in the median. The impact factor normalizes for an age effect, whereby older papers have more citations simply because they are older. The distribution of such normalized citation counts has been found to follow the same log-normal distribution across a wide range of scientific disciplines.^9,10 The papers shown here are in the high-impact tail of this distribution.

We compared the impact distribution of ATI papers both with astronomy as a whole and with PLA papers specifically. To establish the impact distribution of the entire astronomy literature, publications from six journals were used including Astronomy and Astrophysics, The Astrophysical Journal, The Astronomical Journal, The Astrophysical Journal Supplement, Monthly Notices of the Royal Astronomical Society, and Publications of the Astronomical Society of the Pacific. All 157,039 articles from 1988 to 2015 from these six publications were used to determine the astronomy literature citation distribution. Citations up to 2017 were used.

To get normalized distributions, the counts of the raw, unbinned distributions were divided by the number of papers published that year. Normalization means that the sum of all $y$ values will be one. The distributions were first normalized to address different productivity levels in different years. Normalization results in it not mattering if one distribution (for some year) had 1000 articles and another 10,000. If the fraction of articles with $x$ citations is the same, these distributions are the same. For each year, the mean and the median of the number of citations were computed. This is to take into account that older publications had more time to accumulate citations.

The distributions of citations for each year were logarithmically binned using 0.1 decade bins. Then, every individual citation distribution was normalized by the median to get the universal distribution. This basically means that the $x$ axis was shifted by log(median), to take into account that older publications had more time to accumulate citations. Finally, these citations were averaged for all years in the dataset. Averaging is done by summing the values of distributions (at different logarithmic bins) and dividing by the number of distributions (i.e., the number of different years). Figure 7 plots the logarithm of the probability distribution of citation versus the logarithm of median-normalized citations. The citation distribution for astronomy is illustrated by the gray symbols in this figure.

Fig. 7

Distribution of impact (median normalized citations) of papers from (a) ATI (blue circles) and (b) PLA (orange circles) programs compared with the aggregate astronomy literature (gray circles). Impact has been found to track a universal distribution across a wide range of scientific disciplines. ATI and PLA have very similar impact distributions despite ATI being focused on technology and instrumentation development in addition to pure science. ATI and PLA both have fewer low impact papers and somewhat more high impact papers than the universal distribution.

Figure 7 shows the citation distributions for papers that acknowledge an ATI award (left panel; blue symbols) and a PLA award (right panel; orange symbols). These distributions are qualitatively similar to the general astronomy distribution (gray symbols) at medium and high impact values (i.e., the curves have a similar functional form for values of logarithm of normalized citations greater than zero). However, there are more high impact papers among ATI and PLA than in the general astronomical literature, i.e., the blue symbols lie above the gray symbols toward the right half of the plots. Furthermore, there are comparably fewer low impact papers (logarithm of normalized citations less than zero) among the ATI and PLA samples than in the general astronomical literature, i.e., the blue symbols are below the gray symbols toward the left side of the plots. Thus, ATI- and PLA-supported research outperforms the general astronomical literature in terms of literature citations. We interpret this effect as a selection bias resulting from the peer review process: proposals that are awarded by NSF have been selected through competition.

As shown in Fig. 7, the impact distributions of ATI and PLA are essentially the same. Technology development and instrumentation papers are no less impactful in the scientific literature than pure science without a technology or instrumentation component. This surprising conclusion countermands the naive intuition that might have supposed that instrument builders devote less time and effort to writing papers because they spend more time in the laboratory. In fact, papers from technology and instrumentation programs are just as impactful as pure science papers.

3.4.

Sources of Uncertainty

Several factors may affect the accuracy of literature acknowledgments and resulting paper citations as well as their interpretation. Papers may acknowledge multiple grants, and it can be impossible to determine solely from published literature how much or how critical the contribution of any one grant was to a larger project. Indeed, in rare cases, grants are acknowledged in papers on subjects that are not obviously aligned with the original grant. Research is a creative process that can lead to unexpected directions; furthermore, progress in one field may enable serendipitous advances in other fields.

Acknowledgments are also subject to certain types of error. A false positive acknowledgment (type I error) would be an instance of the automated search reporting a paper to acknowledge an award when in fact no such acknowledgment exists. No such errors were found among manual checks of several dozen papers selected at random from the search results.

A false negative (type II error) would be an instance of a paper that was in fact derived from grant-supported research not being found by the automated search. There are several possible explanations for such false negatives. A product of research may be published outside of the ADS accessible repository; these publications may be books or journals that are not included within ADS. Alternatively, there may be a missing or incomplete acknowledgment of an NSF award; such delinquent acknowledgments are the most common source of false negatives. They may arise if the award number is not included in the acknowledgment, if the acknowledged award number is incorrect, or if the authors neglect to include any acknowledgment. Due to all of the false negatives, acknowledgments from the automated search are believed to underestimate the true numbers of publications by approximately a factor of two.

4.1.

Charge Coupled Device Cameras

Among the most transformative devices for astronomy in the past half century is the charge coupled device (CCD) camera. CCDs were invented in 1969 by Willard S. Boyle and George E. Smith at Bell Laboratories, for which they were awarded the 2009 Nobel Prize in physics.¹¹ A CCD is a metal oxide semiconductor device that stores charge in well-defined regions of the semiconductor and moves it according to applied voltages. The first CCD imaging device consisted of a one-dimensional scanner that had 8 pixels.¹²

In the 1980s, CCDs brought about a revolution in imaging for astronomy.¹³ By the early 1990s, several companies had produced CCDs that were suitable for scientific applications; they featured low noise, negligible dark current, and high charge-transfer efficiency. However, such devices had low overall quantum efficiency, which was a disadvantage for astronomical applications. Several ATI awards to the University of Arizona (9121801/Lesser, 9500290/Lesser, and 9618760/Lesser) supported postprocessing of commercial detectors to optimize their performance for astronomy. Devices were thinned for backside illumination, backside charging, and antireflection coating. Such techniques quickly matured and became routinely applied to detectors of various formats and manufacturers. A further ATI award (9876630/Lesser) explored the use of a competing technology, complementary metal oxide semiconductor (CMOS) detectors, for astronomical use.

Other ATI awards in the 1980s and 1990s were notable for disseminating CCD cameras to astronomical observatories for research and teaching. Maunakea, Lick, and Lowell as well as many smaller observatories received CCD cameras through ATI awards in this time period. CCD sensors are now the most common sensor used for visible light astronomy.¹⁴

The camera for the Vera C. Rubin Observatory (formerly Large Synoptic Survey Telescope, LSST) includes CCD sensors with combined 3.2 GigaPixels;¹⁵ it represents an improvement in pixel count by more than eight orders of magnitude compared with the original CCD imager. LSST focal plane development was initiated in 2002 with an ATI award (0243144/Tyson) that supported a trade study between CCD and CMOS detectors for LSST. This award was the first NSF investment in what ultimately became the highest-priority for ground-based astronomy, according to the 2010 decadal review. The ATI award came at a critical phase in the growth of the project; without this essential seed funding, the focal plane may not have ever been developed.¹⁶ Subsequent development of the LSST camera was supported by the Department of Energy.

4.2.

Large Optical Telescope Mirrors

The optical telescope, the most iconic symbol of astronomy, has undergone a revolution in design since the 1970s. New approaches to building the primary mirror have enabled transformative growth in the size and hence the light gathering power of the world’s largest telescopes. Three techniques have been particularly successful for making big glass.¹⁷ A bend and polish method of fabricating segments that fit together to form a large mirror was pioneered by Jerry E. Nelson at the Lawrence Berkeley National Laboratories in the 1970s. This method was ultimately used for the twin 10-m telescopes of the W. M. Keck Observatory and is planned for the Thirty Meter Telescope. Thin meniscus mirrors in which the lack of stiffness is dynamically compensated for to maintain their optical figures were developed by commercial interests including Corning, which has a history of making mirrors for spy satellites. Such mirrors were employed on the Gemini and Subaru telescopes and the European Very Large Telescope.

An alternative technique, spin cast mirrors, was pioneered by J. Roger P. Angel at the University of Arizona, and it was supported in part by NSF/ATI. This technique grew out of backyard experiments with a homemade kiln. The successful spin casting method was ultimately pursued in a dedicated laboratory at the University of Arizona; see Fig. 8. ATI award 8911701/Angel provided $5M of early support for this lab; other funders included the Air Force Office of Scientific Research and private institutional funds. This lab provided primary mirrors for the MMT, the Large Binocular Telescope (LBT), and the New Solar Telescope for the Big Bear Solar Observatory and is being used to make mirrors for the 30-m class Giant Magellan Telescope (GMT) that is currently under construction as well as others.

Fig. 8

(a) Large optical mirror fabrication using spin-cast, honeycomb mirrors was developed with early support from ATI. (b) Loading 17.48 tons of E6 borosilicate glass. Inspecting the surface of the mirror after the annealing process and the furnace has been removed. Image credit: Damien Jemison, GMTO, Richard F. Caris Mirror Lab, The University of Arizona.

ATI support also provided early seed funding for an early precursor to GMT. 0138347/Angel provided $1.9M to support the early conceptual design and development including mechanical and optical components.^18,19 The ATI program is conceived as an incubator for potentially transformative technology concepts for ground-based astronomy—relatively small awards to seed ideas that may one day transform astronomy. In these cases, ATI awards have indeed seeded initiatives that have subsequently grown to be large astronomical facilities.

4.3.

Adaptive Optics

Ever since telescopes turned toward the sky, astronomers realized that atmospheric turbulence degrades their images.²⁰ Constrained by the fickle air above, even the most sophisticated ground-based observatory is hobbled to resolution that is not much better than a backyard telescope. This problem seemed insurmountable until 1953 when Horace W. Babcock outlined a solution that came to be known as AO.²¹ The broad outlines were laid bare in his seminal paper: a wavefront sensor detects time-varying optical deformations by locking onto a nearby, bright reference star. Distortions are corrected by an equal and oppositely deforming mirror that operates in closed loop with the wavefront sensor. The method can work for science targets that are fortuitous enough to be nearby a bright reference guide star; this main limitation is effectively one of sky coverage.

The modern application of this technique is known as natural guide star adaptive optics to differentiate it from variants that spawned in the succeeding decades. These variants include laser guide star adaptive optics (LGS-AO), which uses steerable laser beacon(s) to establish a reference for wavefront correction near the science target. Ground layer adaptive optics (GLAO) offers partially improved seeing over a wide field of view by compensating only for turbulence within $\sim 100\mathrm{m}$ above the telescope. Multiconjugate adaptive optics (MCAO) uses multiple wavefront sensors and deformable mirrors to correct turbulence from multiple levels in the atmosphere. Multiobject adaptive optics (MOAO) corrects images of multiple science targets simultaneously. Extreme adaptive optics (ExAO) attains high order correction over small fields of view near a bright reference star to obtain extremely high contrast, especially for exoplanet imaging.

The first demonstrations of AO were done in the 1970s.²² Early research was supported by the Air Force and the Defense Advanced Research Projects Agency for space surveillance applications.²³ However, the first published application of AO to astronomy was the ESO-supported system called COME-ON, which was successfully tested in 1989 at Observatoire de Haute Provence in France;^24,25 it was later upgraded and moved to the 3.6-m ESO observatory at La Silla, Chile.²⁶

In the 1980s early United States development of AO for astronomy continued at NOAO and the University of Hawaii. An innovative approach to wavefront sensing and control, the curvature sensor and bimorph mirror, was developed during this time.²⁷ NSF/ATI award 9319004/Roddier partially supported the deployment of this technology at the Canada France Hawaii (CFHT) telescope at Maunakea and its use to study circumstellar environments and protoplanetary disks.^28,29 This curvature sensing approach was later adopted by the 8-m class Subaru telescope.³⁰

4.3.4.

Multiconjugate adaptive optics

MCAO uses multiple wavefront sensors and multiple deformable mirrors to achieve wide-field correction.⁵⁵ The region of sky over which turbulence can be considered uniform is described by the isoplanatic angle, for which a typical value may be 30” at an observing wavelength of $2\mu \mathrm{m}$. Anisoplanatism introduces nonuniformity to the point spread function across the corrected field, limits the size of the corrected field, and effectively limits the sky coverage of AO. These restrictions motivate the use of multiple LGS and corresponding optics in GLAO, MOAO, and MCAO.

Although it was first proposed in 1975,⁵⁶ the first on-sky demonstration of MCAO was not accomplished until 2008, by the ESO-supported multiconjugate adaptive optics demonstrator on the VLT.⁵⁷ This was followed in 2011 by the Gemini multiconjugate adaptive optics system at the Gemini South telescope, which was the first system to demonstrate MCAO using multiple sodium LGSs.⁵⁸ The 30 years from genesis to on-sky implementation of MCAO illustrate the technical challenges that needed to be overcome to make this idea a reality and the long time that it can take to develop technology for astronomy. These first-generation MCAO systems will pave the way for systems that are planned for the 30-m class Extremely Large Telescope (ELT)s, including the GMT, Thirty Meter Telescope (TMT), and the European ELT.

Recently, MCAO was first demonstrated for solar imaging at the Big Bear Solar Observatory as part of an ATI-supported research program (1710809/Goode).⁵⁹ Figure 9 is a result of this research; it illustrates the use of AO for improving solar observations and of the wider field of correction that is enabled by MCAO. Images were made at a $0.705\mu \mathrm{m}$ wavelength (corresponding to TiO spectral line) and have a ${50}^{″}\times {50}^{″}$ field of view. Using three deformable mirrors, this MCAO system corrects more than three times the field of conventional (single conjugate) AO. This project is a pathfinder for instrumentation intended for the 4-m Daniel K. Inouye Solar Telescope.

Fig. 9

Images of the Sun taken with AO at the Big Bear Solar Observatory. Observations at $0.705\mu \mathrm{m}$ wavelength; ${50}^{″}\times {50}^{″}$ field of view. Images taken on October 27, 2017. (a) Conventional AO. (b) MCAO. Images courtesy of Phil Goode, New Jersey Institute of Technology.

4.4.

High Precision Radial Velocity Measurements of Exoplanets

4.5.

Infrared Detectors

IR detectors were primarily developed for military and commercial applications and then applied to astronomy.⁸⁸ The most successful detectors to date have been based on HgCdTe semiconductor, for which development began in the 1980s. Teledyne imaging sensors (formerly Rockwell Scientific Company) developed processes of HgCdTe growth by liquid phase epitaxial deposition of a Sapphire substrate—the Rockwell designation was designated as a producible alternative to CdZnTe epitaxy since molecular beam epitaxy was not available for astronomy applications. These detectors had high dark current due to the lattice mismatch between the HgCdTe and the Sapphire substrate. Rockwell was not able to measure dark current down to astronomical levels; this was done by the University of Hawaii researchers to a level orders of magnitude lower than the Rockwell measurements.⁸⁹

Independently, Raytheon Vision Systems developed processes of HgCdTe growth on silicon and CdZnTe substrates.⁹⁰ Potential advantages of the latter approach include substantially reduced cost resutling fromusing established silicon processing technology.

Most of IR astronomy would not exist without NASA and NSF support. NASA substantially funded HgCdTe detector development; such detectors were used in the NICMOS and WFC3 instruments aboard the Hubble Space Telescope. Later generation HAWAII 2RG detectors are installed in the James Webb Space Telescope and the European Space Agency will deploy 16 of these detectors on the Euclid spacecraft as well as on ground-based instruments. There may be 30 to 50 HAWAII 2RG detectors deployed at ground-based observatories worldwide, and the NASA WFIRST mission is base lined with 18 HAWAII 4RG-10 arrays.⁸⁹

ATI awardees partnered with the industrial collaborators to characterize devices and tailor them to the needs of astronomy. The largest single ATI award supported the development of the HAWAII 4RG mosaic camera (0804651/Hall; $7M, made with American Recovery & Reinvestment Act funds) in collaboration with Teledyne; see Fig. 10. Results of 0804651/Hall have been reported in conference proceedings^91,92 (excluded from the automated ADS search mentioned above). Similarly, ATI awards 1207827/Figer and 1509716/Figer constituted substantial investments in the competing Raytheon technology. At present, the Raytheon detectors have difficulty reaching sufficiently low dark current.⁹³ However, if further development can meet the dark current specification, Raytheon detectors could be transformative.

Fig. 10

HAWAII 4RG IR detector developed with support from NASA and NSF/ATI. Figure from Ref. 91. Reproduced with permission.

Traditional IR detectors are charge-integrating devices, and they are constrained by read-noise in fast readout applications. As an alternative, linear mode avalanche photodiode arrays have been developed with support from ATI (1106391/Hall) and other funding sources in collaboration with industrial partner Leonardo (formerly Selex ES). The Selex Advanced Photodiode HgCdTe Infrared Array (SAPHIRA) has emerged as a leading detector for near-IR wavefront sensing in AO.^94,95 On-sky tests with a near-IR pyramid wavefront sensor and SAPHIRA device were recently accomplished at the Subaru SCExAO system using a camera supported by NSF⁹⁶ and on the Keck II telescope, as funded by a separate ATI investigation (1611623/Wizinowich).⁵⁰ Low read-noise, fast time response detectors will be crucial for pushing the limits in exoplanet studies.⁹⁷

4.6.

Microwave Kinetic Inductance Detectors

Another promising technology for low read-noise, fast time response detectors are microwave kinetic inductance detectors (MKIDs).⁹⁸ Unlike CCDs or photodiodes, these detectors rely upon thin superconducting films that are arranged in a microwave resonant circuit. Incident photons break Cooper pairs and generate quasiparticles, which change the complex impedance and hence the resonance of the circuit. High quality factor resonant circuits (operating at cryogenic temperatures) can be sensitive to individual photons. Individual detector elements can be made with unique microwave resonance frequencies and multiplexed into large arrays with simple readouts. This technology has broad application in addition to optical-IR astronomical detectors.⁹⁹

The first application of an MKID detector to ground-based, optical-IR astronomy was the array camera for optical to near-IR spectrophotometry.^100,101 Its successor, the DARK-speckle near-infrared energy-resolving superconducting spectrophotometer (DARKNESS) was supported by an ATI award (1308556/Mazin); see Fig. 11; it was recently installed behind the PALM-3000 AO system and the Stellar Double Coronagraph at Palomar Observatory.¹⁰² In subsequent research from this same research group that was supported by NASA, a 20 kilo-pixel MKID array was fabricated for the MKID exoplanet camera (MEC) intended for use at the Subaru telescope.¹⁰³ The sensitivity and fast time response of MKIDs make them amenable to ExAO and speckle imaging/nulling applications, which are the subjects of active ATI supported investigations.

Fig. 11

MKID from the DARKNESS instrument supported by ATI. (a) MKID device mounted in package. (b) Detail image showing several rows of pixels, transmission line, and bondpad for one feedline. (c) Detail of several MKID pixels; densely meandered patches at the top of each pixel are the photosensitive inductors, and the large sparse sections are the interdigitated capacitors used to tune each MKID pixel to a unique resonant frequency. Figure from Ref. 102. © The Astronomical Society of the Pacific. Reproduced by permission of IOP Publishing. All rights reserved.

4.7.

Multiobject Spectroscopy

Multiobject spectroscopy has been developed through ATI awards. The FLoridA mulitobject infrared grism observational spectrometer (FLAMINGOS) system was the first cryogenic, multiobject spectrograph, and it was developed through an ATI award (9731180/Elston). This instrument accomplished the technical challenges of masking the focal plane to allow for the selection of specific objects for study in a manner that is reconfigurable, dispersing the light and finally detecting it, all while operating at cryogenic temperature at the telescope.¹⁰⁴ FLAMINGOS cryostats, as well as adaptation of the optical imaging spectrograph approach to cryogenic IR optical materials, were significant innovations. Selectable slit masks for multiobject target selection at the focal plane were laser-machined and placed in a rapidly thermal-cycled fore-dewar, while the rest of the instrument could be kept cold for long durations. More recent designs use dynamically reconfigurable slit units.

Although the particular approach of a selectable machined mask used in FLAMINGOS was not widely adopted, this instrument served as a proof of concept. It demonstrated that such a complex system could be made to work at a telescope, and it undoubtedly inspired other efforts. Although this award was acknowledged in the literature only five times, it nevertheless had a profound effect. Subsequent multiobject spectrographs include the multiobject infrared camera and spectrograph for the Subaru telescope,¹⁰⁵ FLAMINGOS-2 on Gemini,¹⁰⁶ EMIR on Gran Telescopio Canarias,¹⁰⁷ multi-object spectrometer for infra-red exploration on Keck,¹⁰⁸ MMT and Magellan infrared spectrograph that was based explicitly on the FLAMINGOS design,¹⁰⁹ and the LBT NIR-spectroscopic utility with camera and integral-field unit for extragalactic research on the LBT.¹¹⁰ Also notable is the K-band multiobject spectrograph (KMOS) on the VLT, which is really a multi-IFU spectrograph that does not have slit masks.^111,112

4.8.

Optical Interferometry

Interferometry coherently combines light from multiple apertures or telescopes to synthesize data for a single object. Optical interferometry provides exquisite resolution that is unobtainable with filled aperture telescopes; modern optical interferometers achieve limiting resolutions in the submilliarcsecond regime. Such instruments have been used to measure the diameters, distances, and/or other properties of solar system objects, individual stars, and recently even active galactic nuclei.

Optical interferometry was first demonstrated in the 1920s as a method to determine stellar properties.^113,114 Interferometry was further developed for astronomy in the late 1970s and 1980s, which led to instruments such as the Sydney University Stellar Interferometer¹¹⁵ and the Mark III interferometer.^116–119 The current generation of interferometers includes the Navy Precision Optical Interferometer (NPOI),¹²⁰ the Palomar Testbed Interferometer (decommissioned),¹²¹ the Center for High Angular Resolution Astronomy array (CHARA),¹²² the (decommissioned) Keck Interferometer,¹²³ the Magdalena Ridge Optical Interferometer (MROI; not yet operational),¹²⁴ the Large Binocular Telescope Interferometer (LBTI),¹²⁵ and especially the Very Large Telescope Interferometer, VLTI, experiment GRAVITY.^126,127 GRAVITY is supported by the European Southern Observatory, whereas the major US-based initiatives include LBTI, NPOI, MROI, and CHARA.

CHARA has been supported through six NSF/ATI grants beginning in 1992 and a current MSIP award, as well as through substantial private foundation support. ATI has been “essential in many ways” for CHARA.¹²⁸ These grants have supported the use of AO in CHARA, which improves the sensitivity of interferometric images (as opposed to improving the resolution, which is the usual provenance of AO). They have also supported the development of the Michigan Young Star Imager at CHARA (MYSTIC),¹²⁹ which is a K-band cryogenic six-beam combiner to image disks around young stars to study planet formation. An MSIP award supports public access to CHARA for 50 to 75 nights per year; observing time is administered through the National Optical Astronomy Observatories.

In addition to measuring diameters of bright stars, interferometry has determined orbits of spectroscopic binaries with high precision,^130–134 studied the mass-radius relation of dwarf stars,¹³⁵ and studied the effects of rotation on stellar surfaces.^136–138 Future science applications include the study of AGN cores, young stellar objects, and exoplanet host stars.

In the 1970s, interferometry was extended into the mid-IR waveband. This advance was enabled by the availability of IR detectors with a fast time response ($<1\mathrm{ns}$) for use as mixing elements in heterodyne detectors.¹³⁹ A two-element interferometer that studied planet forming disks was deployed at Kitt Peak.^140,141 Another instrument at Mount Wilson Observatory was the product of ATI support (9016474/Townes and 9119317/Townes); this Infrared Spatial Interferometer provided the first detailed, resolved observations of dust shells around late type stars.¹⁴²

Most implementations of optical interferometry have consisted of amplitude interferometry, which exploits correlations in the amplitudes of electric fields. However, in the 1950s, Hanbury Brown and Twiss exploited correlations in intensity (intensity interferometry) to interferometrically measure the diameters of bright stars.^143,144 Their methods are now being revisited using advances in fast electronics and photon detectors that were accomplished in subsequent decades.^145,146

Furthermore, the technical requirements of telescopes for intensity interferometry are met surprisingly well by modern instruments that were intended for a much different application. Cerenkov telescopes, which study celestial gamma-ray sources via flashes of light emitted by particle cascades in the upper atmosphere, have the potential for dual-use as optical intensity interferometers.¹⁴⁷ This approach is being pursued with the European Cerenkov Telescope Array.¹⁴⁸ Meanwhile an ATI-funded effort used the Very Energetic Radiation Imaging Telescope Array System, VERITAS, to demonstrate intensity interferometry (1806262/Kieda).¹⁴⁹ This effort recently demonstrated measurements of stellar diameters of two submilliarcsecond stars with a precision better than 5%.¹⁵⁰

4.9.

Radio Wave Digital Signal Processing

Ever since the discovery of celestial radio emission by Karl Jansky in the 1930s at Bell Laboratories,¹⁵¹ astronomers have benefited from technology to detect and process radio waves. Radio astronomy entered the digital era in the 1960s with the advent of the first astrophysical digital radio spectrum analysis.¹⁵² Both single dish radio antennas and antenna arrays have benefitted from tremendous progress in signal processing. There are diverse approaches to processing of contemporary radio astronomical data. Some typical functions that are important for arrays may include: (1) digitizing the analog sky signal, (2) channelizing this digitized signal into discrete frequency bins, and (3) combining signals from multiple antennas. This last function may consist of a weighted addition of antenna signals, known as beamforming or as a correlation (i.e., multiplication) of antenna signals. A common design pattern consists of dividing the “front end” digitization and channelization functions from the “back end” higher level tasks that may be performed on a different computer architecture.¹⁵³ Front-end systems may use field programmable gate arrays (FPGAs), which are reconfigurable integrated circuits. Back-end systems may employ GPUs, which are massively parallel computer engines. In general, the specialized computational methods of radio astronomy, such as those outlined above, add a considerable burden of complexity and a barrier to learning for the students and early career researchers who may ultimately build these systems.

The Collaboration for Astronomy Signal Processing and Electronics Research (CASPER) has provided open source technology and a supportive community for making this technology available for more than a decade. CASPER hardware and software powers more than 45 radio astronomy instruments worldwide.¹⁵⁴ CASPER has been supported by seven ATI grants from 2002 to 2019 that have included awards for highly regarded conferences/workshop tutorials in FPGA and GPU programming. CASPER hardware included the once popular Reconfigurable Open Architecture Computing Hardware (ROACH) series of standalone FPGA processing boards. CASPER powered spectrometers include the Allen Telescope Array Fly’s Eye radio transient search,¹⁵⁵ the Green Bank Ultimate Pulsar Processing Instrument,¹⁵⁶ the workhorse back-end system for the GBT,¹⁵⁷ the versatile GBT astronomical spectrometer (VEGAS), and many others. CASPER-based correlators and beamformers have been used for the Precision Array for Probing the Epoch of Reionization (PAPER),¹⁵⁸ Hydrogen Epoch of Reionization Array (HERA),¹⁵⁹ Large Aperture Experiment to Detect the Dark Ages (LEDA),¹⁶⁰ and the Focal L-band Array for the GBT (FLAG)¹⁶¹ discussed below, as well as the Smithsonian Astrophysical Observatory’s submillimeter array (SMA) Wideband Astronomical ROACH2 Machine (SWARM),¹⁶² the Breakthrough Listen,¹⁶³ and other projects to search for extratrerrestrial life. CASPER hardware was used for the Very Long Baseline Array (VLBA),¹⁶⁴ and it was subsequently used for the millimeter wavelength Very Long Baseline Interferometry (VLBI) as part of the EHT.¹⁶⁵ As of 2017, this hardware enabled VLBI data recording at 64 gigabits per second, which is a nearly five orders of magnitude improvement over the original VLBI recording rates circa 1969. One trend that CASPER has encouraged in recent years is the routing of data via Ethernet switches, a so-called “packetized” approach.

CASPER hardware has also been applied outside of traditional radio astronomy, notably to MKID readouts, which require similar processing functions. Examples include the ATI-supported Multiwavelength Sub/millimeter Inductance Camera (MUSIC)¹⁶⁶ that was deployed at the Caltech Submillimeter Observatory, a millimeter wavelength system developed at Columbia University,¹⁶⁷ and the optical instruments DARKNESS and MEC.¹⁰² A transition edge sensor (TES), discussed below, that uses CASPER hardware is the Muliplexed SQUID TES Array at Ninety Gigahertz (MUSTANG-2),¹⁶⁸ which is deployed at the GBT. Hickish et al. includes a census of CASPER deployments for radio astronomy as of 2016.¹⁵⁴

In addition to CASPER, several other general purpose digital signal processing platforms have been developed for radio astronomy. The Uniboard¹⁶⁹ FPGA platform was developed under a Joint Research Activity in the RadioNet FP7 European program; the radionet consortium coordinates the European radio astronomy community.¹⁷⁰ Several generations of digital signal processing platforms, named Redback, have been developed independently for the Australian Square Kilometer Array Pathfinder (ASKAP) array, and these platforms were partly inspired by CASPER.¹⁷¹

4.10.

Radio Wave Phased Array Feeds

The extreme faintness of celestial sources motivates telescopes with large collecting areas, which necessarily have correspondingly small fields of view. Therefore array feeds consisting of multiple antennas located at the telescope focal plane have been used to increase the field of view compared with a single feed antenna. Feeds consisting of dense arrays of electrically small antennas, called Phased Array Feeds (PAFs), offer the potential to sample a larger area of the focal plane than possible with a single-pixel horn feed and thereby increase mapping speed for surveys. Multiple beams are formed by combining signals sampled by the array elements with complex weights. PAFs were initially used for radar and satellite transmission; however in the late 1990s and early 2000s, they began to be applied to astronomy.¹⁷²

Early work was done by the National Radio Astronomy Observatory (NRAO) on the 43-m Green Bank Telescope (GBT),^172,173 with continuing collaboration with Brigham Young University (BYU). In particular, ATI award 9987339/Jeffs supported real-time DSP-based radio frequency interference mitigation strategies that used multiple antennas for tracking and nulling sources of interference.¹⁷⁴ This research grew to encompass PAFs, and it was supported by three more ATI awards as well as ten awards from other NSF programs in the 2000s and 2010s to support PAF development. These awards included support for the first millimeter-wave PAF, PHAMASS, in collaboration with University of Massachusetts, Amherst.¹⁷⁵ ATI supported a collaboration between BYU and West Virginia University to develop the FLAG instrument (see Fig. 12), which was recently commissioned,^176,177 and ATI and MSIP awards are supporting BYU and Cornell in the development of the Advanced L-band cryogenic Phased Array Camera for Arecibo (ALPACA). This instrument is expected to deliver the largest PAF to date for a U.S. telescope, and the only fully cryogenically cooled (LNAs and antennas) L-band PAF in the world.^178,179

Fig. 12

(a) Phased array feeds supported by ATI. (b) The FLAG instrument deployed at Green Bank Observatory. Image courtesy of Brian Jeffs, Brigham Young University. Engineering prototype PAF at Arecibo Observatory. Image courtesy of Mark Philbrick, Brigham Young University.

From the beginning, PAFs for astronomy were pursued independently by several research groups, including the National Research Council of Canada,¹⁸⁰ and especially the Commonwealth Scientific and Industrial Research Organization (CSIRO) in Australia for the Parkes telescope¹⁸¹ and ASTRON in the Netherlands.¹⁸² Research at ASTRON included technology demonstrators that paved the way for designs of the Square Kilometer Array (SKA) and separately for LOFAR.^182,183 The Australian group fielded a PAF on the Australia Square Kilometer Array Precursor (ASKAP) telescope.^184–186

Science results have begun to emerge from PAF technology development. The APERture Tile In Focus (APERTIF) instrument installed at the Westerbork Synthesis Radio Telescope (WSRT) in the Netherlands presented early results of on-sky testing.¹⁸⁷ Since the discovery of Fast Radio Bursts (FRBs),¹⁸⁸ the study of radio transients has taken on particular urgency. Surveys for transients and pulsars have been accomplished using PAF installed at ASKAP^189–191 and dedicated monitoring of the Vela pulsar with the Parkes 12-m antenna and Mark I PAF.^192,193 Notably, the PAF survey with ASKAP accomplished in days a survey of a comparable area (30 sq. deg) as was done in weeks-months with the Karl Jansky VLA,¹⁹⁴ albeit to a somewhat shallower depth.¹⁹⁰ A pilot survey of 10 sq deg with the FLAG instrument on GBT demonstrated instrument performance on-sky, although it did not report FRB or pulsar detections.¹⁷⁶ The ALPACA PAF, for which ATI supported a concept study¹⁷⁹ is expected to be the most sensitive PAF to date and will discover pulsars up to an order of magnitude faster than others.¹⁷⁶ These early results, coming nearly two decades after the initial papers on PAFs for astronomy, demonstrate the long time that it can take for technology to mature.

4.11.

Cosmology and FIR/Microwave Detectors

The study of the origin and evolution of the universe as a whole became the precision science of cosmology due to observations of the cosmic microwave background (CMB). This background radiation was famously discovered in 1965 at Bell Laboratories.¹⁹⁵ The original discovery, as well as the later discovery, by the COBE satellite launched by NASA in 1989, of the blackbody spectrum and anisotropy of this radiation^196–198 were awarded separate Nobel prizes. These discoveries provided crucial evidence for the Big Bang and the origins of cosmic structure.

Subsequent ground-based and balloon-based measurements revealed acoustic peaks in the angular power spectrum of the CMB, which provided evidence of flat spacetime and supported claims of accelerated expansion as revealed by observations of Type Ia supernovae. Polarization in the CMB was discovered,^199–201 and the cosmic microwave background remains an extremely active field. Community-wide plans for an ambitious “CMB S4” experiment to map polarization—from gravitational lensing and potentially a primordial background of inflationary gravitational waves—are underway.²⁰²

Motivated in large part by CMB science, detectors for the FIR-microwave regime, herein defined as radiation from microns to millimeters in wavelength or equivalently from gigahertz through terahertz frequencies, have been developed. They included coherent heterodyne receivers adapted from radio astronomy. They also included incoherent detectors, mainly bolometers that were developed separately.⁸⁸

4.11.1.

Coherent (heterodyne) detectors

Key early progress on the CMB, including its original discovery, as well as the discovery of fluctuations by COBE and the discovery of CMB polarization by ground-based observations were done with coherent detectors. Such detectors amplify the detected signal and change its frequency while preserving its phase.²⁰³ The large amplification ($\sim 100\mathrm{dB}$) needed to detect faint astronomical signals makes amplifiers highly susceptible to feedback oscillation; therefore, to achieve stability, the amplified signal is mixed with a local oscillator to change its output frequency. This system forms a heterodyne receiver. Front-end amplifiers for such systems have included High Electron Mobility Transistors (HEMTs), Superconductor-Insulator-Superconductor (SIS) mixers, and Hot Electron Bolometers (HEBs); however, maser amplifiers have the lowest noise.

ATI award 8024119/Readhead supported the development and use of a ruby maser for the 40-m telescope at Owens Valley Radio Observatory (OVRO). It was used to place early limits on CMB anisotropy.²⁰⁴ These limits were the most stringent at the time, and the results, which were published in 1989, required both dark matter and dark energy.²⁰⁵ They presaged evidence for the latter from observations of Type 1a supernovae by a decade.

The DASI instrument, which was supported by the Center for Astrophysics Research in Antarctica (CARA, an NSF Science and Technology Center), the NSF Office of Polar Programs, and partially by ATI, measured the angular power spectrum of the CMB; these results were the most widely cited in the history of the ATI program.⁸ DASI also accomplished the first detection of CMB polarization.¹⁹⁹ As mentioned above, the DASI electronics and correlator were partially supported by ATI, which principally supported a sister instrument, the CBI (see Fig. 13) through several awards between 1994 and 2002. CBI made the second measurement of CMB polarization in 2004.²⁰⁰

Fig. 13

The Cosmic Background Imager measured CMB polarization and was supported by several ATI awards. Image courtesy of Anthony C. S. Readhead.

ATI also supported millimeter wavelength interferometry through awards to the University of Chicago, beginning with an NSF Young Investigator award to John Carlstrom in 1992. Later ATI awards 0096913/Carlstrom and 0604982/Carlstrom supported an array of six 3.5-m telescopes and a wideband digital correlator for millimeter wavelength interferometry that were deployed at OVRO as the Sunyaev-Zel’dovich Array (SZA). These telescopes were used to make measurements of the Sunyaev-Zel’dovich (SZ) effect on galaxy clusters for cosmology and astrophysics.

The SZ effect is an absorption of the CMB caused by inverse Compton scattering of CMB photons by hot electrons, principally in the vicinities of clusters of galaxies.²⁰⁶ It is important for cosmology. Measurements of this effect can be combined with other diagnostics to measure the cosmological distance scale, and the number density of SZ-detected galaxy clusters can be used to probe dark energy. The first unambiguous detection of the SZ effect was accomplished with the single dish 40-m telescope at OVRO;²⁰⁷ this effort was made possible by the aforementioned ruby maser, and it was supported by an additional NSF award (8210259/Cohen). However, interferometer arrays were preferred over single dish telescopes due to their stability and spatial filtering capability.

The SZA continued operation until 2007. By combining its microwave measurements with X-ray measurements of hot cluster gas, independent distance estimates to galaxy clusters were obtained, which provided an independent measurement of the Hubble constant, $H_0$.²⁰⁸ In 2008, these telescopes were combined with OVRO telescopes as well as telescopes from the Berkeley-Illinois-Maryland Association (BIMA) and re-deployed to the Inyo Mountains in California near OVRO to complete the Combined Array for Radio Millimeter Astronomy (CARMA). CARMA was supported by NSF through cooperative agreements until 2015. It was an important precursor to the Atacama Large Millimeter-submillimeter Array (ALMA), and it provided essential training for instrumentalists.

ATI award 0905855/Church developed a prototype scalable 4-pixel heterodyne array based on MMIC amplifiers operating at 70-116 GHz. In a subsequent ATI award, 1207825/Church, this technology was used to build the 16-pixel Argus spectroscopic array, currently deployed on the Green Bank Telescope.

4.12.

Cosmic Dawn, the Epoch of Reionization, and Low-Frequency Radio Experiments

Understanding the origins of the first stars and galaxies is a great challenge of astrophysics. Cosmic Dawn refers to the period following the formation of the first stars. During this time, Lyman-$\alpha$ radiation from these stars suffused the universe. Subsequently, the predominantly neutral intergalactic medium became ionized during a watershed known as the EOR. Numerous projects aim to study these epochs with low-frequency radio waves.

Low-frequency radio waves probe redshifted cosmic background radiation from this epoch. Lyman-$\alpha$ pumping of cold atomic hydrogen causes an overpopulation of the 21-cm ground state. Level populations fall into equilibrium with the gas temperature because the Lyman-$\alpha$ line profiles are shaped by the thermal motions of atoms. Absorption follows from the radiation temperature cooling less quickly than gas temperature in an expanding adiabatic system. Thus, absorption at the restframe 21 cm wavelength caused by atomic hydrogen in the presence of UV light from the first stars, known as the Wouthuysen-field effect,^249,250 is expected to create an absorption feature. Similar effects may lead to 21 cm radiation alternating between absorption and emission as the early universe evolves.²⁵¹ Such features are redshifted into the observed FM radio frequency band.

The Murchison wide-field array,²⁵² the PAPER,²⁵³ and the subsequent HERA²⁵⁴ each received NSF support. They were designed to study the power spectrum of fluctuations in the brightness temperature of 21 cm emission from hydrogen. HERA will probe the universe out to redshifts, $z\lesssim 12$.

The LEDA seeks to study the sky-averaged 21 cm signal and power spectrum from primordial hydrogen out to redshifts $z\lesssim 20$ with an interferometer-based approach.^255–257 The advantage of this approach over single dipole antennas (discussed below) is the potential for joint estimation of the instrument calibration and a sky model to better control systematic uncertainties. LEDA received several NSF awards beginning with an ATI grant in 2011 (1106059/Greenhill), and it is one of several efforts pursuing this sky-averaged signal.

Notable among these efforts is the Experiment to Detect the Global EOR Signature (EDGES). The EDGES experiment consists of a table-top sized, single dipole antenna that was deployed at the Murchison Radio-astronomy Observatory in Western Australia; see Fig. 14. EDGES received three grants from ATI; the initial experiment placed constraints on the abruptness of the EOR transition.²⁵⁸ Subsequent work improved the sensitivity and calibration of the instrument.²⁵⁹

Fig. 14

EDGES low-band antenna and ground plane located at the Murchison Radio-Astronomy Observatory in Western Australia. Image courtesy of CSIRO.

In 2018, EDGES reported the first evidence for the detection of the 21-cm absorption signal corresponding to the birth of the first stars.²⁶⁰ This result earned widespread attention in the scientific community and the popular media. The public announcement of the scientific paper was aided by a 6-min popular science video about the project that was produced by NSF.²⁶¹ Physics World magazine later hailed the discovery as a “top 10 breakthrough of the year.”

A companion paper proposed to explain the unexpected size of the EDGES signal (a factor of two greater than expected) through interactions between dark matter and baryonic matter.²⁶² The EDGES signal was also notable for its unexpected width in redshift and its flat profile shape. These unexpected features generated considerable dialog in the scientific community; questions about the analysis and interpretation of the observations were raised²⁶³ and rebutted.²⁶⁴ Polarized foreground emission may have contaminated the observations.²⁶⁵ However, independent analyses of the data are unable to refute the principal claims.²⁶⁶ There is consensus that the EDGES results need confirmation.²⁶⁷ The EDGES team received a third ATI award in 2019 to continue their work (1908933/Bowman).

4.13.

Event Horizon Telescope

The EHT has been a remarkable success of the ATI program and of NSF support in general. On April 10, 2019, the EHT collaboration announced the first-ever direct image of a black hole at the center of the galaxy M87 and published six papers documenting their methods and findings.^268–273 The public announcement itself consisted of an unprecedented six coordinated press conferences in Washington, DC, Brussels, Santiago, Taipei, Tokyo, and Shanghai. The NSF played a lead role in organizing and coordinating these events. Resulting media exposure is estimated to have reached billions of people.²⁷⁴ The EHT image of a black hole, shown in Fig. 15, became a popular culture phenomenon, and it was the subject of a dedicated hearing of the U.S. House of Representatives Committee on Science, Space, and Technology.²⁷⁵ The EHT collaboration won the Breakthrough Prize for Fundamental Physics in 2020.

Fig. 15

Image of the black hole at the center of M87 released on April 10, 2019, by the Event Horizon Telescope Collaboration. Image courtesy of Event Horizon Telescope Collaboration.

The EHT image of a black hole in M87 was the culmination of decades of research and development in VLBI. VLBI is a technique that synchronizes telescope facilities around the world to effectively synthesize an Earth-sized telescope. Radio wavelength VLBI began in the late 1960s and the field advanced as new facilities were constructed. Developments were supported by NSF for astronomy and subsequently by NASA and other agencies for geodesy. Beginning in 1975, VLBI experiments at progressively shorter wavelengths sought to constrain the size of the black hole at the center of the Milky Way, Sgr A*, by showing that the source was dominated by scattering (apparent size dependent on wavelength squared). From 1994 to 2003, the millimeter wavelength VLBI program at MIT Haystack Observatory under Alan Rogers pursued exploratory observations, supported by several NSF awards.

ATI award 0352953/Doeleman continued this research and accomplished giga-bit per second (Gb/s) recording bandwidths for the first time, which along with the wideband digital backends developed with support from 0521233/Whitney, was a critical step in getting the first 1.3 mm fringes on Sgr A*.²⁷⁶ It led to the discovery of event horizon scale structure in Sgr A*.²⁷⁷ ATI award 0905844/Doeleman continued development of high speed data capture, demonstrated sustained data recording rate of $16\mathrm{Gb}/\mathrm{s}$, and resulted in detection of horizon scale structure in the galaxy M87.²⁷⁸ This marked the beginning of a larger push to reinstrument the VLBA. 1207752/Marrone first deployed 1.3-mm VLBI at the SPT, enabling the longest possible baselines to greatly enhance VLBI coverage. 1310896/Doeleman increased the millimeter wavelength VLBI recording capacity to $32\mathrm{Gb}/\mathrm{s}$, more than an order of magnitude greater than normal VLBA recording rates.²⁷⁹

The international EHT collaboration grew from early research collaborations circa 2009 that were formalized at a meeting in Waterloo in 2014. The collaboration received support from the European Research Council, other international agencies, and private funders. Some participants provided extensive support out of institutional budgets, and the project earned 22 separate NSF awards between 2000 and 2018.

The 2017 EHT observing campaign included eight sites in the intercontinental array and notably included a phased ALMA for the first time. The large collecting area of ALMA allowed for sufficient sensitivity to enable successful fringes and finally imaging of the black hole at the center of M87. Observations and analysis, especially of Sgr A*, are ongoing. EHT Founding Director Doeleman acknowledged the importance of early NSF funding to the project in congressional testimony: “NSF support was crucial. NSF support enabled the small EHT team to grow and carry out key proof of concept experiments.”²⁸⁰