Every instrument I have worked on shares a theme: measuring signals thousands of times fainter than the glow of the instrument itself. That pushes everything to cryogenic temperatures — superconducting sensors a fraction of a degree above absolute zero, optics cooled until their own emission disappears — and it makes calibration and systematics control as important as raw sensitivity. This page collects the instrument work behind the science.
Antenna-Coupled Transition-Edge Sensors
Antenna-coupled TES Bolometers used in BICEP2, Keck Array, and SPIDER (ApJ 812, 176, 2015)
BICEP/Keck detects the CMB with antenna-coupled transition-edge sensor (TES) bolometers. Each pixel is a planar array of slot antennas that collects millimeter-wave power and deposits it on a thermally isolated island; a superconducting film biased on its transition converts tiny changes in absorbed power into large changes in resistance, read out by SQUID amplifiers. Because the antennas, band-defining filters, and sensors are all lithographed onto silicon, entire focal planes can be mass-produced photolithographically — the technology that let BICEP2, Keck Array, and SPIDER field thousands of background-limited detectors.
BICEP3: A Compact, Modular 95 GHz Receiver
BICEP/Keck XV: The BICEP3 Cosmic Microwave Background Polarimeter and the First Three-year Data Set (ApJ 927, 77, 2022)
BICEP3 packs 2560 detectors at 95 GHz — 20 silicon tiles, each carrying 128 polarization-sensitive TESs — behind a compact refracting telescope. I led the design of its cryogenic focal plane and detector readout as my Ph.D. work. The modular packaging provides the thermal stability, magnetic shielding, and precise alignment the detectors require, while allowing any individual tile to be swapped without disturbing the rest — a design that carried forward into BICEP Array. BICEP3 has observed every season since 2016 and anchors the 95 GHz channel of the BK18 result.
Ultra-Low-Noise Far-Infrared Bolometers
High-sensitivity transition-edge-sensed bolometers: Improved speed and characterization with AC and DC bias (Journal of Applied Physics 134, 094503, 2023)
A cryogenically cooled space telescope removes nearly all photon background, so the detectors themselves set the sensitivity of a far-infrared spectrometer. At JPL I developed and characterized TES bolometers for a proposed 87–230 µm grating spectrometer, reaching noise equivalent powers below 2×10−19 W/√Hz — among the most sensitive bolometers ever demonstrated — and integrated their frequency-division multiplexed readout. This detector technology is now being carried toward PRIMA, a proposed NASA far-infrared probe mission.
Calibrating SPHEREx
SPHEREx measures diffuse signals across the entire sky, so its calibration has to hold from arcseconds to the full sphere. I led the telescope's integration and calibration campaign — verifying focus to better than 15 µm and spectral calibration to better than 1 nm, with the whole telescope operating at 45 K — and its in-orbit commissioning after launch. The full story, from component-level testing to on-orbit performance, is in the instrument paper (Korngut et al. 2026).
The SPHEREx instrument: telescope, focal plane assemblies, and photon shields (Korngut et al. 2026).
Spectral Response of SPHEREx (ApJS 284, 10, 2026)
In SPHEREx, every pixel sees a different wavelength, set by the linear variable filters mounted over the six detector arrays — so the spectral calibration is a per-pixel measurement across six 2k×2k arrays. We scanned a monochromatic beam across the focal plane in a cryogenic test chamber, measuring the response of each pixel as a function of wavelength, and delivered band centers accurate to better than 1 nm in Bands 1–4 and better than 10 nm in Bands 5–6. In flight, the helium airglow line at 10830 Å provides a continuous cross-check of the calibration.
Left: the spectral calibration source — a monochromator with order-sorting filters and a xenon-lamp wavelength reference — feeding the cryostat window. Right: the measured spectral response of all 102 SPHEREx channels (Hui et al. 2026).
Improving H2RG Performance in SPHEREx Brassboard Model (ApJS 276, 43, 2025)
Intensity mapping demands noise stability on large angular scales, exactly where infrared detector arrays are ordinarily dominated by 1/f noise. Using the SPHEREx brassboard camera — a flight-like H2RG detector with the custom Video8 readout electronics — we developed the sampling strategies the mission now uses in flight: nonsequential row reads, continuous drift monitoring against stable reference voltages, multiple weighted visits to the reference pixels, and onboard slope fitting with cosmic-ray removal. Together these suppress the large-scale noise power that would otherwise limit measurements of diffuse light.
Left: the brassboard camera in its light-tight enclosure with diffuse illumination optics. Right: two-dimensional noise power spectra of dark maps — the readout scheme removes the excess power at low spatial frequencies (Nguyen et al. 2025).
The SPHEREx Instrument: Calibration, Testing, and Performance from the Laboratory to In-Orbit Commissioning (Korngut et al. 2026)
SPHEREx has no focus mechanism in flight, so its focus had to be set correctly on the ground — at 45 K, through a vacuum window. Each focal plane assembly was iteratively shimmed onto the telescope's curved focal surface, and the focus was measured with a collimator through a large vacuum window and a reflective cold filter (Applied Optics 63, 3453, 2024), reaching ~15 µm systematic and ~5 µm statistical error — comfortably inside the telescope's depth of focus. In-orbit commissioning confirmed the image quality on the sky.
Setting the focus: metrology of the mid-infrared focal plane during assembly — the measured detector surfaces (colored) converge onto the target focal surfaces (grey) after shim adjustment (Korngut et al. 2026).
Detector Development: AlMn TES and TKIDs
With my students at Caltech, I am developing two detector technologies: AlMn transition-edge sensor arrays for CMB-S4 prototype focal planes, characterized through dark load curves across bath temperatures; and thermal kinetic inductance detectors (TKIDs), which combine bolometric operation with a natively frequency-multiplexed microwave readout — thousands of detectors on a single line, with no SQUIDs.