forty years stuck

at the same sensitivity

Problem statement

The far-infrared wavelength regime (25-400 μm) contains irreplaceable spectral diagnostics for the formation of galaxies, stars, and planets — including the brightest cooling lines of interstellar gas and the peak thermal emission of cold dust — but no space mission has observed in the far-IR since the Herschel Space Observatory ended operations in 2013. The fundamental bottleneck is detector technology: far-IR detectors must operate at temperatures below 100 millikelvin and current state-of-the-art bolometer arrays contain only a few thousand pixels, one to two orders of magnitude fewer than needed for the next generation of science. Transition edge sensor (TES) bolometers and kinetic inductance detectors (KIDs) are the two leading technologies, but neither has demonstrated the combination of pixel count (>100,000), sensitivity (noise equivalent power <10⁻²⁰ W/√Hz), and flight readiness needed for a future far-IR flagship mission.

Why this matters

The far-IR is the only spectral window that traces the "cold universe" — the formation of new stars inside dusty molecular clouds, the assembly of galaxies in the early universe obscured by dust, the water vapor trail from protoplanetary disks to forming planets, and the composition of debris disks around nearby stars. The Herschel Space Observatory's 3.5m mirror operated at ~80K, meaning its own thermal emission dominated the background above ~100 μm. A cryogenically cooled (~4.5K) telescope with modern detectors would gain factors of 100-1000 in sensitivity, enabling entirely new science. The Astro2020 decadal survey identified far-IR detector maturation as a key technology investment for a future probe-class or flagship mission, and the Origins Space Telescope concept study estimated it would need ~60,000 detector pixels at sensitivities approaching the cosmic background photon noise limit.

What’s been tried and why it hasn’t worked

Herschel's PACS instrument used 2,560 bolometer pixels — the largest far-IR array ever flown — achieving NEP ~2×10⁻¹⁶ W/√Hz. The SPICA/SAFARI instrument (cancelled in 2020 when SPICA was descoped by ESA) was developing arrays of ~3,500 TES bolometers at NEP ~2×10⁻¹⁹ W/√Hz, but the cancellation halted that development. Ground-based KID arrays have reached ~2,300 pixels (NIKA2 at IRAM), but ground-based detectors operate in atmospheric windows with much higher photon backgrounds and don't need the ultra-low NEP required in space. Scaling TES arrays beyond a few thousand pixels is limited by the wiring complexity of individual SQUID readouts per pixel. KIDs offer multiplexed readout (hundreds of detectors per readout line), potentially solving the wiring problem, but their noise performance at the lowest backgrounds has not yet reached TES levels. Neither technology has been demonstrated in arrays larger than ~5,000 pixels at the NEP levels needed.

What would unlock progress

KID technology maturation to achieve photon-noise-limited sensitivity (NEP <10⁻²⁰ W/√Hz) in arrays of 10,000+ pixels, with multiplexing factors of 500-1000 per readout line. This requires advances in superconducting thin-film fabrication uniformity, low-noise cryogenic amplifiers, and digital readout electronics. Alternatively, TES array architectures with microwave SQUID multiplexing that reduce per-pixel wiring to enable 10,000+ pixel arrays. A cryogenic testbed that simulates the photon background of a 4.5K space telescope would allow realistic detector characterization without the cost of a space mission.

Entry points for student teams

The readout bottleneck is open to any team with a laptop: develop and test digital readout algorithms for frequency-multiplexed KID arrays — tone placement, tone tracking as resonances shift under optical load, and noise removal — against a simulated array, using the open-source tools the field actually uses as the reference implementation and the fitting stage. mkidgen3, the RFSoC-based multiplexed MKID readout from the Mazin Lab (GPL-3.0, https://github.com/MazinLab/mkidgen3), shows how a real readout chain is structured, and scraps (MIT-licensed Python for fitting superconducting-microresonator S21 data, https://github.com/FaustinCarter/scraps) handles resonator characterization; a team can generate resonator populations with realistic scatter and collision rates and report how many pixels per readout line survive at a target NEP. Fabricating and characterizing a real 10–50-pixel KID array is the other door, and it is a collaboration rather than a purchase: it takes a superconducting thin-film deposition and patterning facility, a sub-kelvin ADR or dilution refrigerator, and a multiplexed cryogenic readout chain — a combination held by a handful of groups worldwide (NIST Boulder, JPL, SRON, and a few university detector labs), so the realistic route is joining an existing group's cooldown as a collaborator, with the team's contribution being the measurement and analysis plan rather than the fabrication. Relevant disciplines: electrical engineering, condensed matter physics, signal processing, cryogenic engineering.