CEA-Leti Reports 42 to 55 Percent Single-Photon Detection Efficiency and a 150 Hz Intrinsic Dark Count Rate in HgCdTe Avalanche Photodiodes at 80 K, arXiv 2611.24461, 18 November 2026: TRL 4

Quentir Defense Monitor

Evidence-based insights for quantum defense and security. Published by Quentir Systems LLC · November 19, 2026.

A low-angle macro view inside an open brushed-stainless laboratory cryostat at dawn: a tiny mirror-polished semiconductor photodiode die on a white ceramic carrier sits inside two frost-rimmed ice-blue cold-shield rings, a separate silicon readout chip on a gold-plated pedestal hovers a clear gap away joined only by hair-thin gold bond wires, and a single pale-gold photon filament enters through the shield aperture and ends on the diode, with snow-covered Alpine peaks under a clear sunrise sky through the tall window behind

On 18 November 2026 a team at CEA-Leti in Grenoble, with Jean-Marc Merolla of the FEMTO-ST institute in Besançon, posted to arXiv a characterization of mercury cadmium telluride avalanche photodiodes operated as single-photon counters at 80 kelvin. The paper by Arthur Charrier, Johan Rothman and twelve co-authors was written for the SPIE conference Optical Sensing and Detection IX in Strasbourg in April 2026, and it reports two numbers that look incompatible until the measurement is understood. A diode bonded directly to its silicon readout amplifier detected single 1.55 micrometer photons with an external efficiency of 42 to 55 percent while registering 2.6 to 17 million dark counts per second. A sister diode from another wafer, wired to an amplifier held 0.7 centimeters away, showed a measured dark current of 10.9 femtoamperes, which Section 4.2.2 converts to about 150 dark counts per second using a gain of 450. The abstract inconsistently labels the 10.9 femtoampere value gain-normalized; the gain-normalized current corresponding to 150 counts per second is approximately 0.0242 femtoamperes.

The gap between those two figures is the finding. The authors argue that the million-count rate comes from infrared light emitted by the readout circuit itself, an effect known as amplifier glow, and from residual thermal radiation, while the diode's own contribution is roughly five orders of magnitude smaller. For a defense reader this matters because single-photon detection in the short-wave infrared decides how far a lidar can range a low-reflectivity target, how little laser power a free-space optical link needs, and whether a quantum key distribution receiver can be built without a superconducting cryostat. This post reads the result as a sensing development in the imaging-in-low-light sub-branch, dual-use in posture, and places it at the laboratory-testing rung of the readiness ladder.

What CEA-Leti and FEMTO-ST measured at 80 K and 1.55 micrometers: two 30-micrometer HgCdTe diodes, one on a CMOS amplifier and one wired 0.7 centimeters away

The devices are planar p-n photodiodes grown by liquid-phase epitaxy, 30 micrometers in diameter, with a cutoff wavelength near 3 micrometers at 80 kelvin. In HgCdTe with a cadmium fraction below 0.5, electrons dominate impact ionization while holes barely ionize, so the avalanche adds almost no excess noise even at high gain, and a linear-mode diode, whose output stays proportional to the number of photons, can lift a one-photon event above the readout noise. The two samples came from different wafers but share the design and process.

The first diode was flip-chip bonded onto a CMOS readout chip carrying a resistive transimpedance amplifier with 410 megahertz of bandwidth and about 50 electrons of noise at unity gain. It was tested in a liquid-nitrogen cryostat behind a cold shield with no cold filter, illuminated from the back without an anti-reflection coating by a 1.55 micrometer laser emitting 40 picosecond pulses at 4 megahertz. The unity-gain quantum efficiency was 63 percent, and the photon-counting measurements were taken at 15.5 volts, where the gain was 470.

The second diode sat on a passive interconnection circuit and was wire-bonded to a capacitive transimpedance amplifier about 0.7 centimeters away. A capacitive amplifier integrates charge, so the slope of its output voltage gives the dark current directly. The dark current stayed below the 1 femtoampere floor of the setup up to 14 volts and rose sharply past 15 volts as band-to-band tunneling set in.

Why a 150 Hz intrinsic dark count rate and a 3.2 MHz measured rate are both true: amplifier glow in the hybridized assembly

The photon-counting results depend on where the detection threshold sits relative to the amplifier's root-mean-square noise. At 2.5 times that noise the external photon detection efficiency was 54.6 percent, with 17.7 megahertz of dark counts and a timing jitter of 715 picoseconds. At 4.5 times the noise the efficiency fell to 42.9 percent, the dark counts to 2.6 megahertz and the jitter to 461 picoseconds. Dividing by the 63 percent quantum efficiency gives an internal efficiency of 66 to 86 percent, which is why the authors expect an anti-reflection coating, lower amplifier noise or higher gain to push the external figure past 90 percent. They treat the 4 sigma threshold, with 3.2 megahertz of dark counts, as the reliable operating point.

A dark count rate of 3.2 megahertz would disqualify the detector from most quantum optical uses if it were intrinsic. The capacitive-amplifier measurement is their answer. At 15.5 volts the second diode drew 10.9 femtoamperes; dividing by the electron charge and by a gain of 450 gives about 150 electron-hole pairs per second, the dark count rate the diode itself would produce. The authors attribute the remaining four to five orders of magnitude to photons emitted by transistors in the adjacent CMOS circuit, and to stray 300 kelvin thermal radiation that a cold filter could block. The paper does not demonstrate a photon-counting diode with 150 hertz of dark counts. It demonstrates that the physics of the diode permits one, and that the engineering of the readout is what currently prevents it.

The comparison the authors draw is with superconducting nanowire single-photon detectors, which reach efficiencies above 90 percent, dark count rates below 100 hertz and jitter below 10 picoseconds, and which operate below 3 kelvin. They write that HgCdTe diodes could reach comparable performance at 80 kelvin while running without dead time and with photon-number resolution. That is a projection. The measured device trails nanowire detectors on dark counts by four orders of magnitude and on jitter by more than one.

Quantum pillar: sensing (imaging in low light). Use posture: dual-use. Technology readiness: TRL 4 of 9. Two assembled single-pixel diodes were characterized on a laboratory bench in a liquid-nitrogen cryostat, with no coating, no cold filter and no array, so the result validates the detector physics under laboratory conditions and nothing has yet been tested against a realistic target, link or key-exchange traffic.

What a photon-counting infrared diode at 80 K would let a force do: lidar, laser links and QKD receivers without a 3 K cryostat

The applications the authors list are the defense-relevant ones: free-space optical communications, lidar, and quantum optical systems including key distribution and entanglement-based networks. Each is photon-starved. A lidar that registers single returned photons at 1.55 micrometers can range a dim or distant target with a lower-power transmitter, and a laser terminal that counts photons closes a link budget an analog receiver cannot. As in the squeezed-light receiver of the Michigan and USC RF photonic imaging post, the receiver's noise floor sets the range.

The operating temperature is the practical point. A Stirling cryocooler that holds 80 kelvin fits an aircraft pod or a satellite bus with far less difficulty than a system that holds 2 kelvin for a nanowire detector. NASA Goddard and DRS reported in 2018 an integrated detector cooler assembly built around a linear-mode photon-counting HgCdTe array for space lidar, with more than 60 percent photon detection efficiency from 0.9 to 4.3 micrometers, a dark count rate below 250,000 per second, and cooling to 110 kelvin at 7 watts, qualified through vibration, thermal-cycling and thermal-vacuum testing. The CEA-Leti work pushes the same material toward the dark-count floor that quantum receivers demand.

The posture is dual-use in the plain sense. The same diode serves a key distribution receiver, which protects one's own communications, a covert or long-range imaging lidar, which finds what another party intends to conceal, and civil uses such as deep-space communications and automotive ranging. CEA-Leti's own description of its avalanche photodiodes lists lidar for defense, space and mobility alongside free-space optical communication and quantum optics, and states that a spinoff, Moon Photonics, was founded in January 2026 to commercialize the short-wave infrared version. The paper contains no operational claim; it establishes detector figures of merit and the mechanism that limits them.

What stands between a conference paper and a program office: arrays, coatings, jitter and a spinoff founded in January 2026

The first gap is that no single device in the paper delivers both headline numbers. The efficiency figures come from the diode bonded to the fast resistive amplifier, and the 150 hertz figure from a different diode read by a capacitive amplifier with 3 to 30 megahertz of bandwidth, far too slow for photon counting. A buyer should read the result as a diagnosis: the hybridized assembly must be redesigned so that the amplifier's own infrared emission no longer reaches the diode. The authors state this as future work and give no schedule.

The second gap is scale and finish. The measurements are on single 30 micrometer pixels without an anti-reflection coating, and an imaging lidar or a free-space receiver needs an array with uniform gain across pixels. Jitter of 464 picoseconds at the reliable threshold is more than four times the 100 picosecond figure the authors cite as the requirement for quantum communication, and they attribute the excess to a thick absorption layer that a redesign could thin.

The third gap is industrial. The work was funded by the France 2030 plan through the OQuLus project, by the French National Research Agency's Carnot QPIC project, and by the Horizon Europe Qu-Pilot project, a 19 million euro program running from April 2023 to March 2027 that is building federated European pilot production capabilities for quantum technologies across 21 partners, CEA among them. The CORDIS entry states the intent to keep critical intellectual property and a quantum supply chain inside the Union, which connects to the critical-components list and supply-chain monitoring powers in the leaked draft EU Quantum Act that Quentir examined on 18 November 2026; a photon-counting infrared detector made in Grenoble, with a spinoff less than a year old as its commercial channel, is the kind of component such a list would cover.

A program office evaluating HgCdTe avalanche photodiodes for a photon-starved receiver should ask for four things the paper does not yet contain: a hybridized device whose measured dark count rate approaches the 150 hertz the capacitive measurement predicts; a coated array with uniform gain across its pixels; jitter below 100 picoseconds in the same assembly; and an environmental qualification comparable to the 2018 NASA and DRS assembly. Until those arrive, the published result places the technology at laboratory validation, with the physics of the diode no longer the open question.

Sources

Primary source: Charrier, Rothman and co-authors (CEA-Leti, FEMTO-ST), arXiv 2611.24461, 18 November 2026. Other material: CEA-Leti avalanche photodiode page; NASA NTRS record 20190002029 (Sun et al., SPIE Defense and Commercial Sensing, 2018); CORDIS record for Qu-Pilot, grant 101113983. The capability reading, the posture argument, the readiness placement and the buyer items are Quentir analysis.

  1. paper by Arthur Charrier, Johan Rothman and twelve co-authors
  2. integrated detector cooler assembly
  3. description of its avalanche photodiodes
  4. Qu-Pilot project