A quantum dot that emits at 1550 nanometres is useful only if you can find it, put a cavity around it, and get its photons out. The finding and the putting have usually been separate operations, and the putting has usually been luck. A preprint posted to arXiv on 7 November 2026, from Monica Pengerla, Yury Berdnikov, Alexander Kosarev and colleagues at the Technical University of Berlin, the Technical University of Denmark, and Wrocław University of Science and Technology, closes that gap for one specific and important case: metal-organic vapour-phase epitaxy (MOVPE)-grown indium arsenide quantum dots in indium phosphide, emitting in the telecom C-band, deterministically integrated into circular Bragg grating resonators. The headline numbers are a radiative lifetime of 1,400 picoseconds, a post-selected single-photon purity of 95 percent, and a raw two-photon interference visibility of 35 percent — all measured under non-resonant, above-band excitation, with effectively no Purcell enhancement. This is a preprint, not yet peer-reviewed. But the fabrication result is the point: the dots are placed on purpose, and the placement works.

This desk covered a directional quantum-dot molecule result on 17 November and the MoTe₂ emitters on 14 November. This is a different kind of result — a manufacturing claim about a known emitter, not a new emitter physics claim.

Why placement is the hard part

A circular Bragg grating is a bullseye: a central mesa surrounded by concentric rings that redirect a quantum dot’s emission upward into the collection optics. It only works if the dot sits at the centre. The team’s own simulations show that photon extraction stays robust for dot displacements up to 150 nanometres from the mesa centre, but the Purcell enhancement falls away progressively with displacement.

The standard way around this has been marker-based alignment: locate the dot relative to alignment marks, remove the sample, and write the structure in a second lithography step. Every transfer between instruments adds drift, and drift at cryogenic temperatures is the enemy. The Berlin–DTU approach removes the transfer, using deterministic in-situ electron-beam lithography — iEBL — in which low-temperature cathodoluminescence mapping, dot localisation, and pattern writing happen in a single session in one instrument.

The sequence is mechanical and worth describing plainly. The sample is flip-chip bonded to a silicon carrier, with a 100-nanometre aluminium back reflector and a 100-nanometre silicon dioxide spacer beneath the indium phosphide layer that holds the dots. A silicon dioxide hard mask and a layer of CSAR 62 electron-beam resist go on top. A focused electron beam is raster-scanned across the sample at cryogenic temperature, and the resulting cathodoluminescence is spectrally resolved in the near-infrared — locating individual dots and identifying which ones emit inside the target C-band before anything is etched. The grating pattern is then aligned to the extracted dot position and written directly into the same resist, in the same session. The mapping dose is 4 mC/cm², kept low so the resist stays in its positive-tone regime; the patterning dose is 32 mC/cm², which flips it to negative tone.

The resulting geometry is a central mesa of 840 nanometres radius, surrounded by three concentric rings each 589 nanometres wide, separated by 635-nanometre trenches, etched 390 nanometres into a 535-nanometre indium phosphide layer with the dot layer 243 nanometres below the top surface. Electron microscopy confirms the structures, with one honest imperfection: the ring sidewalls are tilted rather than vertical, a systematic deviation the team feeds back into its simulations. Three devices — CBG1, CBG2, and CBG3 — were fabricated, with cavity resonances inside the C-band and brightest emission lines at 1544 nanometres for CBG1 and 1470 nanometres for CBG2.

What the dots do

The optical characterisation, performed at about 4 kelvin, gives a coherent picture of a solid but unhurried emitter. The radiative lifetime is T₁ = (1,400 ± 50) picoseconds — long, by quantum-dot standards, and a direct consequence of the missing Purcell enhancement. The spectral linewidth is (41.0 ± 4.0) µeV, roughly 87 times broader than the Fourier limit of about 0.47 µeV set by the 1.4-nanosecond lifetime.

The grating does what a grating is for. Finite-element simulations predict a photon extraction efficiency above 80 percent across most of the C-band, reaching about 83 percent near the band centre at a collection numerical aperture of 0.5, against about 20–25 percent for the planar reference. That is a simulated figure, not a measured one, and the distinction matters. What was measured is brightness: the integrated photoluminescence of the three grating-integrated dots is on average about twice that of seven randomly selected dots in the planar reference. The measured twofold enhancement falls short of the simulated three- to fourfold, which the authors attribute to slightly off-centre dot positions and the tilted sidewalls. The gratings also substantially suppress broadband background emission. The Purcell factor is effectively unity: the design was optimised for extraction, not for strong cavity enhancement.

Purity, post-selected and integrated

Single-photon purity is where this paper most rewards careful reading: it reports two numbers for the same measurement and explains why they differ.

The measurement is a Hanbury Brown–Twiss autocorrelation under pulsed non-resonant excitation at 800 nanometres, 80 MHz repetition rate, at saturation. For CBG2 at 3.6 kelvin, a fit to the correlation function gives g²(0) = (0.03 ± 0.01); for CBG1, (0.05 ± 0.02). The paper calls these the post-selected values, and the abstract’s headline purity of (0.95 ± 0.02) is one minus the fitted g²(0). The fit is a local measure: it isolates the multiphoton contribution around zero delay.

Analysed differently, the same histograms give larger numbers. Integrating the area of the zero-delay peak over the full 12-nanosecond window between pulses and normalising to the neighbouring peaks gives g²(0) = (0.17 ± 0.03) for CBG2 and (0.22 ± 0.05) for CBG1 at 3.6 kelvin. That is the integrated value, and it is the more honest statement of what a user of the source would actually receive: roughly one in five or six excitation cycles carrying a multiphoton contribution, rather than one in twenty or thirty.

The difference has a physical cause, and the paper names it: carrier recapture. Above-band excitation creates a reservoir of free carriers around the dot; after the dot emits, residual carriers can be recaptured within the same 12.5-nanosecond cycle, producing a second photon. The fitted value describes the multiphoton probability at zero delay; the integrated value describes the multiphoton content of the source’s output per pulse. Coverage should report both, and the abstract’s 95 percent purity is the post-selected one.

Temperature dependence supports the recapture picture. At 20 kelvin, CBG2 retains similar antibunching while CBG1 shifts modestly, to about 0.06 fitted and 0.26 integrated. At 40 kelvin, CBG2 is at about 0.05 fitted and 0.20 integrated, and CBG1 at about 0.06 and 0.30. Still, pronounced antibunching survives to 40 kelvin, and photoluminescence was tracked to 55 kelvin, beyond the range of compact Stirling cryocoolers.

Indistinguishability: 35 percent is the number

The Hong–Ou–Mandel measurement was performed on CBG1’s neutral exciton at 1544 nanometres. Co-polarised photons interfere; cross-polarised photons provide the distinguishable reference. Integrating the coincidence histograms over −6 to +6 nanoseconds — the full 12-nanosecond window between pulses — gives a visibility of (35 ± 2) percent. That is the raw number, and it is the number that should be quoted.

Narrow the integration window and the number climbs, reaching (70 ± 18) percent at the shortest accessible window. The mechanism is the same as in the MoTe₂ result: temporal post-selection keeps only the photon pairs that arrive close together, the ones whose phase has wandered least, and discards the rest. It is a legitimate diagnostic — it shows the coherence is present in the early-arriving photons — but it is not the performance of the source, because a network cannot throw away most of its photons to keep the identical ones. The paper reports both figures side by side in the abstract, and its discussion is candid that indistinguishability “remains limited by dephasing and carrier-induced fluctuations.”

The comparison that puts 35 percent in context is the paper’s own. The best reported C-band result, from Hauser and colleagues in Nature Communications in January 2026, reached 91.7 percent visibility — but under longitudinal-acoustic phonon-assisted excitation, with a Purcell factor of 4.7, in molecular-beam-epitaxy-grown dots. The same group’s dots under above-band excitation managed only (20.6 ± 7) percent. Against that benchmark, 35 percent under plain above-band excitation, with no Purcell enhancement and MOVPE-grown material, is a genuine advance in the simple-excitation regime — and still far from the 90-percent territory interference-based protocols assume. The long lifetime is the core of the problem: at 1.4 nanoseconds, dephasing and spectral wandering have a long time to act on the photon while it is being emitted.

The honest boundary

Five qualifications should accompany any coverage of this preprint, and all five are visible in the paper itself.

First, it is a preprint. It was posted to arXiv on 7 November 2026 as version 1, and it has not been peer-reviewed.

Second, the indistinguishability result comes from a single device. The HOM measurement is reported for CBG1 only. CBG2 gives the best purity numbers; CBG3 appears in the brightness and simulation comparisons. Three devices show the fabrication process works more than once, but the interference claim rests on one dot.

Third, the excitation is non-resonant throughout. That is the paper’s deliberate choice — above-band excitation is simple and robust — but it is also the reason for the recapture that inflates the integrated g²(0) and the charge fluctuations that limit the visibility. The authors’ own route to improvement starts with moving away from it, toward resonant or quasi-resonant excitation.

Fourth, the linewidth is 87 times the transform limit. The photons are pure enough and bright enough, but they are not coherent enough, and no amount of post-selection changes the underlying ratio of coherence time to radiative lifetime.

Fifth, everything was measured at about 4 kelvin. The temperature data to 40 and 55 kelvin are encouraging, but a source that needs a cryostat is a source that needs a cryostat.

None of these qualifications diminishes the fabrication result; they locate it. What Pengerla, Berdnikov, Kosarev and colleagues have done is take the hardest practical step in telecom quantum-dot sources — putting a chosen dot, at a chosen wavelength, at the centre of a chosen cavity — and show that the resulting devices emit photons good enough to measure interference under the simplest excitation conditions available. The 35 percent visibility is not a destination. It is a baseline for a platform that can now be improved by the standard means — Purcell engineering, charge stabilisation, selective excitation — without re-solving placement for every device.

What to watch next

The follow-ups that would matter most are the ones the paper names. The first is radiative-rate engineering: a grating design that trades some broadband extraction for real Purcell enhancement, shortening the 1.4-nanosecond lifetime toward the coherence time and re-measuring the raw HOM visibility without post-selection. The second is excitation: the same devices driven resonantly or with phonon-assisted excitation, which should suppress the recapture behind the inflated integrated g²(0). The third is charge stabilisation — electrical control of the dot’s environment, the fix the Waterloo waveguide-mirror result also needed for its wandering resonance. The fourth is scale: more devices, and eventually two-device interference, the test of whether two deterministically placed sources can interfere with each other.

There is also a quieter lesson in the method. In-situ electron-beam lithography, locating the dot and writing the cavity in one cryogenic session, is the part of this paper most likely to be reused first. The bullseye only works if the dot is at the centre, and this is a way to put it there.

Sources and further reading

  • Monica Pengerla, Yury Berdnikov, Alexander Kosarev, Dominik M. Feder, Avijit Barua, Aris Koulas-Simos, Paweł Holewa, Sven Rodt, Imad Limame, Elizaveta Semenova & Stephan Reitzenstein, “Deterministically fabricated C-band single-photon sources based on quantum dot circular Bragg grating cavities,” arXiv:2611.10988 [quant-ph], posted 7 November 2026 (v1). Primary source for all figures in this article. Corresponding author: Imad Limame (TU Berlin); first three authors contributed equally. Affiliations: TU Berlin; DTU Electro and NanoPhoton, Technical University of Denmark; Wrocław University of Science and Technology.
  • Benchmark cited in the paper: Nico Hauser et al., “Deterministic and highly indistinguishable single photons in the telecom C-band,” Nature Communications 17 (2026) — HOM visibility 91.7% under longitudinal-acoustic phonon-assisted excitation and (20.6 ± 7)% under above-band excitation, Purcell factor 4.7 ± 0.5.
  • Context on this desk: MoTe₂ emitters (7.1% raw HOM visibility), covered 14 November; single-photon reflection from a quantum dot in a nanowire waveguide (Waterloo), covered 16 November. Both are separate results.
  • Note on verification: all quantitative values here are taken from the arXiv v1 preprint text (PDF) and its abstract, which match; the preprint has not been peer-reviewed. The simulated photon extraction efficiency of about 83% is a simulation result, not a measured device efficiency — the measured brightness enhancement is about twofold over the planar reference. No market-size figure, commercial claim, or investment judgement appears in this article; the site’s Terms carry the not-investment-advice disclaimer.