Two-dimensional materials have been promising single-photon sources for a decade, and for most of that decade the promise has carried the same asterisk: the photons are pure enough, but they are not identical enough. A paper published in Nature Communications on 11 November 2026, from Battulga Munkhbat’s group at the Technical University of Denmark, removes part of that asterisk. The team reports reproducible quantum emitters in bilayer molybdenum ditelluride, emitting in the near-infrared between 1090 and 1200 nanometres, with triggered single-photon purity as good as g²(0) ≈ 0.01, electrical tuning over roughly 3 meV, and — for the first time in any transition-metal dichalcogenide (TMD) emitter — a measured Hong–Ou–Mandel visibility of 7.1 percent, rising to about 60 percent only with temporal post-selection. The difference between those two numbers is the story.
This desk has followed the 2D-emitter arc from its visible-light beginnings: tungsten diselenide emitters localized by strain, where the indistinguishability was about 2 percent, and the all-van-der-Waals quantum light sources that made the platform integrable without fixing the coherence problem. The DTU result is the first time the arc has produced a two-photon interference measurement in the near-infrared, and the first time a TMD emitter has been shown to be triggered, tunable, and pure in the same device.
The number that matters, and the number behind it
Indistinguishability decides whether single photons can interfere, and interference is what photonic quantum computing and entanglement distribution run on. Two photons that differ in colour, timing, or polarisation will not bunch at a beamsplitter, and no amount of purity compensates for that. The field’s best quantum-dot sources reach visibilities above 90 percent. The best TMD result before this paper, on WSe₂, was about 2 percent. That gap is the whole problem in one number.
The DTU team measured two-photon interference with a Hong–Ou–Mandel interferometer: consecutive photons from the same emitter, separated by a 12.5-nanosecond fibre delay matched to the 80 MHz laser repetition rate, brought together on a beamsplitter. For their best emitter, an hBN-encapsulated device excited through a p-shell resonance, the visibility integrated over the full 12.5-nanosecond window is 7.1 ± 0.9 percent. For a second emitter under quasi-resonant excitation it is 3.6 ± 1.6 percent.
Then comes the second number. Narrow the coincidence window — counting only photon pairs arriving within 200 picoseconds of each other, then narrower still — and the visibility climbs past 20 percent for one emitter and past 30 percent for the other, reaching about 40 and about 60 percent at the smallest accessible windows. Temporal post-selection discards the photons that arrive late, the ones whose phase has wandered furthest; it measures the indistinguishability of a selected subset, not of the source’s output. It is a legitimate diagnostic, but it is not the performance of a source, because a network cannot afford to throw away most of its photons to keep the identical ones. Coverage should do the same: 7.1 percent is the visibility; 60 percent is the visibility after post-selection.
The reason for the modest raw number is quantitative. From the interference fits, the team extracts coherence times of 62 ± 15 and 81 ± 10 picoseconds for the two emitters, against radiative lifetimes of roughly 240–440 and about 500 picoseconds. The photons simply do not stay coherent for as long as they take to be emitted. The standard relation for a dephasing-limited pulsed experiment, V = τc/2τ, predicts visibilities of about 7–13 percent and about 8 percent — in agreement with what was measured. Fast dephasing, from phonons and charge noise, is what limits the interference.
How the emitters are made
The fabrication is the paper’s other claim to significance, because reproducibility is where MoTe₂ emitters have previously failed. Earlier demonstrations reported lifetimes from about 20 nanoseconds to milliseconds, linewidths broader than 600 µeV, and no reliable tuning — a spread so wide it suggested nobody was making the same device twice. The DTU process is deterministic in two deliberate steps: strain, then defects.
First, the substrate: a distributed Bragg reflector of 20 pairs of gallium arsenide and aluminium arsenide, capped with about 100 nanometres of aluminium oxide, designed for near-unity reflectivity across 1100–1240 nanometres. Onto it the team patterns three-pointed star nanopillars, about 100 nanometres tall. A mechanically exfoliated bilayer MoTe₂ flake is dry-transferred on top and conforms to the pillars, forming nanowrinkles whose directional strain confines excitons into localized states. Second, the promising strained regions are selectively irradiated with a focused electron beam to introduce atomic-scale defects, which activate emitter formation. The team reports efficient generation at about 50 percent of available sites when strain and defect engineering spatially overlap.
The emitters that result are genuinely good single-photon sources by every metric except indistinguishability. Lifetimes across the sample run 130–450 picoseconds, with a mean of about 263 picoseconds — sub-nanosecond, and two orders of magnitude faster than the best previous MoTe₂ emitters. The representative emitter shows a degree of linear polarisation of about 74 percent, aligned with the wrinkle direction. Emission lines narrow to a resolution-limited value of about 150 µeV on short integration times. Emission spans 1090–1200 nanometres across the sample, and the authors note the range could in principle extend toward the second and third telecom windows in thicker MoTe₂ flakes — a point to which we will return.
Why the photons are pure
Purity is where the paper’s numbers are strongest. Under ordinary above-band excitation at 650 nanometres, the emitter is decent but unremarkable: fitted g²(0) values of 0.18 ± 0.07 (above-band) and 0.17 ± 0.07 (quasi-resonant at 1080 nm), with bunching on 10–15 nanosecond timescales from blinking.
The improvement comes from exciting smarter. Using photoluminescence excitation spectroscopy on an hBN-encapsulated emitter, the team mapped the emitter’s higher-energy resonances and found a p-shell-like state at about 1108 nanometres. Exciting through that resonance suppresses the background almost completely and improves the saturated brightness by a factor of 3.3. The triggered, pulsed g²(0) falls from 0.16 ± 0.02 under quasi-resonant excitation to 0.010 ± 0.004 under p-shell excitation: a more than tenfold improvement. Two further emitters measured under p-shell excitation returned consistently low values, so the effect is not specific to one device. “Triggered” here means what it should: the emitter is driven by a pulsed laser, one pulse at a time, and the antibunching is measured pulse by pulse — a photon on demand, not a photon inferred from a continuous stream.
Hexagonal boron nitride encapsulation plays a supporting but honest role. Encapsulated emitters keep their performance across multiple cooldowns over extended periods, while unencapsulated devices degrade with ambient exposure — stability, not better physics, is what the encapsulation buys. It does not reduce charge noise, and it changes the strain landscape from directional wrinkles to a more bubble-like morphology, weakening the polarisation anisotropy. The encapsulated emitters also show slightly longer lifetimes and limited internal quantum efficiency, about 31–33 percent.
The bias knob
Electrical control is the third leg of the result, and the one that most directly serves quantum networking, because networks need independent sources that can be tuned into resonance with each other. The unencapsulated bilayer flake sits in direct contact with one gold electrode, with a second gold electrode about 20 micrometres away laterally as a counter-electrode.
As the bias is swept from −2 V to +15 V, the emission line redshifts from about +4 V onward, with a total tuning range exceeding 3 nanometres — about 3 meV — enough to sweep one emitter’s line across a reference emitter of similar energy.
Bias also does two useful things to the dynamics. It suppresses blinking: bunching times fall from about 10.6 nanoseconds at open circuit to about 3.2 nanoseconds at +5 V and about 0.8 nanoseconds at +10 V, with the internal quantum efficiency rising from about 0.65 to about 0.80. And it shortens the lifetime, from about 440 picoseconds at 0 V to about 240 picoseconds at +15 V — a reduction of more than 45 percent. Because the measured linewidth stays at the resolution limit of about 150 µeV throughout, the ratio of experimental to transform-limited linewidth, R, falls from 100 at zero bias to about 55 at +15 V. R = 1 would be a Fourier-transform-limited emitter; 55 is still far from 1, but the direction of travel is what matters, and it is achieved with a voltage knob, not a new device.
The honest boundary
Three qualifications should accompany any coverage of this paper, and all three are visible in the paper itself.
First, the wavelength. The abstract frames the work as being about sources “operating at telecom wavelengths”; the introduction is more precise, calling MoTe₂’s emission “overlapping with telecom-relevant windows.” The measured range is 1090–1200 nanometres. That is near-infrared, and closer to telecom than the visible emission of WSe₂ — but it is not the telecom O-band at 1310 nanometres, and nowhere near the C-band at 1550 nanometres, where fibre loss is lowest. The authors state this plainly: “the wavelength range of our emitters does not yet reach the telecom band,” and point to thicker MoTe₂ flakes — a prior report observed emitters up to 1550 nanometres in multilayer flakes — as the route there. Until that extension is demonstrated with the same purity and tunability, MoTe₂ is a telecom-adjacent platform, not a telecom platform.
Second, the visibility. Seven percent raw, sixty percent post-selected. The post-selected figure is real physics — it shows the coherence is present in the early-arriving photons — but a source that must discard most of its output to reach 60 percent is not a 60-percent source. The paper’s own framing is careful: the abstract reports the 7.1 percent figure and the post-selected figure side by side, and the discussion calls the modest visibility the “principal limitation” of the device. The path the authors sketch to push indistinguishability beyond 50 percent — high-Q photonic structures for Purcell enhancement, plus stricter resonant or phonon-assisted excitation — is the standard prescription, and an admission that the present device, on its own, does not get there.
Third, the temperature. All measurements were performed at about 4 kelvin in a closed-cycle cryostat. That is standard for this kind of physics, and it is not a flaw — but it is a constraint on any deployment story. A source that needs liquid-helium temperatures lives in a cryostat, alongside the detectors it would feed.
None of these qualifications diminishes the result; they locate it. What the DTU team has done is take a material system whose emitters were previously irreproducible, slow, broad, and untunable, and produce emitters that are reproducible, fast, narrow, pure, and electrically tunable — and then measure, for the first time, how indistinguishable their photons are. The measurement returns a small number with a clear explanation and a credible route to improvement.
What to watch next
The follow-ups that would matter most are easy to name, because the paper names them. The first is a cavity: a MoTe₂ emitter in a high-Q photonic structure, where Purcell enhancement shortens the lifetime toward the coherence time and the raw HOM visibility is re-measured without post-selection. If that number moves from 7 percent toward 50 percent, the platform changes character. The second is wavelength: emitters in thicker MoTe₂ flakes reaching 1310 or 1550 nanometres with the same fabrication recipe, purity, and tunability — the step that would turn “telecom-adjacent” into “telecom.” The third is two-emitter interference: a HOM measurement between two distinct devices, rather than consecutive photons from one, is the experiment that would show whether the bias tuning is good enough for a network. The fourth is resonant excitation: strict resonant driving, or schemes such as SUPER and phonon-assisted excitation.
Two-dimensional emitters were always attractive for the wrong reasons — thin, integrable, easy to make — and held back for the right reason: their photons were not identical. The DTU result does not solve that problem. It measures it, for the first time, in the material most likely to matter for fibre networks, and it shows the problem is dephasing on a tens-of-picoseconds timescale, with every other property of the source already in good order. That is a narrower problem than the field faced a year ago — and a harder one, because dephasing is a property of the material’s environment, not of the device design.
Sources and further reading
- Paweł Wyborski, Athanasios Paralikis, Pietro Metuh, Martin A. Jacobsen, Christian C. Ruiz Madera, Niels Gregersen & Battulga Munkhbat, “Toward triggered generation of indistinguishable single photons from MoTe₂ quantum emitters,” Nature Communications (2026). DOI: 10.1038/s41467-026-78808-y. Published 11 November 2026; received 13 October 2025; accepted 5 October 2026. Primary source for all figures in this article. Preprint: arXiv:2508.20743 (v2, 6 October 2026).
- Prior MoTe₂ emitter work cited in the paper (Zhao et al.): long lifetimes (~20 ns to ms), linewidths above ~600 µeV, emitters observed up to 1550 nm in multilayer flakes, with no reproducibility, tuning, or two-photon interference reported.
- Prior TMD indistinguishability benchmark cited in the paper: WSe₂ emitters at about 2 percent — the figure this result improves on, and the reason the 7.1 percent raw visibility is described as state-of-the-art for the TMD platform rather than competitive with quantum dots.
- Note on verification: all quantitative values here are taken from the Nature Communications paper and its arXiv v2 text, which match the published abstract. A different Nature Communications paper published the same day, on single-photon reflection from a quantum dot in a waveguide (DOI: 10.1038/s41467-026-78984-x), is a separate result and was not used. No market-size figure, commercial claim, or investment judgement appears in this article; the site’s Terms carry the not-investment-advice disclaimer.