Photonic quantum technology has an integration problem disguised as a materials problem. The field knows how to make good single photons, and it knows how to guide light on a chip. What it has not had is a single material family that does both well — a source and a circuit made of the same stuff, assembled without the elaborate hybrid fabrication that currently stitches different material systems together. A result published October 28, 2026, in Nature Communications, from researchers at the Technical University of Denmark, closes one specific gap in that picture: the first integrated quantum light source in which both the emitter and the waveguide are made entirely of van der Waals materials.
The paper, “Integrated on-chip quantum light sources on a van der Waals platform,” is by Pietro Metuh, Paweł Wyborski, Athanasios Paralikis, Frederik Schröder, Nicolas Stenger, Niels Gregersen, and Battulga Munkhbat, all of DTU’s Department of Electrical and Photonics Engineering, with Stenger also affiliated with DTU’s NanoPhoton center. Its claim deserves precision: the team showed that a strain-engineered bilayer tungsten diselenide (WSe2) emitter can be coupled efficiently into a tungsten disulfide (WS2) waveguide, with single-photon purity preserved through the waveguide and measured on-chip. It did not integrate a detector, a modulator, or an interferometer, and it did not measure photon indistinguishability. What it established is a source-plus-circuit platform — a milestone, not a finished component for a quantum processor. The work has been public since the preprint appeared on arXiv in December 2025; the Nature Communications version is the one to cite.
Why a single-material platform is the hard part
The vision the paper opens with is familiar: a chip that generates, routes, and detects single photons, with all the main building blocks on one substrate. The obstacle is that today’s best components come from incompatible material families — silicon or silicon nitride circuits, III–V emitters, amorphous-superconductor detectors such as NbTiN or MoSi. Putting them together requires complex hybrid fabrication with stringent thermal and chemical compatibility, and that burden has become a major bottleneck for scaling photonic quantum technologies.
Transition metal dichalcogenides — the layered semiconductors that include WSe2 and WS2 — offer a way around the incompatibility, in principle. Their layers are held together by weak van der Waals forces and their surfaces have no dangling bonds, so flakes of different materials can be exfoliated and stacked like sheets of paper, without lattice-matching constraints. Different members of the family bring different strengths: WSe2 can host localized quantum emitters created by strain or defect engineering; multilayer WS2 has a high refractive index and low absorption above 700 nanometers, suiting low-loss waveguides; superconducting NbSe2 has been explored as an ultrathin detector material.
The catch, and the reason this result is new, is that no previous demonstration had combined a van der Waals emitter with a van der Waals waveguide into a working, waveguide-coupled single-photon source. Earlier attempts suffered from weak or uncontrolled coupling: monolithic approaches lacked the strain needed to create emitters, while hybrid approaches were limited by the atomic thinness of mono- and bilayer materials, insufficient to reshape the propagating modes.
The device: an emitter draped over a waveguide
The device starts with the passive circuitry. WS2 flakes are exfoliated, selected for thickness and homogeneity, transferred onto silicon chips carrying a 2-micrometer layer of silicon dioxide, and patterned by electron-beam lithography and reactive-ion etching into multimode slab waveguides 500 nanometers wide and 150 nanometers tall. The characterized device uses a 60-micrometer-long waveguide with a 90-degree bend and a grating coupler at each end. The couplers — four concentric half-rings with a 780-nanometer pitch and a 55 percent duty cycle — were optimized for microscope operation, since the emitters are excited with an out-of-plane laser beam and collected by a high-numerical-aperture objective (NA = 0.81). Simulations put the mode-averaged extraction efficiency at 12.5 percent at 802 nanometers, peaking at 18.8 percent for the TM0 mode. That 18.8 percent figure matters later: it is the conversion factor behind the headline rate.
The emitter is then added by transfer, not by growth. A hexagonal boron nitride (hBN) layer under 5 nanometers thick goes onto the waveguide first, followed by a WSe2 bilayer. The hBN is not packaging but part of the device physics: it reduces the chance of the bilayer fracturing during transfer and changes the charge environment and strain profile in the WSe2. Bilayers transferred without it, the paper notes, coupled only weakly to the waveguide.
The emitters themselves are created by strain. Transferring the WSe2 onto the nanostructured waveguide produces edge-induced wrinkles, and the localized strain at those wrinkles confines excitons into quantum-dot-like states that emit single photons. The team finds them by scanning a focused laser along the waveguide at low temperature — the sample is cooled to about 4 kelvin — and watching for a narrow spectral line. Once a line is found, the collection spot moves to a grating coupler to confirm that the emission actually travels through the waveguide.
P-shell excitation: the quieter way to pump
Quantum emitters in WSe2 are often excited above the band gap, with a laser energetic enough to create free carriers that then relax into the emitting state. That approach works, but it also creates background emission from higher-energy states that pollutes the single-photon signal. The DTU team instead used resonant p-shell excitation: tuning the pump laser to a discrete higher-energy level of the emitter itself — here, 783 nanometers, about 10 to 30 nanometers below the 802-nanometer emission line, far enough for spectral filtering of the pump — so the emitter is populated directly, with far less background.
The comparison is stark. Under p-shell excitation, the integrated counts under the spectral line are roughly an order of magnitude higher at equal pump power than under above-band excitation, while the linewidth is comparable, if slightly narrower. The measured decay time drops from 55.0 ± 1.1 nanoseconds under above-band excitation to 31.1 ± 0.4 nanoseconds under p-shell excitation.
Polarization is preserved through the waveguide: the degree of linear polarization is 98 percent confocal to the emitter and 84 ± 18 percent at the upper grating coupler. Additional emitters found by scanning along the waveguide showed bright lines between 795 and 800 nanometers, collected through a grating coupler — a quiet argument for scalability, since the integration process is not a one-off accident of a single flake.
What the team actually measured: purity, twice
The central measurement is the second-order correlation function, g(2)(0), the standard test of single-photon purity. A perfect single-photon source never emits two photons at once, giving g(2)(0) = 0; anything below 0.5 demonstrates single-photon character. The team measured it in three configurations.
First, confocal collection at the emitter, with an external beam splitter before two superconducting-nanowire detectors: g(2)(0) = 0.043 ± 0.027. The emitter is a good single-photon source in free space.
Second, collection at one grating coupler — the light having traveled through the WS2 waveguide — with the same external beam splitter: g(2)(0) = 0.003, with asymmetric uncertainties of +0.030 and −0.003. Purity through the waveguide is not merely preserved; within the measurement’s precision it is essentially at the ideal limit, and the team attributes the improvement over the confocal value to slightly lower excitation power and better coupling of the emitter relative to background emission. Note the uncertainty: the upper bound of the interval is 0.033, so the honest statement is that the waveguide-coupled purity is very high, not that it is exactly 0.003.
Third, and most novel, the on-chip configuration: the external beam splitter is removed entirely, and a collection spot is placed at each grating coupler. Because the emitter couples into the waveguide in both directions, the waveguide itself acts as the beam splitter, and correlating photons detected at the two ports yields an on-chip Hanbury Brown–Twiss measurement — the first, to the authors’ knowledge, with a van der Waals single-photon source. The result: g(2)(0) = 0.076 ± 0.023. Purity survives the on-chip measurement, somewhat worse than the off-chip waveguide value but still far below the 0.5 threshold. It is a more demanding, more realistic test, and it passes.
The rates, and what they do and do not mean
Brightness is where the paper’s numbers need the most careful reading. During the correlation measurements, the detectors recorded approximately 20 and 15 kilohertz at the upper and lower grating couplers. Accounting for losses in the transmission setup, the team estimates the single-photon rate at the first lens — the first optical element collecting the out-coupled light — at approximately 320 kilohertz for the first output and 240 kilohertz for the second, under continuous-wave p-shell excitation.
The headline “1.7 megahertz waveguide-coupled rate” is derived from the 320-kilohertz figure, not measured directly. It divides by the simulated maximum grating extraction efficiency of 18.8 percent: 320 kilohertz out-coupled implies about 1.7 megahertz actually propagating in the waveguide, with 1.28 megahertz for the second direction. The paper notes this is conservative in one respect — fabrication imperfections and multimode coupling make the simulated extraction efficiency an upper bound, so the true rate could be higher — but it is still a model-dependent estimate, not a direct measurement. Quote it as an estimate, with the 18.8 percent assumption named.
The same accounting yields the paper’s efficiency figure: given the 31.1-nanosecond re-excitation time set by the measured decay, the total waveguide-coupled quantum efficiency is estimated at 9.3 percent — roughly one in eleven excitations producing a photon coupled into the waveguide. That is a respectable figure for a first integrated van der Waals device, and a long way from the near-unity coupling a practical source would want. The paper names the routes upward: electrical biasing for spectral stability and tuning, integration with a high-quality-factor optical cavity, and more advanced excitation schemes.
What this does and does not change
Placed in context, the result changes the status of the van der Waals platform as a quantum photonics candidate. Before it, the platform’s components existed separately, but no device had shown a van der Waals emitter feeding a van der Waals circuit with its quantum character intact. After it, there is a demonstrated source-plus-circuit unit, with purity measured off-chip and on-chip and multiple emitters coupling to the same waveguide.
What it does not change is equally important, and the paper is candid about most of it. Detectors and modulators are not integrated: the photons are still collected by an external objective and counted by external superconducting-nanowire detectors, and the paper’s own future-milestones figure lists calibrated beam splitting, single-photon detection, and interferometry as work to come. The device operates at about 4 kelvin, so nothing here addresses room-temperature operation, and the emitters are located by scanning rather than placed deterministically — a manufacturing gap for any scalability claim. The 9.3 percent quantum efficiency leaves most excitations unharvested.
One omission deserves emphasis because it is easy to miss in a story about single photons. The paper reports purity, not indistinguishability. No Hong-Ou-Mandel interference measurement is presented, so nothing here establishes that successive photons from this source are identical enough for interference-based protocols. Purity and indistinguishability are separate requirements, and this device has so far demonstrated only the first. Any retelling that describes these photons as “identical” or “indistinguishable” would be inventing a result the paper does not contain. Nor does the result settle which integrated-photonics platform will win: silicon and silicon nitride circuits with III–V emitters remain far more mature in fabrication, and the van der Waals platform’s advantages in versatility and assembly simplicity are still offset by yield, reproducibility, and cryogenic operation.
What to watch next
The most informative follow-ups are the ones the paper names: integrated NbSe2 detectors that count photons without an external objective; calibrated splitters, interferometers, and modulators in the same material family; a Hong-Ou-Mandel measurement of indistinguishability, with cavity integration raising brightness and indistinguishability together; deterministic emitter placement with predictable wavelengths, rather than emitters found by scanning; and efficiency gains beyond 9.3 percent quantum efficiency and 18.8 percent grating extraction.
The deeper lesson is one this desk has seen before in different materials. Excitonic systems are strongly coupled to light, which makes them bright and optically accessible, and strongly coupled to their environment, which makes them noisy and hard to control. Progress usually comes from engineering the structure around the emitter so the helpful coupling dominates — here, a waveguide draped with a strained bilayer, a spacer layer that quietly does essential work, and an excitation scheme that pumps the emitter without waking its surroundings. The DTU team did not build a quantum computer. They built the first light source in which the emitter and the circuit are made of the same atom-thin family, and showed the photons survive the journey through the circuit with their quantum character intact.
Sources and further reading
- Pietro Metuh, Paweł Wyborski, Athanasios Paralikis, Frederik Schröder, Nicolas Stenger, Niels Gregersen, and Battulga Munkhbat, “Integrated on-chip quantum light sources on a van der Waals platform,” Nature Communications (published October 28, 2026).
- Pietro Metuh et al., “Integrated on-chip quantum light sources on a van der Waals platform,” arXiv:2512.15337 (preprint, December 2025). All figures quoted above are taken from this paper.
- Note on verification: the Nature Communications publication date of October 28, 2026, and the author list are confirmed from the journal’s article listing. No DOI for the journal version is quoted because it could not be verified from the sources available for this article; the arXiv identifier above is the verified reference. No indistinguishability figure is quoted because the paper reports none, and the 1.7 MHz rate is an estimate derived from the simulated 18.8 percent maximum grating extraction efficiency, not a directly measured rate.