There is a problem that has haunted two-dimensional quantum optics for a decade, and it is a problem of scale. On one side sits the atomic structure of a defect: a missing sulfur atom, an oxygen atom in the wrong lattice site, a carbon–hydrogen pair wedged into the crystal. On the other sits the light that defect emits, collected from a spot a thousand times wider than the defect itself. Connecting the two has mostly been a matter of inference, and near-field techniques have only narrowed the gap to tens of nanometres, still hundreds of atoms.
A paper published in Nature Nanotechnology today closes that gap, for one material, in the most literal way available: it puts a scanning tunnelling microscope tip directly above a single, atomically identified defect and reads the light it emits. Lysander Huberich, Eve Ammerman and colleagues — at Empa, Shanghai Jiao Tong University, ETH Zürich and Penn State — report atomically resolved exciton emission from monolayer MoS₂ using scanning tunnelling microscopy-induced luminescence, or STML. Three common point defects, each with its own optical fingerprint, and two of the three turn out to do the opposite of what the literature has long assumed. Every number below comes from the open-access paper itself.
The physics first: why an exciton goes dark on graphene
Put a monolayer of MoS₂ directly on graphene and you get a combination attractive for tunnelling microscopy — graphene is conductive, so the tip can inject charge, and the MoS₂ keeps its direct band gap — but hostile to light. The exciton’s wavefunction hybridises with the graphene beneath it, and any hot exciton transfers its energy non-radiatively into the substrate before it can emit a photon. The emission is quenched, and what little remains is drowned by the tip’s own plasmonic glow.
Huberich and colleagues insert a spacer: a bilayer of hexagonal boron nitride, hBN, between the MoS₂ and the graphene. Two atomic layers is a narrow compromise — thick enough to suppress the non-radiative decay, thin enough that electrons can still tunnel through it and inject charge. The paper quantifies the compromise in two ways.
First, the decoupling is electronic and visible in tunnelling spectroscopy: the MoS₂ band gap grows by approximately 30 meV when the bilayer hBN is present, attributed to reduced dielectric screening from the graphene substrate. Second, the charge-state lifetime of a defect lengthens by a factor of thirty. The evidence comes from z spectroscopy, where the tunnelling current through a defect state saturates once injection outpaces drainage, so the saturation current encodes the lifetime. On graphene alone, sulfur vacancies show saturation currents above 100 nA, implying carrier lifetimes on the order of 1 picosecond. On bilayer hBN, the saturation current falls to about 5 nA, corresponding to a lifetime of about 30 picoseconds — long enough for the injected charge to do optical work before it drains away.
That work is what the STML measurement reads. At negative sample bias, the tip injects holes into the valence band; electrons tunnel in from the substrate to neutralise the charged state, and the resulting exciton recombines radiatively inside the plasmonic nanocavity formed by the tip and the sample. The emission intensity scales linearly with the tunnelling current, as expected for a single-electron process, and the emission onset coincides with the valence band maximum. No excitonic emission appears at positive bias, where the electron attachment energy is insufficient to reach the excitonic state.
The pristine spectrum: an exciton, two trions, and a seven-fold brightening
On a clean region of the sample, the STML spectrum looks like what low-temperature photoluminescence of high-quality MoS₂ should look like, which is itself the point: the tip is not inventing the physics, it is reading it locally. The main line is a sharp emission centred at 1.959 eV, assigned to the neutral A exciton, X. Two weaker side peaks sit at 1.924 eV and 1.905 eV, tentatively assigned to trion species, X₁⁻ and X₂⁻, on the basis of correspondence with literature data and recent magneto-photoluminescence work. A broad feature at 2.095 eV is attributed to the B exciton.
The trion assignment is explicitly tentative — the authors’ word — because the low-energy fine structure of MoS₂ photoluminescence is complicated by the small conduction-band splitting and by many-body effects. It is a reasonable correspondence argument, but a careful reader should keep the qualifier attached.
The decoupling layer’s payoff is most visible in a direct comparison. MoS₂ on monolayer hBN also emits, so a single layer is already enough to decouple the exciton from graphene. But on bilayer hBN the emission is seven times brighter. Seven-fold is not a small factor, and it is the number that makes the rest of the paper possible: without it, the defect-level spectra that follow would be lost in noise.
Three defects, three fingerprints
The defect work is where the paper earns its title. The as-exfoliated MoS₂ carries a very low cumulative defect density — 6.1 × 10¹¹ cm⁻², consisting primarily of oxygen substitutions, which are isovalent and therefore invisible to STM. To study the defects that matter for quantum emission, the team deliberately introduced two more species by mild argon sputtering in ultrahigh vacuum: sulfur vacancies, Vac_S⁻, and negatively charged carbon–hydrogen complexes, CH_S⁻, the latter common in synthetically grown transition metal dichalcogenides. Each defect then gets the same treatment: image it atomically, measure its electronic states, and record the STML spectrum with the tip directly above it.
The oxygen substitution is the quietest case. It has no in-gap defect states, so it produces no additional emission lines; it merely reduces the intensity of the pristine exciton and trion emission, suggesting additional non-radiative decay channels.
The sulfur vacancy is the surprise, and the paper is explicit that the authors found it surprising. Sulfur vacancies have often been proposed as the origin of single-photon emission in far-field optical measurements of MoS₂. Here, the vacancy has distinct in-gap defect states — the STM can see them clearly — but it produces no excitonic emission lines at the defect site within the detectable energy range above 1.1 eV. The pristine emission is suppressed, as with oxygen, and the only signal directly above the vacancy is broad plasmonic emission. The vacancy, in this experiment, is a quencher, not an emitter.
That negative result is the paper’s most consequential claim, and it should be read with the precision the authors give it. It does not say sulfur vacancies never emit. It says that an isolated, negatively charged sulfur vacancy in monolayer MoS₂ on bilayer hBN, measured by STML at liquid-helium temperature, does not produce a defect-bound exciton line in the accessible range, while suppressing the pristine emission around it. The far-field signatures often attributed to sulfur vacancies may belong to something else — and the paper’s third defect is the candidate.
The carbon–hydrogen complex is the one that talks. Directly above a CH_S⁻ defect, a distinct emission line appears around 1.8 eV, approximately 200 meV below the pristine exciton, concurrent with the suppression of the pristine STML emission — a spectral response the authors note resembles the far-field photoluminescence signature often attributed to sulfur vacancies. The defect’s chemical identity is confirmed in a way that only an STM can manage: a voltage pulse of 4.5 V at 8 nA cleaves off the hydrogen, converting CH_S⁻ into C_S⁻, whose characteristic vibronic spectrum then appears in the tunnelling spectrum. The defect is identified not by inference but by surgery.
The authors assign the 1.8 eV line to a free-to-bound recombination, (e, A0): an electron in the conduction band recombining with a hole bound at the neutral acceptor site. In the paper’s many-body picture, a hole injected from the tip is rapidly trapped by the negatively charged acceptor, converting the charged ground state A⁻ into a neutral acceptor state A0 with a localised bound hole; subsequent electron tunnelling from the substrate populates the state (e, A0), and radiative recombination restores the charged ground state.
Subnanometre maps, and the tip’s own electric field
The most technically striking part of the paper is the spatial mapping. Subnanometre-resolved STML spectra acquired across the CH_S⁻ defect show that while the pristine exciton and trion energies stay virtually unchanged, the defect emission lines shift substantially — redshifting as the tip moves away from the defect centre. Beyond the main line, labelled D1, two higher-energy features, D2 and D3, appear; D1 and D3 follow similar curved trajectories, while D2 is flatter.
The photon maps make the structure visible. The MoS₂ exciton intensity is slightly enhanced 2.2 nm from the defect but fully suppressed at the defect centre, while the defect emission forms localised ring-like patterns — maxima at 1.2 nm for D1, 1.4 nm for D3 and 1.7 nm for D2 — over a spatial extent of 1–2 nm. The central suppression is geometric: the emitter’s transition dipole lies in the plane, the tip dipole is vertical, and when the two are perpendicularly aligned the signal cancels.
The large spectral shifts near the charged defect are not intrinsic to the defect at all — they are the tip’s doing. Tip-induced band bending generates a lateral potential well beneath the tip, estimated at ΔE_TIBB ≈ 340 meV at a bias of −2.8 V, assuming a 12 percent voltage drop across the heterostructure. The hole is repelled by the tip potential and can diffuse to the defect site, while the electron stays confined in the tip-induced quantum well. Displace the tip and only the electron energy is lowered, redshifting the emission; bring the tip directly above the defect and both states shift similarly, partially compensating the field effect and recovering an emission energy close to the field-free limit.
This contains a warning about the technique itself: the STML spectrum of a charged defect is not a passive readout, because the probe reshapes the electrostatic landscape it measures by hundreds of milli-electron-volts over nanometre distances. The authors model the effect and use it — the D2 and D3 lines are proposed to be excited electron states, a Rydberg-like series of (e₂s/₂p, A0) states defined by the non-local Rytova–Keldysh screening of monolayer MoS₂ — but the excited-state assignment is explicitly a proposal, and the quantification rests on a model of the tip potential rather than a measurement independent of it.
Why it matters, and what it does not yet show
The immediate significance is identification. For years, the low-energy emission band in MoS₂ photoluminescence — the so-called L-band — has been attributed, variously and inconclusively, to sulfur vacancies and other point defects. This paper ties a spectral feature at that energy to a specific, atomically identified defect, CH_S⁻, and simultaneously shows that the most commonly blamed suspect, the sulfur vacancy, does not emit in this configuration. That is a correction to the literature: you cannot engineer what you cannot identify, and the identification turns out to point somewhere the field was not looking.
What the paper does not show should be stated with equal care. First, this is a cryogenic, ultrahigh-vacuum experiment: the scanning probe measurements were performed below 5 K at pressures under 2 × 10⁻¹⁰ mbar. Nothing here is a device, and nothing here operates at room temperature. Second, the paper reports no photon statistics. There is no second-order correlation measurement, no g⁽²⁾(0), and therefore no demonstration that the defect-bound emission is a single-photon source. Any coverage that calls CH_S⁻ a single-photon source on the basis of this paper would be inventing a result the authors did not claim. Third, the trion assignments are tentative, as the authors say, and the D2 and D3 excited-state assignments are proposals. Fourth, the strong spectral shifts near charged defects are a tip-induced effect, modelled rather than independently measured. Fifth, the sulfur-vacancy negative result is specific to the configuration measured: an isolated vacancy on bilayer hBN at liquid-helium temperature.
None of these limits diminishes the result; they locate it. The paper’s contribution is a clean, atomically precise fact: in monolayer MoS₂ decoupled by bilayer hBN, oxygen substitutions and sulfur vacancies suppress the pristine exciton without emitting of their own, while a negatively charged carbon–hydrogen complex hosts a localised defect-bound exciton about 200 meV below the pristine line, with a spatial extent of 1–2 nm.
What to watch
Three follow-ups would each change the assessment in a different direction. The first is photon statistics: a Hanbury Brown and Twiss measurement on the CH_S⁻ emission, which would settle whether it is a genuine single-photon emitter or a localised but classical source. The second is generality: the same STML fingerprinting applied to other dichalcogenides — WSe₂, WS₂, MoSe₂ — and to deliberately created defects. The third is temperature: whether the defect emission survives as the sample warms toward device-relevant regimes, or dissolves into the thermal background.
That is the paper’s real theme: every defect has a fingerprint, and for the first time in MoS₂, the fingerprints can be read one atom at a time.
Sources and further reading
Huberich, L., Ammerman, E., Gu, Y., Ren, Y., Papadopoulos, S., Dong, C., Robinson, J. A., Watanabe, K., Taniguchi, T., Gröning, O., Novotny, L., Li, T., Wang, S. & Schuler, B. “Atomically resolved exciton emission from single defects in MoS₂.” Nature Nanotechnology (2026). Published 5 November 2026; received 5 January 2026, accepted 25 September 2026. DOI: 10.1038/s41565-026-02314-x. Open access. Corresponding author: Bruno Schuler. Affiliations: Empa; Shanghai Jiao Tong University; ETH Zürich; Université de Strasbourg; Penn State; National Institute for Materials Science, Japan.
For the far-field attribution of MoS₂ defect emission to sulfur vacancies that this paper revises, see Mitterreiter, E. et al., Nature Communications 12, 3822 (2021); and Klein, J. et al., Nature Communications 10, 2755 (2019). See also Barja, S. et al., Nature Communications 10, 3382 (2019).