Solid-state quantum emitters are promising building blocks for quantum technologies such as high-precision sensing and secure communications (1). In particular, atom-like emitters embedded in a host crystal [e.g., color centers in diamond (2) and silicon carbide (SiC) (3)] combine several appealing properties, namely, spin-selective optical transitions (2), room temperature stability (4), exceptionally long coherence times (4), and potential for scalability (5). However, on demand integration of quantum emitters with well-defined optical characteristics remains an unsolved challenge in solid-state quantum photonics. The development of these key capabilities requires achieving high-precision spatial placement; controlling the mesoscopic environment to avert variability between emitters; identifying and eliminating decoherence channels; and developing high-fidelity, scalable pumping schemes that are amenable to on-chip integration, such as electrically driven optical emission (1).
Two-dimensional (2D) materials are emerging as a powerful platform for hosting next-generation quantum emitters (1, 6), excelling in synthetic flexibility (7), high photon extraction efficiency (1), and tunability through external gates and substrate engineering (8). They also provide unparalleled control of emitter placement (9) and integration with photonic (10) and plasmonic (11) nanocavities. Various quantum emitters have been recently discovered in semiconducting transition metal dichalcogenides (TMDs) such as tungsten diselenide (WSe2) (12–17) and molybdenum diselenide (MoSe2) (18). While the exact origin of the emission is still actively debated, the single-photon emission has been attributed to a quantum dot–like confinement of excitons down to tens of nanometers (14), for instance, due to residual strain from the material transfer process (12). These quantum dot–like mesoscopic 2D emitters are structurally not well defined and exhibit large spectral variability, thus precluding their identical generation and precise spatial placement. These are the main challenges that prevent a systematic development and deployment of quantum emitters in 2D materials.
Here, we demonstrate photon emission from individual atomic defects in epitaxially grown monolayer tungsten disulfide (WS2) driven by local electron injection from a Au-coated tip. We correlate the atomic structure, electronic structure, and photon emission of the exact same defects with atomic spatial resolution. Specifically, we demonstrate on demand photon emission from intrinsic tungsten substitutes and deliberately created sulfur vacancies (VacS) by local electrical stimulation. We use a plasmonic scanning tunneling tip to drive optical transitions by sequential inelastic tunneling, tune the emission via the applied bias voltage, and map the photon emission from individual defect orbitals with the highest spatial resolution so far reported on par with that obtained using conventional elastic scanning tunneling microscopy (STM).
RESULTS AND DISCUSSION
Single-defect luminescence in monolayer WS2 is driven by electron tunneling from a plasmonic scanning probe tip (Fig. 1A). Through precise tip positioning, assisted by an extremely local electron transfer mechanism, we can selectively excite a single defect, free from any background luminescence of the host crystal. Electrons injected from the continuum of metallic tip states into discrete defect states of the 2D material generate broad optical emission spectra, with associated photon energies corresponding to the difference between initial (tip) and final (defect) electron energies in the tunneling process (Fig. 1B). The metal tip acts as a plasmonic antenna that assists the coupling between the inelastic tunneling current and the far-field light emission. This allows us to control the emitted photon energy through the applied tip-sample bias voltage, resulting in emission rates directly proportional to the electron density in the defect orbital right under the tip position.
Our single-atom system is fundamentally different from previous reports of mesoscale emitters in 2D materials. All atomic emitters in our study are identical because they have an atomically well-defined structure and very localized electronic wave functions that are much less susceptible to long-range disorder, such as strain or neighboring structural defects. Our observations are fully reproducible when looking at different defects of the same nature. In our experiments, we only observe local relaxation of the lattice in the immediate vicinity of the defect (19). Mesoscopic strain potentials can be excluded because the samples are grown by van der Waals epitaxy and not obtained by exfoliation (19, 20). In addition, atomic defects in TMDs can be generated with high-density control and specificity, for instance, through chemical doping or by annealing, as we show in this work.
We use the Au-coated tip of a STM to inject charge carriers and mediate the coupling between optical near and far fields via tip plasmon modes. This so-called STM luminescence (STML) emission (21) has been successfully exploited in studies of metallic surfaces (22) and molecular systems (23) to beat the light diffraction limit by more than two orders of magnitude due to the extreme localization of initial and final electron states in the tunneling process. In this context, electroluminescence from single molecules has been recently established through self-decoupling (24) or by introducing ultrathin insulating layers (25) between the molecule and a noble metal substrate. In addition, STML has enabled vibronic spectroscopy with submolecular resolution (25, 26), imaging molecular orbitals through photon emission maps (27, 28), studying charge and exciton dynamics (24, 29, 30), and stimulating spin-selective optical transitions (31). STML on indium tin oxide–supported MoSe2 has also been recently reported and attributed to radiative decay of the A exciton in this material (32).
Plasmonic noble metal substrates, which are commonly used in STML, are not a viable option to investigate TMDs because strong hybridization quenches the intrinsic optical emission (33). Instead, we use epitaxial graphene grown on SiC as a substrate, which has been shown to preserve the native TMD band structure (20). While electroluminescence from graphene has been previously observed (34), we find conclusive evidence to assign the optical emission in our TMD/graphene heterostructure to the electronic states of the TMD alone.
We prepare our samples by growing monolayer WS2 islands on epitaxial graphene on SiC using chemical vapor deposition (20). The as-grown sample contains several substitutional atomic defects, such as chromium (CrW) and molybdenum (MoW) replacing tungsten, as well as oxygen substituting sulfur (OS) (19, 35). Then, we selectively generate VacS in both the top and bottom sulfur layers by high-temperature annealing in vacuum (36). Both CrW and VacS defects exhibit unoccupied electronic states placed a few hundred millielectronvolts below the WS2 conduction band edge (Fig. 2B) (19, 36). Elastic tunneling into these in-gap defect states gives rise to pronounced resonances in the differential conductance (dI/dV) spectra measured in scanning tunneling spectroscopy (STS), which roughly yields the local density of states, as shown in Fig. 2E. Their characteristic electronic spectrum results from a combination of crystal field splitting, spin-orbit coupling, and electron-phonon interaction (19).
At tunneling voltages exceeding 1.5 V, we observe electron-induced photon emission originating from inelastic electron tunneling on the CrW and VacS defects, as well as on defect-free WS2 locations. The emission rate strongly depends on the applied bias and position (e.g., on or off a defect, as well as different regions within the defect) where STML spectra are acquired at constant current, as shown in Fig. 2F. We find a clear correlation between the bias onset for photon emission and the energy of the lowest unoccupied states observed in dI/dV. In particular, the difference between the bias onset of VacS (2.2 V) and CrW (2.4 V) is 0.2 V, which agrees with the energy difference of their respective defect states. Similarly, we observe STML at negative bias polarities (hole injection), for which the emission onset relates to the energy of the highest occupied state (see fig. S7). The emission scales linearly with the current, thus suggesting a single-electron process. In what follows, we focus on photon emission from sulfur top vacancies. Similar results for CrW can be found in the Supplementary Materials.
The extremely localized excitation by tunneling electron injection allows us to record atomically resolved photon maps. In Fig. 3A, we show the spectrally integrated photon emission as a function of lateral tip position over a single VacS in the top sulfur layer. The subnanometer-resolved photon maps acquired at high biases closely resemble the STM image of the in-gap defect orbital (Fig. 2C). The emission does not correlate with the simultaneously acquired STM topography at high bias (Fig. 3C), therefore excluding any substantial effect of the slightly varying gap distance on the spatial variation of the emission. Laterally, the defect emission is closely confined within ∼1 nm, concurrent with the electronic orbital dimensions (see Fig. 3B).
The close resemblance between the STM map resonant with the defect state and the STML map is further supported by theory. Photon emission in STML is mediated by the transition dipole p associated with the tunneling electron, acting as a radiation source (37) and giving rise to an emission yield ∝∣p∣2, which is, in turn, proportional to the elastic tunneling current measured in STM (see details in the Supplementary Materials). This current alone bears a dependence on the sampled final state, and therefore, both elastic and inelastic tunneling rates are proportional to the electron density of states at the tip position, given by the defect orbital density ∣ψf∣2. Similar to the enhanced emission rate of an excited atom in a resonant cavity defined by the Purcell factor P(ω), the spectrally resolved STML rate is directly proportional to P(ω), which, for metallic tips, is dominated by light-plasmon coupling and can be observed through optical spectroscopy.
Spectrally resolved STML measurements on the sulfur vacancy defect shown in Fig. 4 reveal broadband photon emission. The large spectral width originates from the finite energy range of tunneling electrons from the metallic tip (cf. Fig. 1B) and is not related to the intrinsic broadening of the defect state or the plasmonic near-field amplification by the tip. We find that the detected far-field spectra generally extend to 2.0 to 2.3 eV, which is consistent with the calculated field enhancement P(ω) between a spherical Au tip and the SiC substrate (fig. S3). At low photon energies, losses in the W tip (∼1 eV) and reduced quantum yield of the photon detectors (∼1.2 eV) produce a cutoff in the spectrum. The Purcell enhancement, and hence the spectral shape of the luminescence, is determined by the tip morphology and material. While broad luminescence results from the continuum of tip states and not the plasmonic enhancement, the latter weights the inelastic electron transitions by the photon density of states, which can introduce changes in the detailed spectral profile and actual intensity of the emission. We refer the reader to the Supplementary Materials for a detailed theoretical analysis of the near-field enhancement as a function of tip material and morphology, as well as luminescence spectra recorded with different tips including a comparison of W and Ag bulk tip wires.
We can identify two distinct regions in the plot of the emission intensity as a function of photon energy and tunneling bias (Fig. 4B): the high-bias, low photon energy corner (top left) is associated with substrate emission from WS2 (i.e., tunneling to substrate states), while the distinct emission band at lower bias is associated exclusively with the defect. For VacS, two emission steps (marked by white arrows in Fig. 4B) are observed, which follow a linear relation between applied bias and photon energy. Each of these steps corresponds to the opening of a new radiative decay channel, which we attribute to a transition from tip states at or below its Fermi level into the two unoccupied sulfur vacancy states. At the threshold energies indicated by the white arrows in Fig. 4B, the entire potential energy of the tunneling electron across the vacuum barrier is converted into a photon with matching energy, hence suggesting a single-electron to single-photon conversion mechanism. Figure 4C shows these transitions for a specific photon energy ħω = 1.85 eV, with each new decay channel indicated in the insets. Emission of a single photon associated with a single-electron transition is the dominant lowest-order radiative process. Accordingly, the highest allowed photon energy is given by the difference between the tip Fermi energy and the defect state energy.
For our Au-coated tungsten tips, we detect ∼10−7 far-field photons per electron. Accounting for all setup related losses, the intrinsic quantum efficiency of the radiative tunneling process is estimated as Y ∼ 10−4 photons per electron. It is interesting to compare this number with the analytical result (see Supplementary Materials) Y ∼ (3πα/2)(d/λ)2(Δω/ω)P(ω), where ω, λ, and Δω are the photon frequency, wavelength, and bandwidth, respectively; d is the tip-sample distance; and α ≈ 1/137 is the fine-structure constant. The Purcell factor P, which scales linearly with the density of available optical states dominated by plasmon modes in the tip-sample nanocavity (38), can be estimated by invoking electromagnetic reciprocity, asserting that the emission intensity is strictly proportional to the enhancement in the near-field intensity under external illumination at the position of the emitting dipole. The latter depends on the precise tip-sample morphology but is rather insensitive to the lateral position of the tip in the nonmetallic SiC substrate under consideration and reaches plasmon-enhanced values of P ∼ 104 (see the Supplementary Materials). Using parameters corresponding to our experimental conditions, we then predict a photon yield per tunneled electron Y ∼ 10−4 in good agreement with the experimental estimate.
Ultimately, the photon generation rate is fundamentally limited by Coulomb blockade, which prevents tunneling of subsequent electrons into the defect state before the previous electron drains to the graphene substrate. Accordingly, the correlation time of single-photon emission events as derived by the antibunching in a g(2)(t) experiment is expected to yield this Coulomb blockade time, as also suggested by recent STML experiments for C60 multilayers on Ag (39). From the dI/dV linewidth of the vacancy state of 8 meV, we estimate a lower bound of the transient charging time of the defect on the order of 100 fs. This number is well beyond the time resolution of state-of-the-art timing electronics in a g(2)(t) measurement. Hence, the single-photon character in our experiments can only be inferred indirectly through the sequential single-electron tunneling and the matching electron-to-photon conversion energy. We note that the Coulomb blockade time could be exponentially increased by introducing a tunneling barrier between the TMD layer and the graphene substrate, to attain a measurable g(2)(t) lifetime in the future. Conversely, a short Coulomb blockade time enables a high single-photon generation rate. In our case, a 100-fs lifetime and 10−4 conversion yield suggest that a 109 Hz single-photon generation rate should be attainable. This number can be even increased using optimized plasmonic or optical cavities to enhance the Purcell factor.
Acknowledgments: Funding: This work was supported as part of the Center for Novel Pathways to Quantum Coherence in Materials, an Energy Frontier Research Center funded by the U.S. Department of Energy, Office of Science, Basic Energy Sciences. This work was performed at the Molecular Foundry supported by the Office of Science, Office of Basic Energy Sciences, of the U.S. Department of Energy under contract no. DE-AC02-05CH11231. B.S. acknowledges support from the Swiss National Science Foundation under project number P2SKP2_171770. C.K. acknowledges support by the Bavaria California Technology Center (BaCaTeC) and the International Graduate School of Science and Engineering (IGSSE) via project “CommOnChip.” A.W.-B. was supported by the U.S. Department of Energy Early Career Award. F.J.G.A. acknowledges support from the Spanish MINECO (MAT2017-88492-R and SEV2015-0522), the European Research Council (advanced grant 789104-eNANO), the Catalan CERCA Program, and the Fundació Privada Cellex. Author contributions: B.S. and A.W.-B. conceived the experiments. B.S. and K.A.C. conducted the SPM measurements. C.K. and A.M.S. synthesized the samples. B.S., E.S.B., E.W., and N.J.B. constructed the optical detection setup. F.J.G.A. derived the theoretical framework. All authors discussed the results and contributed to the manuscript. Competing interests: The authors declare that they have no competing interests. Data and materials availability: All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials. Additional data related to this paper may be requested from the authors.