Electrically pumped single-defect light emitters in WSe₂

The recent observation of direct bandgaps in semiconducting molybdenum and tungsten dichalcogenide monolayers has led to a rise of interest to these two-dimensional (2D) materials and demonstrated their potential for future optoelectronic devices [1–4]. These one-monolayer-thick crystals are characterised by large exciton binding energies [5, 6] and oscillator strengths [7] and can be combined with other layered materials to create heterostructures held together by van der Waals forces [8–12]. This concept has been used to form electrically driven light-emitting structures, where MoX₂ or WX₂ (X = S or Se) monolayers were used as the exciton recombination layers, thin hexagonal boron nitride (hBN) was used for tunnelling barriers and graphene—used for transparent electrodes [10, 11].

In these devices, in addition to the pronounced 2D excitonic peaks that could be excited both optically and electrically, various (usually broad) features have also been detected 50 to 150 meV below the neutral exciton peak in both photo- and electroluminescence [1, 2, 10, 11, 13–17]. These were assigned to luminescence of localised states associated with defects in the monolayer semiconductors. In this energy range it was also possible to observe narrow photoluminescence (PL) features with linewidths 100–400 μeV particularly pronounced in WSe₂ at temperatures below 30 K [14–20]. The origin of the defect states giving rise to such PL response remains unknown. However, it has been shown for WSe₂ that photons emitted into the sharp defect PL peaks exhibit anti-bunching [14–19]. This provides a unique single-photon source contained within an atomic monolayer of a solid state material, which opens the way for integration of such defect light emitters with various photonic structures such as photonic crystal [21], microdisk [22] and scanning cavities [23] as well as waveguides.

Here we demonstrate electrical pumping of such defect states in graphene/hBN/WSe₂/hBN/graphene heterostructures previously used to generate electroluminescence (EL) from 2D excitons. We identify the regimes where EL of single defects dominates the emission spectrum of the device as the filling of the localised states occurs at a lower bias voltage (V_b) than for the 2D states. Overall, EL appears at applied bias around −2.1 V, close to the expected single-particle bandgap in WSe₂. We also show that our device can be used to fine-tune the energy of the defect emitter by varying the bias, constituting observation of the quantum confined Stark effect [24, 25].

The upper part of figure 1(a) shows a schematic of the van der Waals heterostructure that forms the EL device. Here, the graphene layers serve as transparent electric contacts and the hBN layers are used as tunnelling barriers that separate the graphene contacts from the active material, a WSe₂ monolayer. The lower part of figure 1(a) shows a microscope image of one of the heterostructures used in this work, where graphene contacts and a WSe₂ monolayer are clearly visible. In this image the hBN layers have a low contrast and are invisible, but they can be imaged using various optical filter arrangements and dark field microscopy.

Zoom In Zoom Out Reset image size

A schematic of the band structure for a finite V_b applied between the graphene contacts and corresponding to the onset of EL is shown in figure 1(b). The quasi-Fermi levels in the graphene contacts shift with increasing voltage. As the quasi-Fermi level of the left contact reaches the minimum of the conduction band of WSe₂, electrons tunnel through the hBN barrier. Holes tunnel through the other hBN barrier (on the right) when the quasi-Fermi level of the corresponding contact reaches the maximum of the valence band of WSe₂. Upon injection of both electrons and holes into WSe₂ layer the formation of excitons becomes possible. A typical IV-curve for this EL device is shown in figure 1(c), where the EL onset voltage is observed at the 'kink' in the curve for a bias of −1.9 V. The I–V shows some asymmetry possibly caused by differing doping of graphene contacts and differing thickness of the hBN barriers.

In addition, electron–hole pair excitation can be achieved in a wide voltage range using optical pumping above the bandgap, which creates excitons and electron–hole pairs relaxing into the lowest exciton states and giving rise to PL. In our experiments we use a typical micro-PL arrangement, where for PL measurements a laser is focussed in a 2 μm diameter spot on the sample by a short focal length lens positioned above the sample in a gas-exchange cryostat at a temperature of 4.2 K. The lens has a high numerical aperture of 0.55 enabling efficient collection of PL and EL from the sample. The PL and EL are then sent through a single-mode fibre to a spectrometer and measured with a charge coupled device. The single-mode fibre effectively acts as a spatial filter collecting light from a 2.5 μm diameter spot on the sample. In PL measurements described below we used a laser with photon energy 1.94 eV and the laser power was kept at 300 (30) μW for the data in figure 1 (figure 2).

Zoom In Zoom Out Reset image size

A typical PL spectrum of a monolayer WSe₂ is shown in figure 1(d). The PL features X⁰ and X⁻ are attributed to neutral exciton and trion peaks, whereas the features P0–P3 correspond to PL from the localised exciton states. Previously single photon emitters have been found in the whole range of energies where we observe the features P0–P3 [14–19]. The PL bias dependence is shown in figure 1(e). For biases between −1.5 and −1.9 V the quasi-Fermi level of the top graphene layer reaches the minimum of the conduction band resulting in supply of additional electrons. This leads to an increase of the trion contribution in PL just before the onset of EL. At V_b < −1.9 V, the X⁻ intensity drops, but overall the sample luminescence becomes significantly brighter with particular enhancement at lower energies below 1.67 eV. This corresponds to a situation when the tunnelling of both electrons and holes is possible from the graphene contacts. In this regime, the luminescence in the sample is excited mainly through electrical injection. In order to compare with typical PL, an EL spectrum in this regime is shown in figure 1(f). The main differences are in the lower intensity of both X⁰ and X⁻ relative to very pronounced low energy peaks, with the strongest intensity observed for P1. Similar behaviour is observed for positive bias.

At a lower laser power of 30 μW in some locations on the sample it is also possible to isolate narrow PL peaks having linewidths down to 100 μeV. Figure 2(a) shows an example of such peaks in the energy range corresponding to the localised state PL. We refer to these peaks as single-defect emitter (SDE) lines. The measurements were carried out on a number of such peaks, and for clarity we label them on the graphs as SDE1, SDE2 etc. The density of such defects is extremely low. We find a total of around 10 SDEs in our sample giving an areal density of less than 10⁷ cm⁻². The SDE emission is strongly localised, with the localisation size determined by the laser spot diameter of about 2 μm. The PL intensity of the peak falls by about a factor of 2 when the sample is moved by 1 μm.

Similarly to the observation in the III–V quantum dots [26], pairs of orthogonally-polarised PL lines are typically observed. The splitting between the lines is typically about 1 meV, with the low energy component of the doublet being more pronounced. Figure 2(b) shows that the doublet has a strong linear cross-polarisation with a polarisation degree of 90%. The observed splitting is possibly a consequence of a low symmetry of the defect, whereas the difference in intensity may be explained by relaxation with acoustic phonon emission efficient at low temperature [27].

By varying the bias V_b, steady shifts of some defect PL lines have been observed as shown in figures 2(c)–(e). The ability to spectrally tune a quantum emitter is essential for potential quantum interference experiments [28]. However, in our case a range of different behaviours is observed with different peak energy shift rates, appearance and disappearance of new lines in narrow voltage ranges etc. We believe that the different behaviour as a function of voltage arises from the variation of the environment, charge and the size of the electronic wave-function from defect to defect. The intensity switching may be associated with voltage-dependent charging effects. This is similar to the significantly more studied III–V quantum dots, where abrupt intensity thresholds related to dot charging were found [29]. The optical properties of these defects such as the spectral range where they are observed, linewidth and lineshapes, temperature and polarisation dependent behaviour are similar to the WSe₂ defects reported in the literature, which makes us believe that the defects studied here are of a similar nature to the reported structures.

Figures 2(d) and (e) show in more detail a particular voltage-dependent behaviour for a bright PL peak at 1.632 eV labelled SDE4. The peak exhibits a linewidth of 245 μeV at −2.5 V. As the voltage is changed a weaker PL feature appears around 227 μeV above the original peak. In addition to the energy shift with a rate of about 0.4 meV V⁻¹ of the higher intensity peak, fluctuations in the photon energy are observed including discontinuities around ±1.5 and 0 V. This may be related to a long-term recharging of the defects in the vicinity of this light-emitting centre.

From the trend in the peak position we can estimate the permanent dipole formed by the electron–hole pair localised on the defect (figure 2(d)). The change in the emission energy is given by ΔU = −pΔV/d, where p is the electric dipole, ΔV is the change in the applied voltage and d is the distance between the graphene contacts. We obtain that p/ed = 0.4 × 10⁻³, which for d ≈ 3 nm gives p/e ≈ 1.2 pm indicating a vanishing vertical displacement of the electron and hole wavefunctions. This estimation has been enabled by a well-understood distribution of the electric field in our structure, where the most significant contribution is expected to be from the vertical electric field.

We will now discuss the results of the EL measurements on SDEs. The panels of figure 3(a) show typical EL spectra observed for a defect emitter SDE7 for different applied voltages (no laser excitation is present here). Surprisingly, there is a regime where the signal from the SDE7 dominates the EL spectrum as shown in the bottom panel for V_b = −2.15 V, where a single sharp line is observed at 1.607 eV. As the bias is increased, additional broad features first appear at energies around the emitter, where luminescence from localised states is expected (e.g. P3 in figure 1(c)). For lower biases other features become dominant with most prominent peaks P1 and P2. At around −2.3 V, trion EL also becomes visible around 1.7 eV. The peak intensity of the SDE7 line remains high (much stronger than for the X⁻ line) and still comparable with the dominant features P1 and P2 up to the bias of −2.5 V. We find that EL from SDE7 peak and other SDE lines is localised in an area on our sample comparable with the spatial resolution provided by the single-mode fibre in the light collection path, in agreement with the assumption that this signal comes from a single localised defect site.

Zoom In Zoom Out Reset image size

The purity of a single-photon emitter (the depth of the anti-bunching curve) depends on how efficiently other luminescence can be suppressed, i.e. ideally a spectrally isolated emitter is required [30]. Although in our case the SDE7 EL is superimposed on a broad background of EL from other states, there are bias regimes where its EL is the strongest. In order to demonstrate this further, in figure 3(b) we show bias dependences of the normalised integrated intensity for the SDE7 peak, EL of the localised states (integrated up to 1.635 eV) constituting the background around the SDE7 peak, and EL above 1.635 eV including P2, P1 and X⁻. Up to about −2.15 V both the localised states emission and the high energy features are practically not observed in EL, whereas the SDE peak intensity is already considerable. Below −2.2 V, the high energy EL grows fast with bias, whereas the SDE EL gradually saturates and then decreases for V_b < −2.35 V.

Figure 3(c) shows a typical bias dependence of the full-width-at-half-maximum (FWHM) of the SDE7 EL peak, with a minimum of 0.6 meV observed at low biases where EL emerges and a subsequent increase in the saturation regime above −2.3 V to 1.4 meV. Typically SDE EL has an asymmetric shape, with a pronounced sharp maximum mostly contributing to the FWHM (see figure 3(d)). The main maximum in the SDE spectra is also accompanied with a high energy shoulder about 1 meV above and with a more pronounced low energy shoulder 2–2.5 meV below. While the high energy shoulder may be caused by the fine-structure splitting observed in figure 2(a), the origin of the low energy shoulder is not known. The low linewidth of the main maximum may allow efficient spectral filtering to improve the purity of the single photon source based on the SDEs. To demonstrate the potential of this approach, in figure 3(d) we show an SDE8 EL spectrum that is filtered using a 2 nm bandpass filter, which is sufficient to remove most of the background EL.

In conclusion, both photo- and electroluminescence from spatially and spectrally isolated SDEs has been observed in van der Waals heterostructures using WSe₂ as the optically active material. In the EL regime SDEs with linewidths of around 0.6 meV are most efficiently pumped at low biases close to the onset of the EL in the whole structure, where their luminescence is observed with little background. At higher biases, presence of emission from other localised states leads to a stronger background in EL, which would result in the reduced anti-bunching contrast. This effect could be partially suppressed by spectral filtering before light reaches the detection system. We also observe tuning of the single-defect emitter frequency by applied bias, an important feature in quantum interference experiments. This tuning enables estimation of the permanent electric dipole p/e ≈ 1.2 pm of the electron–hole pair associated with the defect emission.

At the last stages of preparation of this manuscript we have become aware of two papers where similar electrical pumping has been observed in TMDC/graphene heterostructures of a different design [31, 32].

Acknowledgments