Direct exciton emission from atomically thin transition metal dichalcogenide heterostructures near the lifetime limit

We demonstrate the reduction of the inhomogeneous linewidth of the free excitons in atomically thin transition metal dichalcogenides (TMDCs) MoSe2, WSe2 and MoS2 by encapsulation within few nanometre thick hBN. Encapsulation is shown to result in a significant reduction of the 10 K excitonic linewidths down to ∼3.5 meV for n-MoSe2, ∼5.0  meV for p-WSe2 and ∼4.8 meV for n-MoS2. Evidence is obtained that the hBN environment effectively lowers the Fermi level since the relative spectral weight shifts towards the neutral exciton emission in n-doped TMDCs and towards charged exciton emission in p-doped TMDCs. Moreover, we find that fully encapsulated MoS2 shows resolvable exciton and trion emission even after high power density excitation in contrast to non-encapsulated materials. Our findings suggest that encapsulation of mechanically exfoliated few-monolayer TMDCs within nanometre thick hBN dramatically enhances optical quality, producing ultra-narrow linewidths that approach the homogeneous limit.

In the group of atomically thin two-dimensional (2D) materials the transition metal dichalcogenides MoS 2 , MoSe 2 , WS 2 and WSe 2 reveal fascinating photophysical properties owing to their direct gap and strong light-matter interactions 1,2 . The weak dielectric screening results in emission dominated by excitonic processes, with exciton binding energies on the order of several hundred meV 3,4 that follow a non-hydrogenic Rydberg series 5 . However, in the vast majority of reports to date the linewidths of the free excitons exhibit significant inhomogeneous broadening. This is typically attributed to the local spatial inhomogeneity of the substrate, adsorbed atoms and molecules on the surface due to the large surface-to-volume ratio and different doping and dielectric screening conditions that are highly sensitive to the choice of substrate. Broad linewidths of the exciton emission of ∼50 meV for MoS 2 6 , ∼40-75 meV for WS 2 7,8 , ∼5 meV for MoSe 2 9 and ∼10 meV for WSe 2 10 have been reported in photoluminescence experiments(PL), while time-domain spectroscopy [11][12][13] and recent theory 14 report homogeneously broadened luminescence linewidths in the range of ∼2-6 meV depending on the material system. The healing of sulphur defects using sulphuric superacids increases the optical quantum yield and reduces the linewidths at room temperature [15][16][17] from ∼70 meV to ∼55 meV. However, low temperature studies of treated MoS 2 monolayers 17 show that the linewidths still remain in the order of ∼15 meV. Very recently, it has been shown that MoS 2 is particularly sensitive to photo-induced irreversible changes resulting in broad luminescence from overlapping neutral and charged exciton emission 18 . Measurements performed using ultra-low excitation power densities reveal distinct peaks for neutral and charged excitons with linewidths of ∼15 meV for MoS 2 similar to Se-based TMDCs 18 .
In this letter, we present an optical study of TMDCs encapsulated within hBN and demonstrate that encapsulation leads to a significant reduction of the linewidth observed in photoluminescence (PL) experiments, towards the radiative limit. We systematically probe modifications in the luminescence linewidth after each stacking step and extract key parameters such as the exciton peak position, relative intensities of exciton and trion recombination and peak linewidths. We also show that annealing of the heterostructure improves the spatial homogeneity of the TMDC and, thus, of the observed luminescence. From our results, we make three major observations upon hBN encapsulation: (i) the linewidths of free excitons are significantly reduced down to a few meV approaching the homogeneous linewidth limit, (ii) the surface is protected, preventing samples against irreversible photo-induced spectral changes and (iii) encapsulation effectively lowers the Fermi level, reducing emission from negatively charged excitons in MoSe 2 , while increasing the emission from positively charged excitons in WSe 2 due to protection against physisorption and impurities from the substrate.

Results and Discussion
Photoluminescence of encapsulated MoSe 2 . To probe the impact of the proximal substrate and explore the benefits of hBN encapsulation, we performed spatially resolved PL measurements and statistically analyse the emission spectra at different positions on the sample surface. Note, in our analysis we disregard spectra recorded from the edge of the flake and obviously damaged parts of the sample, as identified by conventional light microscopy. From the measurements, we extract the peak positions, full widths at half maximum linewidths (FWHM) and relative intensities of the neutral exciton (X) and charged trion (T) by fitting with Lorentzian peaks. Figure 1 compares examples of spectra of MoSe 2 on SiO 2 , MoSe 2 on hBN, MoSe 2 on hBN after annealing and MoSe 2 sandwiched between hBN and after annealing. The corresponding statistical analysis of peak position, exciton linewidth and peak area for the different MoSe 2 /substrate configurations are shown in Fig. 1b-d, respectively. Note that in order to obtain the best comparison, in the case of MoSe 2 on hBN we scan the same area after subsequent steps of stacking and annealing to trace the impact of the encapsulation on the spectral evolution. A typical spectrum recorded from MoSe 2 on SiO 2 is presented in Fig. 1a (top). It exhibits pronounced emission from trions, typically attributed to extrinsic effects such as doping from the substrate, mediated through trap states 6,9,10,19 and intrinsic doping resulting from chalcogen vacancies and adsorbates that are reported to occur in mechanically exfoliated flakes 20,21 . We obtain a qualitative measure of the doping by analysing the areas of the neutral and charged exciton A X and A T and their relative spectral weight R = A T /(A T + A X ). Figure 1b shows the peak areas, while the corresponding relative spectral weights are presented in Fig. 1c. The emission intensity for MoSe 2 on SiO 2 is higher for trions than for neutral excitons which is reflected by values of R > 0.5 in Fig. 1c. This remains unchanged when MoSe 2 is stacked on top of ∼14 nm thick hBN, and also after annealing that only results in a slightly decreased total peak area. However, fully encapsulated MoSe 2 exhibits a higher X peak-area compared to T with R < 0.5. This behaviour is indicative of an effective lowering of the Fermi level in the crystal inhibiting trion formation. We attribute this to protection against surface charge contributions from the SiO 2 layer on the substrate material 6 , as hBN naturally shows low defect densities over large areas 22 . This effect is strongest in the fully encapsulated configuration. Since the MoSe 2 is exposed to ambient conditions during and after fabrication in the previous configurations, the TMDC surfaces are free to physisorption of ambient molecules 23,24 , most likely H 2 O due to its polarity. Thus, we assume that the impact of the hBN substrate is reduced due to frozen adsorbates possibly at defects such as selenium vacancies on the TMDC surface.
Changes in the dielectric environment and doping can further influence the exciton peak positions and the linewidths (Fig. 1d) 25 . Here, we directly correlate peak positions and linewidths. The exfoliated MoSe 2 on SiO 2 shows exciton peak positions of P X = (1663.1 ± 1.2) meV and P T = (1631.8 ± 1.3) meV with a binding energy of E T ∼ 31 meV which is typically observed in literature 9,26 . Stacking MoSe 2 on hBN results only in a slight redshift by ΔE ∼ −2 meV and a slightly broader distribution as can be seen in Fig. 1b. This redshift is consistent with recent calculations 25 and measurements 27 considering the change in the refractive index of the substrate from = . n 1 457

SiO 2
to n hBN = 2.2 at the neutral exciton resonance 28,29 . Annealing results in an additional redshift, and a total shift to a lower energy by ΔE ∼ −12 meV compared to pristine MoSe 2 . This is accompanied by a much broader distribution of peak positions. The sandwiched and annealed MoSe 2 structure exhibits the strongest redshift of ΔE ∼ −12 meV. Yet, the statistical spread of the peak position distribution is significantly reduced from s X = (6.8 ± 0.1) meV to s X = (2.8 ± 0.1) meV (see Supplementary Fig. S2), as depicted in the bottom panel in Fig. 1d. Moreover, the trion binding energy decreases from E X − E T = (31 ± 3) meV to (28 ± 3) meV after the annealing step possibly resulting from the modification of the dielectric environment and a change in extrinsic doping 25 . In general, we observe that the symmetric dielectric hBN environment of the MoSe 2 flake combined with annealing results in the sharpest distribution of emission energies, indicative of the highest homogeneity within the MoSe 2 flake. The statistical analysis of the linewidths for MoSe 2 on SiO 2 reveals average values of w X = (5.0 ± 0.5) meV and w T = (7.0 ± 0.8) meV for X and T excitons, respectively. Stacking MoSe 2 on hBN results in significantly higher linewidths of w X = (6.3 ± 1.0) meV and w T = (8.4 ± 1.3) meV with a much broader variation in obtained values. Annealing reduces the linewidth to w X = (5.7 ± 1.5) meV and w T = (4.8 ± 1.5) meV while capping with hBN further reduces the X linewidth to w X = (4.7 ± 0.9) meV, keeping the T linewidth at w T = (4.9 ± 1.3) meV. Interestingly, annealing reveals much higher variance of values which is significantly narrowed upon capping. However, for investigating the linewidths not only the average values are important but also the lowest values obtained. Importantly, for MoSe 2 encapsulated in hBN we observe values as low as w X ∼ 3.5 meV, almost reaching the homogeneous linewidths recently reported in time-resolved four-wave-mixing experiments 12,13 and theoretical calculations 14 of w X ∼ 2.1 meV, w X ∼ 3.4 meV and w X ~ 5.5 meV, for lattice temperatures of T = 6.5 K and 10 K, respectively.
With the dependence γ rad ∝ 1/n Substrate for the radiative linewidth broadening 14,30 , changing the substrate material reduces γ rad by a factor of n SiO 2 /n hBN ≈ 0.66. This then produces a radiative rate which would be quantitatively consistent with the narrowest linewidths measured in our study. Beside radiative broadening, primarily exciton-phonon coupling has been identified as broadening mechanism 14 . Moreover, we attribute the observed remaining broadening of the linewidth to spatial inhomogeneities of the TMDC as a result of the exfoliation procedure and residual polymer bubbles between the interfaces of the monolayer crystal and the surrounding hBN.
Photoluminescence of encapsulated WSe 2 . We repeated the fabrication scheme and optical experiments discussed above for MoSe 2 with WSe 2 . Since we found the most significant improvement in optical quality for TMDCs that are fully encapsulated in hBN, we compare only the two cases of WSe 2 on SiO 2 and WSe 2 encapsulated in hBN after annealing. Typical spectra for WSe 2 on hBN are presented in Fig. 2a. The relative spectral weights R are shown in Fig. 2b. Comparing the relative peak areas of the neutral and charged excitons, results in a trend opposite to that for MoSe 2 . For WSe 2 , the relative intensity of the charged exciton increases by a factor of two upon encapsulation with hBN. We explain this trend by the difference in intrinsic doping of TMDCs present in our experiments. The MoSe 2 crystal employed in this work is n-doped, consistent with measurements on electrically contacted monolayer devices (see Supplementary Section S1), which results in a negatively charged exciton. In contrast, the WSe 2 is p-doped, resulting in emission from positively charged excitons (see Supplementary  Fig. S3). Thus, the hBN encapsulation enables a higher positively charged exciton formation rate.
Upon encapsulation, we observe a reduction of the neutral exciton emission linewidth from (10.3 ± 0.7) meV to (9.8 ± 1.4) meV, whilst the trion emission linewidth reduces from (10.1 ± 2.1) meV to (8.4 ± 1.9) meV. However, this effect is accompanied by a higher overall spread in the linewidth distribution for the capped material. A similar trend is observed for the distribution of peak positions. However, here only a slight redshift is observed. Notably, we observe linewidths as low as w X ∼ 5 meV for the neutral exciton and w T ∼ 5.5 meV. Recent four-wave-mixing measurements 11,13 and theoretical work 14 report and predict homogeneous linewidths of w X ∼ 6.1 meV, w X ∼ 4.7 meV and w X ∼ 6.5meV, respectively.
Optimised stacking processes, reducing bubble formation and wrinkling of the 2D materials could lead to desired purely lifetime broadened emission of the TMDCs.
Photoluminescence of encapsulated MoS 2 . In addition to the Se based TMDCs, we also applied our encapsulation scheme to MoS 2 which in past experiments showed comparatively broad emission from the A-exciton 1,2,6 . This is attributed to inhomogeneous broadening of the emission from neutral and charged excitons that is typically so large that the two peaks are not resolved. Typical PL from MoS 2 exfoliated on SiO 2 is presented in Fig. 3a. For very low excitation power densities of 0.33 kWcm −2 , the spectrum (black curve) reveals emission from the neutral exciton X at (1947.4 ± 0.3) meV, charged excitons T at (1910.7 ± 0.3) meV and pronounced  for a low (black) moderate (red) and high (blue) excitation power featuring the A-peak (blue spectrum) neutral and charged exciton emission and the L-peak at lower energies. (b) Typical μ -PL spectrum of hBN encapsulated MoS 2 for a low (black) and high (red) excitation power reveals sharp neutral and charged exciton emission and no emission from the L-peak. emission from the low energy L-peak is observed located ∼100 meV below X. This broad emission is attributed to defect-related exciton emission 1,2,31 . Upon increasing the excitation power density to 5.27 kWcm −2 (red curve in Fig. 3a) the emission from the neutral exciton vanishes while charged exciton emission dominates. Meanwhile the emission from the L-peak saturates, and its contribution reduces compared to the charged exciton emission. When further increasing the excitation power density to values as high as 83 kWcm −2 (blue curve in Fig. 3a), the emission merges to the broad A-exciton peak normally observed in luminescence studies of MoS 2 with a linewidth of w A ∼ (53.6 ± 0.8) meV 2 . Note that these photo-induced changes in the form of the PL spectrum in our studies were found to be irreversible, consistent with recent findings 18 . For the lowest excitation power densities investigated, the neutral and charged excitons exhibit linewidths of w X ∼ (14.7 ± 0.7) meV and w T ∼ (23.4 ± 0.8) meV. Here, a full statistical analysis was not possible due to the photo-induced changes in the optical spectra. In strong contrast, encapsulation of MoS 2 and annealing significantly enhances the optical emission properties. The PL (Fig. 3b) exhibits bright emission from free excitons. The neutral exciton at (1955.8 ± 0.5) meV and the trion emission at (1926.2 ± 0.5) meV is now blue shifted by (8.4 ± 1.0) meV and (15.5 ± 1.0) meV compared to the MoS 2 on SiO 2 configuration, respectively. By comparing the bare monolayer on SiO 2 at ∼5 kWcm −2 to the encapsulated MoS 2 at ∼3 kWcm −2 (red curves in Fig. 3a and b), we observe that the relative spectral weight strongly shifts from ∼ 0.94 towards lower values of ∼ 0.75. This behaviour of the relative spectral weight of the charged trion emission indicates an effectively lowered Fermi level in the MoS 2 . The overall blueshift is accompanied by a strong decrease in X and T linewidths down to w X ∼ (4.8 ± 1.0) meV and w T ∼ (6.8 ± 0.9) meV, consistent with recent work by Dey et al. 13 reporting a homogeneous linewidth of w X ∼ 6.6 meV in time-resolved four-wave-mixing measurements.
Importantly, we observe no emission from the L-peak indicative of defects and adsorbates 1,2,31 for the fully encapsulated sample. Such features are observed for all other sample configurations, further highlighting the importance of surface protection. Furthermore, both exciton species are well resolved and we observe no photo-induced changes even for the highest excitation power (83 kWcm −2 ) used in our experiments.

Conclusion
In summary, we have investigated the impact of hBN encapsulation on the optical properties of several TMDCs through statistically analysing low temperature photoluminescence experiments. Encapsulation distinctly reduces exciton linewidths and further protects the TMDCs against unwanted doping contributions from substrates or ambient molecules. Moreover, surface protection especially enhances the optical quality of MoS 2 , resulting in very clean spectra and revealing sharp emission from neutral and charged exciton without the presence of any irreversible photo-induced changes. Our findings suggest that encapsulation of TMDCs is essential for accessing the interesting photophysical properties of MoS 2 and enables more sophisticated future optoelectronic devices.
During the writing of this manuscript we recognised related work reported by Cadiz et al. 32 and Ajayi et al. 33 .

Methods
Sample preparation. The monolayer TMDCs studied in this letter are mechanically exfoliated onto degenerately n-doped Si substrates covered with a 285nm thick layer of wet-thermally grown SiO 2 . The heterostructures are stacked using the dry viscoelastic transfer method 34 , whereby we iteratively stacked hBN/TMDC/hBN onto the Si/SiO 2 substrate. The hBN layer thicknesses range from 10nm to 70nm (AFM measurements). After stacking, the heterostructures were annealed at 150 K for 20 min to remove water and polymer accumulated into bubbles and improve the sample homogeneity (see Supplementary Fig. S1).
μ-PL measurements. All photoluminescence (PL) experiments were performed using a confocal microscope at 10 K. The continuous-wave excitation energy was kept at 2.33 eV (Nd:YAG) and an excitation power density of 0.66 kWcm −2 , unless otherwise noted. The spatial mode field diameter of the focal spot (1/e 2 contour) was ∼1.1 μm. The detected light was filtered with a steep fluorescence filter with a transmission cut-on energy 11.7 meV below the laser excitation energy. The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.