Spectral and spatial isolation of single tungsten diselenide quantum emitters using hexagonal boron nitride wrinkles

Solid-state quantum emitters (QEs) such as semiconductor quantum dots and color centers in solids have promising applications in quantum information science, optoelectronics, and nano-sensing because of their versatility, stability, and sensitivity.^1–4 Monolayer transition metal dichalcogenides (TMDs) are direct bandgap semiconductors with strong light–matter interactions, large excitonic effects, and a valley degree of freedom that locks excitons to a given photon helicity.⁵ Quantum emitters in TMDs that inherit valley properties could be used for the generation of single photons with orthogonal polarization for polarization-encoded flying qubits.⁶ One TMD, tungsten diselenide (WSe₂), hosts bright and stable QEs,^7–10 which have been used in a broad range of experiments that show their potential: fine-structure splitting manipulation with the electric field,¹¹ linear-to-circular helicity conversion with the magnetic field,⁸ quantum confined Stark tuning of excitons,¹² and single charge injection to the trion state.^13,14

While most reported emitters in WSe₂ monolayers are located at random, a few reports demonstrate quasi-deterministic activation of QEs via local strain by placing the WSe₂ monolayer over a substrate patterned with nanopillars,^15,16 a slotted waveguide,¹⁷ or hexagonal boron nitride (hBN) nano-bubbles.¹⁸ Based on these works, it has become clear that strain plays a role in activating WSe₂ QEs, although their precise nature—whether excitonic or defect-bound^15,19—is still being actively investigated. There is still no reliable method to strain-engineer single QEs in WSe₂ deterministically. In most reports, the strain in WSe₂ monolayers that is created at random or at a pillar apex is complex and non-uniform, resulting in the creation of multiple QEs per site. Eventually, this results in the excitation and emission from several QEs at once leading to degraded single-photon purity and the need for tunable spectral filtering of specific emitters.

Here, we report on an alternative approach to create spatially isolated single QEs in WSe₂ monolayers with the help of hexagonal boron nitride (hBN). Unlike previous WSe₂ QE studies involving hBN as an insulating layer in heterostructures including graphene,^11–13 we use hBN as a strain buffer from a nanostructured substrate to control the amount of strain applied to WSe₂. By stacking WSe₂ with sub-10-nm hBN, wrinkles nucleate from a nanostructured substrate and create local strain in WSe₂ at a spatial location away from the substrate pattern. We find that these wrinkle points frequently host a single QE with very low background. These two properties—the number of QEs created per site and the peak-to-background ratio at the emitter frequency—are important figures-of-merit to quantify since spectral overlap from other emitters and background light on top of a QE of interest are intrinsically detrimental to the single-photon purity of the source. We extract a peak-to-background ratio as high as 0.99 ± 0.05 for wrinkle-based QEs with an average of 0.88 ± 0.03 for a 3 nm spectral width. By comparison, pillar-based QEs have an average peak-to-background ratio of 0.74 ± 0.03 due to the typical formation of a much richer spectrum composed of several emitter lines and a larger background. We also find that using a thin hBN capping layer above WSe₂ reduces spectral wandering and blinking. Finally, we discuss approaches to controllably introduce local wrinkles in hBN in desired locations, thus enabling deterministic positioning of single WSe₂ QEs. In this vein, we present a sample patterned in different designs to exemplify that many strain profiles, besides the use of nanopillars, are suitable to activate QEs.

Strain engineering of WSe₂ monolayers is a proven method for creating QEs on demand and with position control.^15,16,20 Previous methods have employed WSe₂ monolayers strained directly on top of a nano-patterned substrate to create QEs. Our approach uses a nano-patterned substrate as an indirect means to communicate strain to WSe₂ via a thin hBN layer. By conforming to the uneven substrate, the thin hBN layer supports wrinkles and small height variations, which strains WSe₂ differently than if WSe₂ were to conform to the lithographically defined structures by itself. In the first part of this work, we use a substrate patterned with nanopillars following the fabrication recipe of Proscia et al.²¹ and make a comparative study of emitters formed directly on the pillars vs along hBN wrinkles that propagate between pillars.

The nanopillars are fabricated from a SiO₂(500 nm)/p-Si(500 µm) substrate via electron-beam lithography (JEOL 6300) using the M-aN 2403 negative resist.²¹ The nanopillars (and other structures) are etched into SiO₂ and are cylindrical with a diameter of 200 nm and a height of 300 nm. The pillar aspect ratio greater than 1 is chosen following the findings of Palacios-Berraquero et al.,¹⁶ who found it beneficial to decrease the number of emitters created per pillar. The spacing between pillars is 4 µm, which we found is phenomenologically the minimum spacing that allows the WSe₂/hBN stack to both conform to the pillars and adhere to the substrate via van der Waals forces. Both the hBN and WSe₂ materials used in this work are commercially available from HQ Graphene. To place an exfoliated WSe₂ monolayer deterministically over the fabricated nanopillars, we employ a polycarbonate (PC) on polydimethylsiloxane (PDMS) stamp technique,²² which is commonly used to make heterostructures of 2D materials. The PC is used to pick up WSe₂ and hBN (or vice versa) successively and melted at 180 °C on top of the target substrate. The melted PC is finally dissolved in chloroform for 10 min. The hBN/WSe₂ stack is transferred such that the WSe₂ monolayer is in contact with a Au electrode. A small bias of 1 V is applied across the SiO₂ layer between the WSe₂ monolayer and the back silicon layer, which is used to passivate the electrostatic environment and stabilize QEs. In this section, we study two samples made from the same substrate: sample 1 contains a hBN(bottom)/WSe₂ stack and sample 2 contains a WSe₂/hBN(top) stack.

We study the samples using a homebuilt confocal microscope setup with a 637 nm continuous wave laser focused with 50× Olympus (NA = 0.7) to the diffraction limit (<1 μm). The sample is placed in a helium flow cryostat at 10 K. The laser light is filtered out in collection with a long-pass filter and sent onto silicon avalanche photodiodes (APDs) for time-resolved measurements or a spectrometer (Princeton Acton SP-2500) with focal length f = 500 mm and a 300 g/mm grating. Single QE lines are filtered through a Semrock tunable filter with a 3 nm transmission bandwidth.

Previous studies using nanopillars to activate QEs show that WSe₂ strained by a nanopillar apex become the brightest centers of photoluminescence (PL) on the monolayer.^15,16,23 The spectra associated with these bright spots are composed of several sharp lines and a broad background of weakly bound excitons (720 nm–780 nm). In this work, we are able to reproduce similar results when we study the bright emission centers directly on top of nanopillars. Besides this, we identify a single QE with low surrounding background emission along a hBN wrinkle. We will show that QEs formed at wrinkles reproducibly show low background emission, which is a primary finding of this work.

Figure 1(a) plots an atomic force microscopy (AFM) image of sample 1 consisting of a hBN/WSe₂ heterostructure on a SiO₂ nanopillar substrate. The photoluminescence (PL) map of this sample area obtained after exciting at 637 nm with 500 nW laser power is plotted in Fig. 1(b). The brightest emission is recorded on top of pillars (center and top of the image) and at the folded monolayer edge (left). A fluorescent spot is also visible along a hBN wrinkle, as indicated by a white dashed circle in Figs. 1(a) and 1(b). The spectrum from the center of the nanopillar is shown in Fig. 1(c) and features a broad emission peak from 755 nm to 775 nm, on top of which two peaks may be identified. We attribute the broad background emission to multiple emitters and weakly localized excitons that are simultaneously excited within the collection area of the confocal setup. Figure 1(d) shows a spectrum collected from the wrinkle shooting off the nanopillar (see the white circle). This spectrum features a comparably single sharp peak at 765.8 nm with minimal background emission surrounding the peak. The AFM image reveals a kink in the wrinkle, which can explain the creation of a QE at this specific site. At this location, the wrinkle is 90 nm high and has a full-width at half-maximum (FWHM) of 160 nm. The collected light from this spot is filtered through a 3 nm bandpass filter and sent to a Hanbury Brown–Twiss interferometer for auto-correlation measurements. The time correlation between the two APD signals for the wrinkle QE is plotted as g⁽²⁾(τ) in Fig. 1(f). We normalize the data based on the average count rate at a longer time delay, consistent with the lack of bunching in previous auto-correlation measurements of WSe₂ QEs.^{8,9,15,16,24–26} The data are fitted with g⁽²⁾(τ) = 1 − A₁ exp(−|τ|/t₁), where t₁ and A₁ are a characteristic time and amplitude of photon antibunching, respectively. From the fit, we extract g⁽²⁾(0) = 0.08 ± 0.04, unequivocally demonstrating the single-photon nature of the emission. Figure 1(e) shows g⁽²⁾(τ) of a filtered QE collected from a pillar (spectrum shown as the inset), from which we extract g⁽²⁾(0) = 0.39 ± 0.08. The detrimental effect of spectral proximity to other emitters and background emission collected with the QE clearly impacts the single-photon purity. In the following paragraph, we quantify the amount of unwanted background emission associated with several QEs. This sample provides a first insight into the difference in spectral quality between pillar and wrinkle QEs, an effect that we study extensively in the next sample where we observe reproducible behavior.

We now study sample 2 where the hBN layer is on top of WSe₂ instead of the bottom. As in the previous case, single QEs are found on the pillars as well as between pillars where hBN forms wrinkles or nano-bubbles.^18,27 Figure 2(a) shows an AFM image of the sample where six nanopillars are investigated. All on-pillar and off-pillar emission centers are labeled in Fig. 2(b). Again, a PL map—recorded from the area inside the red square box in Fig. 2(a)—shows that pillars emit the most light while there are other emission centers away from the pillars, see Fig. 2(c). We compare emission spectra of six on-pillar emitters and seven off-pillar emitters in Figs. 2(d) and 2(e). The spectra collected from the top of the pillars show one or several lines attributed to QEs and a broad emission background. We quantify the peak-to-background ratio r₀ = I_peak/(I_peak + I_background) for a 3 nm spectral window, corresponding to our experimental conditions. To extract I_peak and I_background, each spectrum is fitted with the function $a0/((λ−λ0)2+γ2)+b0(λ)$⁠, where a₀ is proportional to the peak intensity, λ₀ is the peak center wavelength, 2γ is the peak full-width at half-maximum (FWHM), and b₀(λ) is the wavelength-dependent background level. I_peak is computed as the integrated intensity of the peak and I_background is the integrated b₀(λ), both of which are limited to a 3 nm spectral window around the emitter center wavelength. For a graphical representation of how the ratio is extracted from a spectrum, refer to Fig. 6 in  Appendix A. Quantum emitters stemming from pillars have an average r₀ of 0.74 ± 0.03. That is, a fourth of the filtered emission comes from sources other than the single-photon source of interest. This background emission cannot be experimentally separated from the single photons, and such sources will intrinsically have a low single-photon purity and limited applicability.^2,28 In Fig. 2(e), we sample spectra that are collected along wrinkles that formed away from the pillars. The spectra consistently display a single peak, i.e., a single QE probed at a time, while the background level is significantly reduced compared to that in the spectra of Fig. 2(b). The average r₀ for wrinkle-based QEs is 0.88 ± 0.03, meaning one photon in eight comes from unwanted sources in the filtered signal, which is a 2× improvement. Four emitters out of seven (labeled A, E, F, and G) have r₀ > 0.9 with the highest value being 0.99 ± 0.05, all allowing for high single-photon purity. We speculate that the variation in r₀ is due to the different wrinkle sizes and morphology, which influence the confining potential and the presence of weakly bound excitons that contribute to background fluorescence. We also note that there are multiple instances in the literature that report spectrally isolated single quantum emitters, formed at random over a thick exfoliated hBN substrate²⁴ or from high aspect ratio nanopillars with sub-100-nm diameter.¹⁶ 

The peak-to-background ratio estimation can be extended to the full spectrum as a way to probe the overall amount of background and other emitters collected in the signal along with the QE of interest. In practice, we compare a QE peak intensity to the sum of all pixels in the collected spectrum. On average, the main QE peak makes 60% of the signal along wrinkles and 26% on pillars. That is, for equal peak intensity, single photons from wrinkle-based QEs are collected with roughly four times as little unwanted emission in the unfiltered spectrum compared to those from pillar-based QEs. In the best case, we find that one wrinkle QE has a full-spectrum peak-to-background ratio of 0.96 ± 0.03. Further improvement of this ratio could allow for direct use of the single-photon source without spectral filtering, which is currently a default setting for semiconductor sources.²⁹ A known way to clean a QE’s spectrum is to use quasi-resonant excitation, whereby the QE is excited through a higher energy state and other emitters do not respond to the same excitation energy.²⁴ This requires a tunable laser close to the QE ground state energy and most often a cross-polarization excitation scheme.

Another aspect that influences the applicability of single-photon sources is the emitter’s spectral and temporal stability. Spectral diffusion and blinking are commonly observed in all solid-state emitters because of fluctuations in the electrostatic environment. Such spectral instabilities are detrimental because they reduce the two-photon indistinguishability and overall excitation/collection efficiency with resonant excitation or spectral filtering.²⁹ Spectral wandering is commonly reported for WSe₂ QEs on SiO₂ substrates; however, blinking is seldom observed.^7,10 Hexagonal boron nitride has been previously used with WSe₂ as a buffer layer to isolate from the substrate;²⁴ however, the effect of the substrate on QE stability has not been fully understood. Tonndorf et al.²⁵ studied the effect of using hBN vs SiO₂ substrates and found no clear influence on QE stability although hBN was found to increase non-radiative recombination channels. Substrate engineering with InGaP was also shown to yield WSe₂ QEs with no detectable spectral diffusion.³⁰ For strain-activated QEs, pillars with a 2:1 aspect ratio (200 nm tall, 100 nm width) create less emitters per pillar with low spectral wandering and blinking, whereas 1:1 aspect ratio pillars host more QEs, which tend to display substantial blinking and spectral diffusion over time.¹⁶ 

In the process of working with hBN/WSe₂ stacks on SiO₂ substrates, we have also examined the effect of using a passivation layer (hBN) on the spectral stability of QEs. Figure 3 shows a spectral time trace for three different configurations of WSe₂ QE environments, in particular with respect to the presence of hBN. Figures 3(a) and 3(b) show two QE spectra from sample 2 taken from two nanopillars in an area of the sample where hBN does not cover the WSe₂ monolayer. Figures 3(c) and 3(d) show two QE spectra from sample 1 where WSe₂ is not capped, but hBN separates the QEs from the SiO₂ substrate. Data from Figs. 3(c) and 3(d) come from a pillar QE and wrinkle QE and correspond to the same QEs shown in Figs. 1(e), 1(d), and 1(f), respectively. Finally, Figs. 3(e) and 3(f) plot data for two wrinkle QEs from sample 2 (spots F and G) that are capped by a thin hBN layer. Overall, we observe that emitters not capped by hBN show significant spectral wandering with a standard deviation around the center wavelength of 0.75 nm and blinking in four emitters out of five. We find that using hBN as a capping layer creates the most spectrally stable emitters. On sample 2, we record spectral wandering with a standard deviation around the center wavelength of 0.11 nm and 0.20 nm, for wrinkle-based and pillar-based emitters, respectively, that is, about five times lower than that for uncapped QEs. Additionally, there is no sign of blinking for all types of capped emitters in sample 2. Details of spectral wandering statistics for all emitters can be found in Table I in  Appendix B. We speculate that the fivefold reduction in spectral wandering and cancellation of blinking when emitters are capped (sample 2) come from the reduction of surface adsorbates and other surface states, which renders the electrostatic environment more stable. The study of spectral wandering should be refined in a further study but is currently limited to the larger linewidth of the emitters at 10 K (FWHM >1 meV).

Multiple experiments suggest that single-photon emitters in WSe₂ are quantum dot-like, i.e., originating from the quantum confinement of single excitons.^15,16,18,27 Tensile strain locally shrinks the WSe₂ bandgap energy, which may lead to a potential well that hosts discrete energy levels similarly to semiconductor quantum dots.³¹ Some theoretical works speculate that the tensile strain gradient serves to funnel excitons toward a defect, the presence of which is necessary for single-photon emission.^19,32,33 Despite the difference in the underlying capturing process of the exciton, controlling the size scale over which strain is applied to WSe₂ is equally crucial in both cases. When using nanopillars with diameter >100 nm to purposefully strain WSe₂, it is likely that confinement over the full size scale of the nanopillar is likely not happening. Based on the nanopillar data of Figs. 1(c) and 2(e) and other works,^15,16 we hypothesize that multiple confining sites are created around the nanopillars due to a complex and uncontrolled strain profile much larger than the size scale suitable to create quantum confinement.²⁹ As a result, nanopillars larger than few tens of nanometers are likely to host several QEs within the excitation/emission spot.

Figures 1–2 show that the use of a thin hBN flake over a nano-patterned substrate to regulate the amount of strain in WSe₂ may be a better choice to create single WSe₂ QEs than using substrate features directly to strain WSe₂. However, the positioning of QEs along hBN wrinkles in these samples is random, whether at a kink or in nano-bubbles. The remainder of this paper focuses on ways to create hBN wrinkles so as to control the positioning of QEs. A single wrinkle only provides one in-plane dimension of confinement, similar to semiconductor nanowires. We have considered several ways to extend the confinement provided by wrinkles to two in-plane dimensions. As a first approach, we designed the structures shown in Fig. 4, which are meant to have two propagating wrinkles cross and create a high strain point. Figures 4(a)–4(c) show an optical micrograph and an AFM image of sample 3a containing star-triangle structures overlaid by a thin hBN flake. In most cases, wrinkles connect nearest-neighbor ridge structures instead of propagating independently and in the direction of the ridge, i.e., creating a crossing of two wrinkles is not straightforward. Moreover, since the transfer process is manually operated and uses a visco-elastic PDMS stamp, the details of the transfer matter: propagation direction, transfer speed, and amount of pressure applied, all influence the outcome. In Figs. 4(d)–4(f), we show sample 3b containing propagating ridges with varying separation angles. At shallow angles (30°), the ridges do form wrinkles that propagate in the direction of the ridge; however, rather than crossing, the two wrinkles avoid one another, ending up propagating in a roughly parallel path. At larger angles, the wrinkles bind directly between nearest-neighbor ridge ends as seen in Fig. 4(c). Figures 4(b) and 4(e) sum up our expected and observed wrinkle propagation directions as black and red dashed lines, respectively. We conclude that a simple approach such as trying to cross two wrinkles (Fig. 4) does not work straightforwardly. Additional work with either a more complex substrate or a refinement of the transfer process will be required to create wrinkle-based quantum emitters deterministically.

Several groups have demonstrated that strain activation of QEs extends beyond the use of nanopillars, by using, e.g., slotted waveguides^17,26 or hBN nano-bubbles.¹⁸ We also exemplify the universality of strain induction of QEs in WSe₂ monolayers by using a star-triangle patterned substrate (sample 3a) with a WSe₂/hBN heterostructure. The heterostructure is transferred on top of 12 star-triangle shapes: we record bright distinguishable emission centers on every structure with an average of 2.9 bright centers per star-triangle feature and an average of 3.1 emitters per bright center (out of 21 studied). Figure 5 shows results from two star-triangles. The AFM images in Figs. 5(a) and 5(d) show how the heterostructure conforms to the star-triangle shape. Interestingly, we notice a wrinkle pattern on top of each star that resembles a triangle. In both instances, bright emission centers appear at the tip of one branch and from the center of the structure, see the PL map in Figs. 5(b) and 5(e). The star-triangle shape is indicated by a yellow dashed line and was inferred from cross referencing the PL map, AFM scan, and optical image of the sample. The associated spectra and filtered g⁽²⁾(τ) are shown in Figs. 5(c) and 5(f). The spectra are reminiscent of those collected from pillars in samples 1 and 2, where the background level is rather pronounced. The bright emission centers recorded at the center of the star-triangle may be coming from the wrinkles that formed on top of the stars in the AFM images. However, we speculate that the tightness of the wrinkles along with the complexity on top of the star structures creates a complex strain profile and hence the more crowded spectra. The filtered g⁽²⁾(0) extracted from the fit is 0.14 and 0.16. This sample shows that working with any structure that produces local strain is likely to create QEs in WSe₂ and that the spectral quality of emitters can vary based on the transfer process, hBN thickness, and morphology.

We have reported the observation of WSe₂ quantum emitters formed from strain created along hexagonal boron nitride wrinkles. These emitters have been shown to be frequently isolated from other emitters and background emission. Quantum emitters found along hBN wrinkles typically have a peak-to-background ratio exceeding 90%, which defines the fraction of a QE intensity over all the collected light in a 3 nm spectral window. Remarkably, an analysis of the unfiltered spectrum shows that on average, bright emission centers associated with QEs along hBN wrinkles contain four times as little residual emission (background and other emitters) compared to emitters formed directly on substrates features such as a nanopillar. We also found that using a capping hBN layer to protect WSe₂ QEs from the environment confers the best configuration to reduce spectral wandering and suppress blinking in our study. Finally, we have studied alternative substrates to control hBN wrinkle propagation and concluded that crossing two wrinkles to create a high-strain point is not straightforwardly achievable and requires additional tuning of the substrate features or transfer process. Our combined results—single QE created per site, low surrounding background, and spectral stability—make wrinkle-based WSe₂ QEs promising for applications requiring highly pure single-photon sources.^1–3,29 Further research that enables the use of the valley properties of WSe₂ excitons would bring even more functionality to the source, such as control over photon polarization.⁵ 

R.S.D., T.A.F.V., and G.D.F. designed the experiment and wrote the manuscript. R.S.D. and T.A.F.V. prepared samples and collected and analyzed the data. A.M. and A.N.V. performed additional measurements on the samples. Z.W., K.F.M., and J.S. helped with sample preparation. All authors reviewed the manuscript.

The data that support the findings of this study are available from the corresponding author upon reasonable request.

This work was primarily supported by AFOSR MURI (Grant No. FA9550-18-1-0480) with partial support from the Cornell Center for Materials Research (CCMR) with funding from the NSF MRSEC program (Grant No. DMR-1719875), NSF Career (Grant No. DMR-1553788), and AFOSR (Grant No. FA9550-19-1-0074). We acknowledge facility use from the CCMR and from the Cornell NanoScale Facility, a member of the National Nanotechnology Coordinated Infrastructure (NNCI), which is supported by the National Science Foundation (Grant No. NNCI-1542081).

Figure 6 shows the spectrum of wrinkle QE B from sample 2 [see Fig. 2(e)] along with a fit of the QE line. The fitted function is composed of a Lorentzian lineshape I_peak(λ) plus a wavelength-dependent background I_background(λ) as discussed in the main text. The peak area and background are calculated by summing the pixels of the red and green areas within a spectral window of 3 nm indicated as a dashed box. In this example, the computed r₀ = 0.76 ± 0.02.

Table I collects the center wavelength (λ₀), the standard deviation around λ₀, the maximum recorded deviation around λ₀, and the blinking amount of all measured QEs in samples 1 and 2.