Monolayer WS₂ electro- and photo-luminescence enhancement by TFSI treatment

We use 1L-WS₂ as the active light-emitting layer since it has a direct bandgap [77–80], its PL emission is ∼60 times stronger than 1L-MoS₂ [39, 78] at RT, η_EL can be up to ∼50 times larger than 1L-MoS₂ [19, 20] at RT, while references [55, 57–68] demonstrated that TFSI treatment increases up to ∼10-times its PL intensity.

Figure 1(a) shows the 1L-WS₂/hBN/SLG tunnel junction configuration used here, where the metallic electrodes provide contacts to apply a voltage (V) between SLG and 1L-WS₂. This is prepared as follows.

Zoom In Zoom Out Reset image size

WS₂ crystals are synthesized using a two-step self-flux technique [81] using 99.9999% purity W and S powders without any transporting agents. Commercial (Alfa Aesar) sources of powders contain a number of defects and impurities (Li, O, Na, and other metals as determined by secondary ion mass spectroscopy). Before growth, W and S powders are thus purified using electrolytic [82] and H₂ [82] based techniques to reach 99.995% purity. WS₂ polycrystalline powders are created by annealing a stoichiometric ratio of powders at 900 ^∘C for 3 weeks in a quartz ampoule sealed at 10⁻⁷ Torr. The resulting powders are re-sealed in a different quartz ampoule under similar pressures and further annealed at 870 ^∘C–910 ^∘C with thermodynamic temperature differential (hot to cold zone difference) ∼40 ^∘C. The growth process takes 5 weeks. At the end of the growth, ampoules are cooled to RT slowly (∼40 ^∘C h⁻¹) [83]. We use this material as bulk source because we previously [83] demonstrated that this has a defect density ∼10⁹–10¹⁰ cm⁻², on par or better than previous reports [84].

Bulk WS₂, hBN (grown by the temperature-gradient method [85]), and graphite (sourced from HQ Graphene) crystals are then exfoliated by micromechanical cleavage using Nitto-tape [86] on 285 nm SiO₂/Si. Optical contrast [87] is first used to identify 1L-WS₂, SLG, FLG (3-10 nm), and hBN($\lt$5 nm). The LMs are then characterized by Raman spectroscopy as discussed in Methods. After Raman characterization of all individual LMs on SiO₂/Si, the FLG/1L-WS₂/hBN/SLG LMH is assembled using dry-transfer, as for references [88, 89]. FLG is picked-up from SiO₂/Si using a polycarbonate (PC) membrane on a polydimethylsiloxane (PDMS) stamp (as mechanical support) at 40 ^∘C. We use 40 ^∘C because this is sufficient to increase the adhesion of the PC film [90], to pick all LMs from SiO₂/Si. Then, FLG is aligned to one edge of 1L-WS₂ on SiO₂/Si and brought into contact using xyz micromanipulators at 40 ^∘C, leaving the majority of 1L-WS₂ without FLG cover to be used as active area (AA). AA is the region from where light emission is expected, and it is the overlap area between 1L-WS₂ and SLG (green-shaded part in figure 1(b)). Next, FLG/1L-WS₂ is aligned to a hBN flake deposited onto SiO₂/Si and brought into contact using xyz micromanipulators at 40 ^∘C. Finally, FLG/1L-WS₂/hBN is aligned to a SLG on SiO₂/Si and brought into contact using xyz micromanipulators at 180 ^∘C, whereby PC preferentially adheres to SiO₂ [88], allowing PDMS to be peeled away, leaving PC/FLG/1L-WS₂/hBN/SLG on SiO₂/Si. PC is then dissolved in chloroform for ∼15mins at RT, leaving FLG/1L-WS₂/hBN/SLG LMH on SiO₂/Si [88, 89]. After LMH assembly, Cr/Au electrodes are fabricated by electron beam lithography (EBPG 5200, Raith GMBH), followed by metallization (1:50 nm) and lift-off.

The tunnel junction based on a MIS structure consists of a LMH with 1L-WS₂ as the light emitter, FL-hBN (typically from 2 to 4 nm) as tunnel barrier, and a SLG electrode to inject holes (h) into 1L-WS₂. We use FL-hBN$\lt$5 nm so that a low (typically $\lt$5 V) driving voltage is sufficient for charge injection to the 1L-WS₂ [91, 92]. We employ FLG (∼3–10 nm) to contact 1L-WS₂, because FLG reduces the contact resistance [93], while Cr/Au electrodes give Ohmic contacts to SLG and FLG [93]. SLG could also be used to contact 1L-WS₂, however, as the optical contrast is higher in FLG than SLG [87, 94], using FLG makes it easier to align it to 1L-WS₂ during transfer. Since TFSI treatment requires direct exposure of 1L-TMDs [57], we place 1L-WS₂ on top of the stack to compare device performance before and after treatment. We TFSI-treat 4 samples for EL and gated-PL measurements. These are immersed in a TFSI solution (0.2 mg ml⁻¹) in a closed vial for 10mins at 100 ^∘C [57–59], then removed, dried by a N₂ gun, and annealed on a hot plate at 100 ^∘C for 5mins [57–59]. Figure 1(b) is an image of the 1L-WS₂-LEDs. The FLG electrode is placed on the side of the SLG to avoid direct tunneling of carriers from SLG to FLG, hence keeping as AA the LMH region extended over SLG and 1L-WS₂, green-shaded in figure 1(b). If there is a FLG/SLG overlap, tunneling through FLG-SLG may be possible, not resulting in e-h recombination into 1L-WS₂, hence no EL [6, 25, 38].

Figures 1(c) and (d) sketch the band diagram of our LEDs for V = 0 V and V $\gt$ 0 V, respectively. For V = 0 V (at thermodynamic equilibrium as indicated in figure 1(c)), the Fermi level, E$_\textrm{F}$, is constant across the junction, and the net current (I) is zero [6, 21, 25, 28, 38]. For V $\gt$ 0 V (positive potential on SLG), the SLG E$_\textrm{F}$ is shifted below the 1L-WS₂ valence band (VB) energy E$_\textrm{V}$, figure 1(d), and h from SLG tunnel across the hBN barrier into 1L-WS₂, promoting EL emission by radiative recombination between the injected excess h and intrinsic e [21–24, 28, 35, 38]. The EL emission is expected to increase as a function of tunneling current because of the increasing h injected into 1L-WS₂ available for e-h recombination.

The LMs are characterized by Raman, PL, EL spectroscopy using a Horiba LabRam HR Evolution. The Raman spectra are collected using a 100x objective with numerical aperture (NA) = 0.9, and a 514.5 nm laser with a power ∼5 µW to avoid damage or heating. The voltage bias dependent PL and EL are collected using a long working distance 50x objective (NA = 0.45). For the PL spectra, we use a λ = 532 nm (2.33 eV) continuous-wave laser in order to excite above the X⁰ emission (∼2 eV) [9, 10]. The laser power is kept ∼80 nW to avoid laser-induced thermal effects [2, 9–11]. The spot size A_spot for the PL measurements is calculated as A $_{\textrm{spot}} = \pi[1.22\mathbf{\lambda}$/2NA]$^2\sim1.6\,\mu$m². The voltage (V) and current (I) between source (SLG) and drain (1L-WS₂) electrodes are set (V) and measured (I) by a Keithley 2400 (see section 3 for details of electrical measurements).

Figure 2 plots the Raman spectrum of 1L-WS₂/hBN/SLG on Si/SiO₂ after device fabrication and before current–voltage (I-V) measurements. The Raman modes of each LM can be identified, table 2. For 1L-WS₂, Pos(A$_{1}^{{^{\prime}}}$) and its full width af half maximum, FWHM(A$_{1}^{{^{\prime}}}$), change from ∼418.9$\,\pm\,0.2$ cm⁻¹; 3.9$\,\pm\,0.2$ cm⁻¹, before assembly, to ∼419.8$\,\pm\,0.2$ cm⁻¹; 3.4$\,\pm\,0.2$ cm⁻¹, after. All the changes in the other modes are close to our spectral resolution and errors, as for reference [95]. Pos(A$_{1}^{{^{\prime}}}$) and FWHM(A$_{1}^{{^{\prime}}}$) are sensitive to changes in n-doping [96, 97]. The mechanism responsible for this effect is an enhancement of electron-phonon (e-ph) coupling when e populate the valleys at K and Q simultaneously [97]. The energy of the K and Q valleys is modulated by the A$_{1}^{{^{\prime}}}$ ph [97]. Since the K and Q energies are modulated out-of-phase, charge transfer between the two valleys occurs in presence of the A$_{1}^{{^{\prime}}}$ ph [96, 97]. When the K and Q valleys are populated by e, these are transferred back and forward from one valley to the other [97, 98]. This increases the e-ph coupling of out-of-plane modes, such as A$_{1}^{{^{\prime}}}$ [97]. The same process does not occur for p-doping [97]. The reason for this asymmetry between n- and p-doping is due to a much larger energy separation (∼230 meV [97]) between the VB Γ and K valleys than that (∼100 meV [97]) of the conduction band (CB) K and Q valleys. From the changes in Pos(A$_{1}^{{^{\prime}}}$) and FWHM(A$_{1}^{{^{\prime}}}$), and by comparison with reference [97], we estimate a reduction in n-doping ${\sim}5\times10^{12}$ cm⁻².

Zoom In Zoom Out Reset image size

Table 2. Pos and (FWHM) in cm⁻¹ of WS₂ Raman peaks, before and after LMH assembly, and TFSI treatment.

Peak	Bulk-WS2 Assignment	Bulk-WS2	1L-WS2 Assignment	1L-WS2-SiO2	1L-WS2-LMH	TFSI + 1L-WS2-LMH
1	LA(M)	174.5 (11.1)	LA(M)	175.6 (14.5)	175.6 (14.6)	174.9 (14.4)
2	LA(K)	194.8 (3.3)	LA(K)	193.3 (4.5)	193.8 (3.3)	193.3 (4.7)
3	A$_{1g}$(K)-LA(K)	213.7 (4.2)	A$_{1}^{{^{\prime}}}$(K)-LA(K)	214.5 (5.7)	214.5 (5.2)	213.5 (6.0)
4	A$_{1g}$(M)-LA(M)	232.8 (5.7)	A$_{1}^{{^{\prime}}}$(M)-LA(M)	231.5 (6.7)	231.9 (7.1)	231.4 (5.9)
5	A$_{1g}$(M)-ZA(M)	266.8 (6.9)	A$_{1}^{{^{\prime}}}$(M)-ZA(M)	265.3 (6.9)	265.9 (7.2)	265.4 (7.0)
6	E$^2_{2g}$(Γ)	297.6 (4.2)	E$^{^{^{\prime\prime}}}$(Γ)	297.7 (2.8)	298.5 (3.1)	298.7 (2.6)
7	LA(M)+TA(M)	311.2(2.4)	LA(M)+TA(M)	311.2 (2.5)	311.8 (2.3)	311.2 (2.4)
8	E$^2_{2g}$(M)	324.6 (17.5)	E$^{^{^{\prime\prime}}}$(M)	326.7 (25.5)	325.9 (24.7)	327.7 (25.7)
	2LA(M)	350.6 (8.3)	2LA(M)	352.4 (9.3)	352.7 (9.2)	352.7 (8.0)
	E$^{1}_{2g}$(Γ)	356.9 (1.5)	E$^{{^{\prime}}}$(Γ)	357.2 (3.3)	357.4 (3.1)	357.2 (2.9)
	A$_{1g}$(Γ)	420.8 (2.1)	A$_{1}^{{^{\prime}}}$(Γ)	418.9 (3.9)	419.8 (3.4)	419.9 (3.4)

For hBN in figure 2, Pos(E$_{2g}$) ∼1366.4$\,\pm\,0.2$ cm⁻¹ and FWHM(E$_{2g}$) ∼9.2$\,\pm\,0.2$ cm⁻¹. Although FWHM(E$_{2g}$) changes within the error, Pos(E$_{2g}$) downshifts ∼2.1 cm⁻¹ after assembly, suggesting a contribution from strain (see section 3 for comparison between FL- and bulk-hBN Raman). Uniaxial strain lifts the degeneracy of the E$_{2g}$ mode and results in the splitting in two subpeaks E$_{2g}^+$ and E$_{2g}^-$, with shift rates ${\sim}-8.4$ and −25.2 cm⁻¹/$\%$ [99, 100]. For small levels of uniaxial strain ($\lt$0.5$\%$) splitting cannot be observed and the shift rate is ${\sim}-16.8$ cm$^{-1}/\%$ [99, 100]. For biaxial strain, splitting does not occur and E$_{2g}$ shifts with rate ${\sim}-39.1$ cm⁻¹/$\%$ [99]. Since we do not observe splitting, the E$_{2g}$ shift can be attributed to uniaxial or biaxial tensile strain ∼0.13$\%$ or ∼0.06$\%$, respectively.

For SLG in figure 2, no D peak is observed after LMH assembly, indicating negligible defects [101–103]. In figure 2, Pos(G) ∼1585.1$\,\pm\,0.2$ cm⁻¹, FWHM(G) ∼9.0$\,\pm\,0.2$ cm⁻¹, Pos(2D) ∼2692.3$\,\pm\,0.2$ cm⁻¹, FWHM(2D) ∼20.9$\,\pm\,0.2$ cm⁻¹, I(2D)/I(G) ∼2.4, and A(2D)/A(G) ∼5.6. These indicate that the SLG is p-doped, with E$_\textrm{F}\sim$150$\,\pm\,$50 meV [102–104] by taking into account the average dielectric constant (∼3.85) of the environment ($\varepsilon_\mathrm{SiO_2}\sim$3.8 [105] and $\varepsilon_{hBN}\sim$3.9 [106]). E$_\textrm{F}\sim$150 meV should correspond to Pos(G) ∼1584.1 cm⁻¹ for unstrained SLG [107]. However, Pos(G) ∼1585.1$\,\pm\,0.2$ cm⁻¹, which implies a contribution from compressive uniaxial (biaxial) strain ∼0.04$\%$ (∼0.01$\%$). The strain level for SLG and hBN are different, most likely due to the fact that the SLG is directly exfoliated onto SiO₂/Si, while hBN is picked up and transferred by PDMS stamps, hence, this could induce a larger amount of strain on hBN.

Figure 3(a) plots the I-V characteristics. For V = 0 V the current is zero (figure 1(c)). When V is applied, an electrical rectification (i.e. diode behavior) with negligible leakage current (I $\lt$10⁻¹¹A) for V $\lt$0 is seen. A tunneling onset, (i.e. exponential increase of I) is seen at V_ON ∼4.1 V, figure 3(a). V_ON is related to the breakdown electric field (E_bd) across the junction, which depends on the voltage drop on the hBN tunnel barrier and hBN thickness (d) accordingly to E_bd = (V_bd/d) ∼0.7–1 V nm⁻¹ [91, 92], where V_bd is voltage breakdown V_bd = qnd²/($\varepsilon_0\varepsilon_{hBN}$), q is the e charge, n is total charge concentration, ε₀ = 8.854×10⁻¹² F/m and $\varepsilon_{hBN}\sim$3.9 [91, 92], so that V$_{ON}$ can vary between different devices. When V $\gt$ V $_{ON}$, h from SLG tunnel across the hBN barrier into 1L-WS₂, promoting EL emission by radiative recombination between the injected h and majority e in 1L-WS₂, figure 1(c) [21–24, 35, 38]. The EL intensity ∼634 nm (∼1.956 eV) increases with tunneling current, as in figure 3(b). The red-shifts ($\lt$2 nm) in EL emission in figure 3(b) for larger I can be assigned to the E$_\textrm{F}$ shift induced by the MIS structure [6, 31, 33]. FWHM(EL) also increases with I, attributed to the scattering of excitons with extra carriers [6, 108]. This shift and EL broadening is assigned to heating effects at the layered junction [28]. A red-shift ∼48 meV is observed in EL emission ∼634 nm (∼1.956 eV) with respect to the PL X⁰ emission of the unbiased device (dashed black line, figure 3(b)). Figure 3(b) shows a EL peak position close to X⁻ of unbiased PL (dashed black line, figure 3(b)), implying a trionic EL emission, due to excess e in 1L-WS₂ [28, 38]. In contrast, no light emission is observed for V $\lt0V$ and small positive ($0\lt$ V $\lt$ V$_{ON}$ ) bias, below the tunneling condition (V_ON $\lt$4.1 V).

Zoom In Zoom Out Reset image size

To further understand the EL emission origin, we perform EL and PL spectroscopy at the same V. Figure 4(a) plots PL spectra at different V. At V = 0 V, the PL peak is ∼619.2 nm (∼2.002 eV), assigned to X⁰ [9, 78]. By increasing V (i.e. increasing e density in 1L-WS₂), a second peak appears at longer wavelengths (∼630 nm, ∼1.968 eV), due to X⁻ [9–11, 109]. The dashed lines in figure 4(a) describe the peak position evolution of X⁰ and X⁻ emission for different V. For V $\gt$ 0 V, the X⁰ intensity gradually decreases and nearly vanishes, while X⁻ shifts to longer wavelengths, figure 4(a), reaching a similar position for both EL and PL emission (see methods for further details). Figure 4(a) demonstrates that the gated-PL data show multipeak emissions at certain V (i.e. 0 V$\lt$ V $_{ON}$) due to injection of charge carriers from SLG to 1L-WS₂, allowing the recombination of both X⁰ and X⁻, whereas for n-doping (V $\gt\gt$0 V), only X⁻ recombination takes place. This is expected for trionic emission, due to e-doping induced by V [9–12, 38, 109]. Similar effects were observed in 1L-MoS₂/SiO₂/Si [110], hBN/1L-WSe₂/hBN/SiO₂/Si [6], and hBN/1L-WS₂/hBN/SiO₂/Si [28]. Therefore, for similar tunneling current, EL agrees in energy and shape with the PL emission, further verifying the same origin for EL and PL, figure 4(b). This is confirmed by figure 4(c), where EL and PL peak positions are plotted for 4 devices, showing EL and PL emission at similar wavelengths. Thus, EL predominantly originates from X⁻ [6, 9, 10, 21, 38]. The variations in X⁻ for different LEDs are due to changes in charge carriers density across different samples. The charge density variation can be due to the number of vacancies in 1L-WS₂ [41] and external impurities (PC residues and adsorbed water) after LED fabrication, which may vary from sample to sample.

Zoom In Zoom Out Reset image size

We now consider the origin and consequences of excess e in 1L-WS₂ for EL emission induced by V. Besides the intrinsic charge carriers in 1L-WS₂ (typically n-type due to S vacancies [41]), there is also an electrostatically induced charge in 1L-WS₂ when V $\gt$ 0 V. A SLG/hBN/1L-WS₂ tunneling junction acts as a MIS capacitor [6, 28, 38]. When V $\gt$ 0 is applied to SLG, inducing positive charges in SLG, there is an opposite (negative) charge induced in 1L-WS₂ [6, 28, 38], thus making the charge density on 1L-WS₂ larger than for V=0. When V $\gt$ V$_{ON}$ , h will be injected by tunneling into 1L-WS₂ (figure 1(d)), hence, h will recombine with e. Consequently, the EL emission originates from X⁻ states. However, the radiative recombination efficiency (defined as the number of e-h pairs that recombine by emission of a photon divided by the total number of e-h pairs) of X⁻ is lower than X⁰ because of the small (∼30 meV) binding energy of trions [42]. Thus, to gain higher η_EL one should favor X⁰ EL emission by lowering the unbalanced free-carriers concentration in 1L-TMDs by either gate modulation [6, 12, 28, 31, 36, 38], physical [111, 112], or chemical doping [11, 25].

We thus treat 1L-WS₂ using TFSI to reduce doping and favor X⁰ emission under bias and investigate the effects on EL emission and gated-PL. Figure 5 plots representative Raman spectra before (black) and after (red) TFSI treatment. By comparing the spectra before and after TFSI treatment, and the fits for 1L-WS₂ in table 2, we do not observe significant changes in peak position and FHWM. However, there is an overall intensity increase of the Raman peaks of ∼50$\%$, compared to the Si peak. This indicates a reduction of n-doping induced by TFSI treatment, because S vacancies in 1L-TMDs are commonly associated to n-type behaviour and the reduction of these defects will reflect in p-type doping [57–59, 69]. Pos(A$^{{^{\prime}}}_{1}$) is unaffected by TFSI treatment, which suggests that the reduction in the intrinsic 1L-WS₂ n-doping induced by TFSI is$\lt\lt$10¹² cm⁻² [97]. Although TFSI is able to p-dope SLG when it is in contact with the TFSI solution [113], figure 5 shows negligible (within the errors [95]) changes in SLG (e.g. before (after): Pos(G) ∼1585.1 (1585.0)$\,\pm\,0.2$ cm⁻¹, FWHM(G) ∼9.0 (9.1)$\,\pm\,0.2$ cm⁻¹, Pos(2D) ∼2692.3 (2692.2)$\,\pm\,0.2$ cm⁻¹, FWHM(2D) ∼20.9 (20.8)$\,\pm\,0.2$ cm⁻¹, I(2D)/I(G) ∼2.4 (2.4), and A(2D)/A(G) ∼5.6 (5.6)) and hBN (e.g. before (after): Pos(E$_{2g}$) ∼1366.4 (1366.5)$\,\pm\,0.2$ cm⁻¹ and FWHM(E$_{2g}$) ∼9.2 (9.1)$\,\pm\,0.2$ cm⁻¹) Raman spectra after treatment, as both are protected by the top 1L-W₂.

Zoom In Zoom Out Reset image size

Figure 6(a) plots a representative PL spectrum of 1L-WS₂ embedded in the LMH before TFSI treatment, and figure 6(b) after. For the pristine case, there are two components, fitted by two Lorentzians ∼618.7 nm with FHWM ∼8.4 nm and ∼629.1 nm with FHWM ∼18.1 nm, which correspond to X⁰ [78, 80] and X⁻ emission, respectively [109, 114]. The observation of X⁻ emission and FWHM(X⁻) $\gt$ FWHM(X⁰) is related to the n-type doping behavior of 1L-WS₂ [115], which favors X⁻ recombination. For non-biased devices, the spectral weight (defined as the area of each peak) of the PL emission indicates a majority emission due to X⁰. After treatment, the PL emission evolves to a main single peak ∼618.1 nm with FHWM ∼8.7 nm, accompanied by a ∼4-fold increase in PL intensity. The PL intensity enhancement induced by TFSI is consistent with that reported in references [8, 49, 55, 57, 59, 60, 69, 70]. The changes in spectral weight of X⁰ and X⁻ emission and FHWM after treatment can be assigned to a reduction in e-density of 1L-WS₂ [57–59], in agreement with our Raman analysis. References [53–55, 57–68] reported that PL enhancement depends on sample quality (defects) and may vary 1 to 10 times. We observe a PL increase ∼5$\,\pm\,$1-times, as for references [53–55, 57–68].

Zoom In Zoom Out Reset image size

Figure 7(a) plots typical I-V characteristics of 3 devices before (solid black lines) and after (dashed red lines) TFSI treatment. I is not affected by the treatment. V $_{\textit{ON}}$ is related to the breakdown electric field E_bd across the junction, which depends on the voltage drop on the hBN tunnel barrier and hBN thickness [91, 92]. Our LED-devices use 2 to 4 nm hBN as tunnel barrier. Consequently, V $_{\textit{ON}}$ can vary between different devices. Our results on the same device demonstrate that TFSI treatment does not induce changes in the tunneling because the chemical treatment does not modify the MIS structure. Figures 9(a) and (b) show EL collected before and after TFSI, respectively, for different I. In both cases, EL is triggered for similar current levels (I $\lt$ 5 nA), and the intensity increases linearly with I, figure 9(c). The EL intensity slope as a function of current density (I divided by AA) is affected by TFSI. For pristine-LEDs we get an average slope $\alpha\sim$1.4$\,\pm\,$0.3, while after TFSI $\alpha\sim$13.5$\,\pm\,$1.1, with 1 order of magnitude η_EL increase, figure 9(c). The red-shifts in the EL emission with I increase in pristine ($\lt$6 nm) and TFSI treated LEDs ($\lt$5 nm), figures 9(a) and (b), can be assigned to E$_\textrm{F}$ shift induced by the MIS structure [6, 31, 33], or heating effects at the layered junction [28].

Zoom In Zoom Out Reset image size

Next, we estimate the external quantum efficiency (EQE) of our LEDs. This is defined as the ratio between the number of emitted photons (N_ph) and that of injected h per second (N_h ) [116]:

$$\begin{equation} \textrm{EQE} = \frac{N_{\textrm{ph}}}{N_{h}} = \frac{\sum_\mathbf{\lambda}N_\textrm{ph-counts}}{N_{h}}\times \frac{A_\textrm{eff}}{\eta_\textrm{sys}}, \end{equation} \tag{ 1 }$$

where $\sum_\mathbf{\lambda}N_\textrm{ph-counts}$ is the sum of the total photons collected by the spectrometer over the measured spectral range, A$_\textrm{eff} = AA/A_\textrm{spot}$, where A $_\textrm{spot}$ is the microscope objective spot size A $_\textrm{spot}\sim$2.2µm², with λ = 618 nm and NA = 0.45, and N_h = I×t/q, where t is the acquisition time, and q the e charge. The efficiency factor (defined as the ratio between the photons collected by the detector and the emitted photons by EL at the sample position) of our setup, including all optical components and spectrometer, is $\eta_\textrm{sys}\sim$0.0051, see section 3.

From equation (1) we get EQE ∼0.025%$\,\pm\,$0.021% and ∼0.195%$\,\pm\,$0.324% for pristine- and TFSI treated-LEDs, respectively, corresponding to a ∼8.7$\,\pm\,$1.5-fold increase, thus demonstrating that TFSI can boost EQE by almost one order of magnitude. EQE ∼0.2% is better than previously reported for 1L-WS₂-based EL (∼0.1%) on a SiO₂/SiN_x microcavity [35], bulk organic (2,7-bis[9,9-di(4-methylphenyl)-fluoren-2-yl]-9,9-di(4-methylphenyl)fluorene) emissive layer (∼0.1%) [117], and semiconducting (6,5) single-wall nanotubes (∼0.1%) [118]. The WS₂-LED performance can be further improved if placed within a photonic cavity [25, 119–127].

To evaluate the cavity enhanced EQE, we extract the performance metrics from our reference measurements on SiO₂/Si and project them on simulation results for a cavity system. We first estimate the reference internal quantum efficiency IQE$_\textrm{ref} = \textrm{EQE}_\textrm{ref}/\eta_\textrm{(extr,ref)}$ [116], where $\eta_\textrm{(extr,ref)}$ is the extraction efficiency on SiO₂/Si, defined as the number of photons emitted into free space over the total number of emitted photons [116]. Emitted photons not reaching free space include those absorbed and/or lost into the reference substrate. By 3d finite-difference time-domain method (FDTD) simulations [128, 129], of a 10×10 µm² reference system with a single transverse electric (TE) (i.e. parallel to the surface) point dipole emitter on the surface at λ = 624 nm, we calculate $\eta_\textrm{(extr,ref)}\sim$14%, i.e. a significant (∼86%) portion of the emitted light is absorbed and/or lost into the substrate. This yields for the TFSI-treated LED IQE$_\textrm{ref}\sim$1.43%. In turn, IQE$_\textrm{ref}$ is related to the ratio of radiative to total recombination rates, i.e. IQE$_\textrm{ref} = \Gamma_\textrm{R,ref}/(\Gamma_\textrm{R,ref}+\Gamma_\textrm{NR}$) [116, 130], where $\Gamma_\textrm{R,ref}$ and $\Gamma_\textrm{NR}$ denote the radiative and non-radiative rates, respectively. For simplicity we assume $\Gamma_\textrm{NR}\equiv\Gamma_\textrm{NR,ref}$ independent of substrate, i.e. the non-radiative relaxation pathways are unaffected by the environment. Thus $\Gamma_\textrm{NR} = \Gamma_\textrm{R,ref}(\text{IQE}_\textrm{ref}^{-1}-1)$. Inside a cavity, the radiative density of states increases, leading to a proportional increase in radiative rate due to the Purcell effect [130, 131]. Thus, $\textrm{IQE}_\textrm{cav} = \Gamma_\textrm{R,cav}/(\Gamma_\textrm{R,cav}+\Gamma_\textrm{NR})$, where $\Gamma_\textrm{R,cav} = \eta_P\Gamma_\textrm{R,ref}$ and η_P is the relative Purcell enhancement factor calculated by FDTD as the ratio of total light (free and lost) emitted by a point dipole in the cavity over the total light emitted on the reference substrate. Combining the above relations, we get $\textrm{IQE}_\textrm{cav}^{-1} = 1+\eta_P^{-1}(\textrm{IQE}_\mathrm{SiO_2/Si}^{-1}-1)$. The external quantum efficiency becomes $\textrm{EQE}_\textrm{cav} = \eta_\textrm{extr,cav} \textrm{IQE}_\textrm{cav}$, with $\eta_\textrm{extr,cav}$ the extraction efficiency of the cavity evaluated by FDTD. The photonic cavity, shown in the top inset of figure 8, is asymmetric to enable maximal unidirectional emission. Nb₂O₅/SiO₂ bilayers are used for the Bragg mirrors, with N_P Nb₂O₅/SiO₂ periods on top, a SiO₂ cavity in the middle, and an Ag back mirror. At λ = 624 nm the refractive indices are $\textit{n}_\mathrm{Nb_2O_5} = 2.325$ [132],$\textit{n}_\mathrm{SiO_2} = 1.457$ [133], $\textit{n}_\mathrm{Ag} = 0.0581+\textit{i}4.212$ [134], $\textit{n}_\textrm{SLG} = 2.787+\textit{i}1.443$ [135, 136], $\textit{n}_{hBN} = 2.12$ [137], $\textit{n}_{WS_2} = 5.38+\textit{i}0.382$ [138]. The layer thicknesses are d $_\mathrm{Nb_2 O_5}$ = 67 nm, d $_\mathrm{SiO_2}$ = 107 nm and d $_\textrm{cav}$ = 191 nm, while 1L-WS₂/hBN(3 nm)/SLG is placed in the middle of the SiO₂ cavity layer. FDTD simulations predict the cavity enhanced EQE as a function N_P as for figure 8. For N_P = 1 the EQE gets a ×5.75 enhancement, reaching ∼1.15%, attributed to a Purcell enhancement of ×1.15 and a ×5 increase in extraction efficiency. Cavity effects also enhance the directionality of light, as shown in the bottom inset of figure 8 for the SiO₂/Si reference as well as for the N_P = 1 and N_P = 2 cavity structures. To quantify the enhancement in the vertical direction (e.g. were we to use a narrow NA = 0.0125 [127]), we also plot the normal emission enhancement through a 1d calculation in figure 8. These are consistent with the angular plots of full 3d systems (lower inset of figure 8) and predict an optimal cavity with N_P = 3 and relative normal emissivity enhancement of ×16. Further enhancement mechanisms can also be considered, such as plasmonic cavities and/or antenna-coupling [139–145], as well as dielectric domes on top of the device to further boost extraction.

Zoom In Zoom Out Reset image size

We now consider the EL emission features induced by TFSI treatment. By comparing EL before and after TFSI (figures 9(a) and (b)), a blue-shift in EL is observed. In pristine-LEDs, the main EL emission can be fitted with a single Lorenztian at ∼641.8 nm with FHWM ∼28.6 nm, figure 10(a), whereas after treatment it can be fitted with two-Lorentzian ∼625.9 nm with FHWM ∼28.6 nm and ∼640.1 nm with FHWM ∼29.2 nm, figure 10(b). The observation of X⁰ emission and similar FWHM(X⁻) and Pos(X⁻) after TFSI treatment suggests that there is a reduction in n-doping, consistent with PL and Raman analysis. The fact that we observe X⁻ emission in EL after TFSI treatment indicates that 1L-WS₂ is still n-doped, which can be caused by e injection from the bias in the MIS structure [6, 28, 33, 108]. Figure 10(c) plots the EL peak position before and after treatment in 4 devices. After treatment, the EL emission shifts to shorter wavelengths, where X⁰ is expected [77, 78] (dashed line in figure 10(c)). In non-biased S-based TMDs devices, this shift could be due to the depletion of excess e in n-doped 1L-WS₂ due to TFSI [57–62, 65–67]. Nevertheless, we cannot neglect the additional charge density induced by V on the MIS capacitor. E.g. the I-V characteristics in figure 7 show that I and V$_{ON}$ do not change before and after TFSI, suggesting the same tunneling condition is maintained across the 1L-WS₂/hBN/SLG junction. In both cases a comparable electric field (and charge) is developed across the junction for a given V. Figure 7 implies that, independent of TFSI treatment, the same amount of negative charge is electrostatically induced in 1L-WS₂ at V $\gt$ 0. However, taking into account the EL spectral shift towards X⁰ emission upon bias, the expected depletion of excess e in 1L-WS₂ cannot explain the electrical behaviour of figures 10(b) and (c). Thus, the emission profile is not compatible with the I-V curves before and after TFSI in figure 7, given that the electric field across the junction should be modified by the e density change in 1L-WS₂.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

To get a better insight on the effects of TFSI on 1L-WS₂ based LEDs, figures 10(d) and (e) plot normalized PL spectra as a function of V before and after TFSI. In the pristine case (figure 10(d)), the PL map shows an evolution in emission spectra from ∼620 nm (∼2.000 eV) to ∼638 nm (∼1.943 eV), corresponding to a spectral shift from X⁰ to X⁻ due to excess e in 1L-WS₂ induced by V. After TFSI treatment (figure 10(e)), the PL exhibits only a minor shift from ∼618 nm (∼2.006 eV) to ∼622 nm (∼1.993 eV), implying that the induced e-charge in 1L-WS₂ does not contribute to the X⁻ emission pathway. Therefore, similar to figures 10(a) and (b), PL also indicates that the emission after TFSI treatment predominantly originates from radiative recombination of X⁰, independent of V. References [57, 59, 61–64] claimed that TFSI treatment reduces the extent of n-type behavior in S-based 1L-TMDs due to S vacancies passivation, consistent with the suppression of X⁻ formation in references [60, 65–68]. Reference [8] reported that TFSI acts as a Lewis acid, i.e. it can accept an e pair from a donor [56], suppressing X⁻ formation. Whereas references [53–55] claimed that TFSI may activate sub-gap states and reduce the n-type behavior in S-based TMDs, as well as reducing X⁻ formation. Our I-V, EL and gated-PL results suggest that TFSI treatment (i) depletes the excess e in 1L-WS₂, acting as a Lewis acid [8] and (ii) favours the radiative recombination of X⁰ independent of bias, due to the activation of trapping states [53, 55] in 1L-WS₂ caused by the treatment. One would expect changes in the excitonic emission at such trapping states at RT, where the thermal energy can assist carrier de-trapping, and radiative recombination from excitons [67]. Therefore, the modification from non-radiative to radiative recombination by activation of trapping states could be further engineered to achieve more efficient optoelectronic devices.

We demonstrated a one order of magnitude enhancement in EL emission of 1L-WS₂-LEDs by performing TFSI treatment. EL predominantly originates from trions in pristine devices, while neutral excitons dominate in treated ones. The neutral excitonic emission is also restored in 1L-WS₂ gated-PL measurements. We attribute these changes to a reduction of n-doping of 1L-WS₂, as well as changes in the relaxation and recombination pathways within 1L-WS₂. This paves the way to more efficient 1L-TMDs-based LEDs, and sheds light into tunability of the excitonic emission of these devices.

3.1. Raman characterization of LMH individual constituents

Raman spectroscopy allows one to monitor LMs at every step of device fabrication. This should always be performed on individual LMs before and after assembly in LMHs and devices. This is an essential step to ensure reproducibility of results but, unfortunately, it is often neglected in literature.

Ultralow-frequency (ULF) Raman spectra in the range ∼10–50 cm⁻¹ probe shear (C), corresponding to layer motion parallel to the planes, and layer breathing modes (LBM), corresponding to the motion perpendicular to them [102, 146–148]. Pos(C)_N can determine the number of layers [146–149] N = $\pi({2\cos^{-1}[\frac{\textrm{Pos}(C)_{N}}{\textrm{Pos}(C)_\infty}]})^{-1}$, with Pos(C)$_\infty$ the bulk Pos(C).

Figure 11 plots the Raman spectra of non-treated 1L-WS₂ and bulk-WS₂. In figure 11(a), the C mode and LBM are not observed for 1L-WS₂, as expected [146–148]. In bulk-WS₂, Pos(C) $\sim26.9\,\pm\,0.14$ cm⁻¹. The spectral resolution $\pm\,0.14$ cm⁻¹ for the ULF region is obtained as for reference [95]. We observe two additional peaks ∼28.7$\,\pm\,0.14$ cm⁻¹ and 46.4$\,\pm\,0.14$ cm⁻¹, respectively, in agreement with references [150–152]. These do not depend on N [150, 151] and are seen because 514.5 nm (∼2.41 eV) is nearly resonant with the B exciton (∼2.4 eV) of 1L-WS₂ [9, 153–156], and ∼20 meV above the bulk-WS₂ B exciton (∼2.38 eV) [9, 154]. This gives rise to a resonant process [9, 153–156], which occurs because the laser energy matches the electronic transition of the B exciton, revealing features associated with intervalley scattering mediated by acoustic ph [157–159]. A similar process also happens in 1L-MoS₂ [150, 151] and other 1L-TMDs [157–159]. Although our ULF filters cut ∼5 cm⁻¹, the LBM is not detected in bulk-WS₂, as its frequency is expected to be $\lt$10 cm⁻¹ [148], because this resonant process with a 514.5 nm laser reduces the signal to noise ratio in this spectral region [150].

Zoom In Zoom Out Reset image size

The high-frequency (HF) Raman spectra of non-treated 1L-WS₂ and bulk-WS₂ (figure 11(b)) show various peaks, table 2. The first order Raman modes are E$^{{^{\prime}}}$, A$^{{^{\prime}}}_1$ in 1L-WS₂ [77–80] and E$^1_{2g}$, A$_{1g}$ in bulk-WS₂ [77–80]. E$^{{^{\prime}}}$ (E$1_{2g}$) and A$^{{^{\prime}}}_1$ (A$_{1g}$) correspond to in-plane and out-of-plane optical ph for 1L(bulk)-WS₂. Their nomenclature for 1L and bulk differs due to the different crystal symmetry [77–80]. In 1L-WS₂ we get Pos(E$^{{^{\prime}}}$) ∼356.8$\,\pm\,0.2$ cm⁻¹, FWHM(E$^{{^{\prime}}})\sim$3.2$\,\pm\,0.2$ cm⁻¹, Pos(A$^{{^{\prime}}}_1$) ∼418.5$\,\pm\,0.2$ cm⁻¹, FWHM(A$^{{^{\prime}}}_1$) ∼4.3$\,\pm\,0.2$ cm⁻¹. In bulk-WS₂ we have Pos(E$^1_{2g}$) ∼356.8$\,\pm\,0.2$ cm⁻¹, FWHM(E$^1_{2g}$) ∼1.5$\,\pm\,0.2$ cm⁻¹, Pos(A$^{{^{\prime}}}_1$) ∼420.8$\,\pm\,0.2$ cm⁻¹, FWHM(A$^{{^{\prime}}}_1$) ∼2.1$\,\pm\,0.2$ cm⁻¹. In 1L-WS₂ the difference in peaks' position [Pos(E$^{{^{\prime}}}$)-Pos(A$^{{^{\prime}}}_1$)] is ∼61.7 cm⁻¹ while this is ∼64.0 cm⁻¹ in bulk-WS₂, further corroborating the identification of 1L [77]. In the HF spectra of 1L- and bulk-WS₂ we also observe the 2LA(M) mode, involving two longitudinal acoustic (LA) ph close to the M point [77–79]. For 1L-WS₂, Pos(2LA(M)) ∼351.9$\,\pm\,0.2$ cm⁻¹ and FWHM(2LA(M)) ∼9.2$\,\pm\,0.2$ cm⁻¹, whereas for bulk-WS₂, Pos(2LA(M)) ∼350.6$\,\pm\,0.2$ cm⁻¹ and FWHM(2LA(M)) ∼8.3$\,\pm\,0.2$ cm⁻¹. The 2LA(M) mode originates from a second-order double resonant process [157–159], where momentum conservation is satisfied by two LA ph with opposite momenta around K- and M [158], therefore sensitive to differences in band structure between bulk and 1L-WS₂ [77, 160].

I(A$_{1g}$)/I(E$_{1g}$) ∼3.2 in bulk-WS₂, with I the peak height, is higher than I(A$^{{^{\prime}}}_1$)/I(E$^{{^{\prime}}}$) ∼0.8 in 1L-WS₂. I(2LA)/I(E$_{1g}$) ∼1 in bulk-WS₂ is lower than I(2LA(M))/I(E$^{{^{\prime}}}$) ∼1.7 in 1L-WS₂. This can be explained considering that the main first-order (E$^{{^{\prime}}}$, A$^{{^{\prime}}}_1$) and second-order (2LA(M)) Raman modes are enhanced for 2.41 eV excitation, due to exciton-ph coupling effects involving B exciton transitions [155, 161]. These depend on mode symmetry (i.e. differ between out-of-plane and in-plane modes) as well as N [157]. In bulk-WS₂, the out-of-plane A$_{1g}$ is resonant with the B exciton, unlike E$^1_{2g}$ [157]. The enhancement of A$_{1g}$ decreases with decreasing N due to the dependence of the lifetime of the intermediate excitonic states on N [157]. The difference between I(2LA)/I(E$^{{^{\prime}}}_1$) in 1L-WS₂ and I(2LA)/I(E$^1_{2g}$) in bulk-WS₂ is due to a change in band structure from direct bandgap in 1L to indirect in bulk-WS₂ [77–80], which changes the double resonance conditions [157–159].

The Raman spectrum of 1L-WS₂ also shows 8 peaks in the range 170–350 cm⁻¹ (figure 10(b) and table 2). LA(M) and LA(K) correspond to one-ph processes originating from the LA branch at M- and K, respectively [77–80]. Since LA(M) and LA(K) and E$^2_{2g}$(M) are one-ph processes from the edge of the Brillouin Zone (q≠0) [77–80], they should not be seen in the Raman spectra considering that, due to the Raman fundamental selection rule [162], one-ph processes are Raman active only for ph with q ∼ 0, whereas for multi-ph scattering the sum of ph momenta needs to be ∼0 [157–160]. However, these modes can be activated in presence of defects, as these can exchange momentum with ph, such that the sum of the momenta in the process is ∼0 [77–80]. A$_{1g}$(K)-LA(K), A$_{1g}$(M)-LA(M), A$_{1g}$(M)-ZA(M), LA(M)+TA(M) in bulk-WS₂ and A$^{{^{\prime}}}$(K)-LA(K), A$^{{^{\prime}}}_1$(M)-LA(M), A$^{{^{\prime}}}_1$(M)-ZA(M), LA(M)+TA(M) in 1L-WS₂ are combinational modes, and Raman allowed [77–80]. E$^2_{2g}$(M) correspond to a one-ph process originating from the transverse optical (TO) branch at the M-point [77–80]. E$^2_{2g}$(Γ) is a degenerate mode originating from the LO and TO branches at Γ [77–80].

Figure 12 plots the Raman spectra of a ∼3 nm hBN flake (black curves) and bulk-hBN (red curves). The latter has 2 Raman-active modes [163, 164], C and E$_{2g}$. In figure 12(a) Pos(C)$_\infty$ = 52.3$\,\pm\,$0.14 cm⁻¹ with FWHM ∼0.7$\,\pm\,0.2$ cm⁻¹ for bulk-hBN and Pos(C)_N = 50.4$\,\pm\,$0.14 cm⁻¹, FWHM ∼0.8$\,\pm\,0.2$ cm⁻¹ for the hBN flake. In bulk-hBN Pos(C)$_\infty$ = $\frac{1}{\pi c}\sqrt{\frac{\alpha}{\mu}}$ = 52.3 cm⁻¹, with µ = 6.9×10²⁷kgA⁻² the mass of one layer per unit area, c the speed of light in cm s⁻¹, and α the spring constant associated to the coupling between adjacent layers [95, 148]. From this, we get $\alpha = 16.9\times10^{18}$Nm⁻³. From N = $\pi \{{2\cos^{-1}[\frac{\textrm{Pos}(C)_{N}}{\text{Pos}(C)_\infty}]}\}^{-1}$, we get N = 6$\,\pm\,$1 for the 3 nm thick flake (measured with a Dimension Icon Bruker AFM in tapping mode) as shown in the inset of figure 12(b)). In figure 12(b) Pos(E$_{2g}$) ∼1368.5$\,\pm\,0.2$ cm⁻¹ and FWHM(E$_{2g}$) ∼9.1$\,\pm\,0.2$ cm⁻¹ for FL-hBN, and Pos(E$_{2g}$) ∼1367$\,\pm\,0.2$ cm⁻¹ with FWHM(E$_{2g}$) ∼7.6$\,\pm\,0.2$ cm⁻¹ for bulk-hBN. The peak broadening ∼1.5 cm⁻¹ in FL-hBN can be attributed to strain variations within the laser spot, as thinner flakes conform more closely to the roughness of the underlying SiO₂ [95]. This is consistent with the fact that thicker hBN have lower root mean square (RMS) roughness [88, 92, 95, 165], e.g. 300 nm SiO₂ has RMS roughness ∼1 nm [92], 2–8 nm hBN has RMS roughness ∼0.2–0.6 nm [95], while $\gt$ 10 nm hBN thick has RMS roughness ∼0.1 nm [88, 92].

Zoom In Zoom Out Reset image size

The red curves in figures 13(a) and (b) are the Raman spectra of SLG on SiO₂/Si before LMH assembly. Pos(G) = 1586.9$\,\pm\,0.2$ cm⁻¹ with FWHM(G) = 7.7$\,\pm\,0.2$ cm⁻¹, Pos(2D) = $2685.2\,\pm\,$0.2 cm⁻¹ with FWHM(2D) ∼29.3$\,\pm\,0.2$ cm⁻¹, I(2D)/I(G) ∼0.85, A(2D)/A(G) ∼3.3. These indicate a p-doping [102–104] with E$_\textrm{F}\sim200\,\pm\,$50 meV. No D peak is observed, thus negligible defects [101–103]. Pos(G) and Pos(2D) are affected by the presence of strain [102, 103]. Biaxial strain can be differentiated from uniaxial from the absence of G-peak splitting with increasing ε [166, 167], however at low ($\unicode{x2A7D}$0.5%) ε the splitting cannot be resolved [166, 167]. Thus, the presence (or coexistence) of biaxial strain cannot be ruled out. For uniaxial(biaxial) strain, Pos(G) shifts by ΔPos(G)/$\Delta\epsilon$ ≈23(60) cm⁻¹/% [166, 167]. Pos(G) also depends on doping [104, 107]. E$_F\sim200\,\pm\,$50 meV should correspond to Pos(G) ∼1584.3 cm⁻¹ for unstrained SLG [107]. However, in our experiment Pos(G) ∼1586.9$\,\pm\,0.2$ cm⁻¹, which implies a contribution from compressive uniaxial (biaxial) strain ∼0.1% (∼0.04%). The black curves in figures 12(a) and (b) show the Raman spectrum of the FLG electrode on SiO₂/Si. Pos(G) $\sim1581.2\,\pm\,$0.2 cm⁻¹ with FWHM ∼12$\,\pm\,0.2$ cm⁻¹, Pos(2D₁) $\sim2694.0\,\pm\,$0.2 cm⁻¹ with FWHM ∼48$\,\pm\,0.2$ cm⁻¹, and Pos(2D₂) $\sim2725\,\pm\,$0.2 cm⁻¹ with FWHM ∼33$\,\pm\,0.2$ cm⁻¹. Pos(C)$_N\sim$41.4$\,\pm\,0.14$ cm⁻¹, corresponding to N = 5.

Zoom In Zoom Out Reset image size

3.2. Electrical characterization

3.3. Trionic PL and EL emission in 1L-WS₂ LEDs

Figures 14(a) and (b) show the gated-PL spectra with a continuous shift for larger V, with a ∼20 nm shift, consistent with reference [28]. To further corroborate that both gated-PL and EL come from X⁻, we fabricate 3 types of 1L-WS₂ based LEDs, figure 14(c): (i) 1L-WS₂ on top of hBN and SLG layers; (ii) 1L-WS₂ encapsulated by 2 hBN layers (top and bottom), and SLG on the top-hBN layer; (iii) 1L-WS₂ sitting directly onto SiO₂/Si, capped with hBN and SLG. All devices are made following similar procedures. Electrical connection is done by applying V between SLG and 1L-WS₂. The hBN tunnel barrier is 2–4 nm thick. All devices are then checked by I-V measurements, gated-PL, and EL. Figure 14(c) demonstrates that all present a good match for EL and PL emission, in agreement with the X⁻ emission of figure 4. We then TFSI treat 4 devices, those with 1L-WS₂ at the top of the stack, allowing direct comparison before and after treatment.

Zoom In Zoom Out Reset image size

3.4. Spectrometer efficiency

We acknowledge funding from the EU Graphene and Quantum Flagships, EU grant Graph-X, ERC Grants Hetero2D, GSYNCOR and GIPT, EPSRC Grants EP/K01711X/1, EP/K017144/1, EP/N010345/1, EP/L016087/1, EP/V000055/1, EP/X015742/1, JSPS KAKENHI (Grant Numbers 21H05233 and 23H02052) and World Premier International Research Center Initiative (WPI), MEXT, Japan.

All data that support the findings of this study are included within the article (and any supplementary files).