In two-dimensional (2D) semiconducting transition metal dichalcogenides (TMDs), new electronic phenomena such as tunable bandgaps1,2,3 and strongly bound excitons and trions emerge from strong many-body effects4,5,6, beyond the spin and valley degrees of freedom induced by spin–orbit coupling and by lattice symmetry7. Combining single-layer TMDs with other 2D materials in van der Waals heterostructures offers an intriguing means of controlling the electronic properties through these many-body effects, by means of engineered interlayer interactions8,9,10. Here, we use micro-focused angle-resolved photoemission spectroscopy (microARPES) and in situ surface doping to manipulate the electronic structure of single-layer WS2 on hexagonal boron nitride (WS2/h-BN). Upon electron doping, we observe an unexpected giant renormalization of the spin–orbit splitting of the single-layer WS2 valence band, from 430 meV to 660 meV, together with a bandgap reduction of at least 325 meV, attributed to the formation of trionic quasiparticles. These findings suggest that the electronic, spintronic and excitonic properties are widely tunable in 2D TMD/h-BN heterostructures, as these are intimately linked to the quasiparticle dynamics of the materials11,12,13.
Coulomb interactions in 2D materials are several times stronger than in their 3D counterparts. In 2D TMDs, this is most directly evidenced by the presence of excitons with binding energies an order of magnitude higher than in the bulk4. Although the excitons in these 2D materials have been widely studied by optical techniques13, the impact of strong electron–electron interactions on the quasiparticle band structure remains unclear. Theory predicts that many-body effects will influence the spin–orbit splitting around the valence-band maximum (VBM) and conduction-band minimum (CBM)14. Although these should be observable by ARPES, a direct probe of many-body effects15, measurements so far have mainly focused on the layer-dependence of the single-particle spectrum and the direct bandgap transition in 2D TMD systems, including epitaxial single-layer MoSe2 (ref. 16) and single-layer WSe2 (ref. 17) grown on doped multilayer graphene, and single-layer MoS2 grown on a metal surface18. On such conductive substrates, the interfacial interactions and screening are known to strongly influence the electronic properties of the single-layer TMD9.
Flakes of single-layer TMDs have been transferred onto oxide substrates such as SiO2, where the substrate screening and interfacial effects are potentially reduced. However, resulting ARPES spectra have been too broad for detailed analysis19, probably owing to surface roughness and charge impurity scattering20. Compared with SiO2 and similar substrates, h-BN has favourable qualities such as atomic flatness, modest screening and a homogeneous charge distribution. This should enable direct investigation of the intrinsic electronic structure and many-body effects of the adjacent TMD. Hexagonal BN is often used as a substrate for graphene heterostructures8,20 with high device performance21 and new exotic electronic states such as quantized Dirac cones22. Unfortunately, the lateral size of mechanically assembled heterostructures is usually of the order of 10 μm, much smaller than the beam spot of typical ARPES set-ups (≳100 μm). Furthermore, sample charging on insulating bulk h-BN substrates would complicate ARPES experiments.
We overcome these challenges as follows. We realize a high-quality 2D semiconductor–insulator interface by mechanically transferring a relatively large (~100 μm) single-layer WS2 crystal onto a thin flake of h-BN that has itself been transferred onto a degenerately doped TiO2 substrate, as depicted in Fig. 1a. Sample charging is avoided because there is electrical contact from the continuous single-layer WS2 flake to both the h-BN and the conductive TiO2. Figure 1b is an optical microscope image of the sample, including a flake of h-BN, approximately 100 μm wide, surrounded by several transferred flakes of single-layer WS2 on the TiO2 substrate (WS2/TiO2), one of which partially overlaps the h-BN.
By using a state-of-the-art spatially resolved microARPES experiment with a 10 μm focused synchrotron beam spot, we are able to collect distinct high-quality band-structure information from the multiple micrometre-scale interfaces. A spatial map of the photoemission intensity around the WS2/h-BN heterostructure is shown in Fig. 1c, which was produced by integrating the intensity over the boxed region of the corresponding k-space band structure shown in Fig.1d, measured at each spatial point. The crossing single-layer WS2 and h-BN bands in this region ensure strong contrast between WS2/h-BN (white arrow), regions of WS2/TiO2 (light purple) and regions of bare TiO2 (dark purple) in the spatial map. The photoemission map (Fig. 1c) corresponds directly to the optical micrograph (Fig. 1b) with contrasts that reflect the intensity of the WS2 and h-BN features in the red box in Fig. 1d. The band structures from bare TiO2, WS2/TiO2 and several spots within the WS2/h-BN heterostructure are presented in Supplementary Fig. 1. The slight intensity variations within the WS2/h-BN heterostructure arise from areas with pinholes introduced in the single-layer WS2 during transfer23, as sketched in Fig. 1a. The sensitivity towards such features, which are not resolved by the optical micrograph, demonstrates the ability to identify optimum sample areas directly in the ARPES experiment, which is critical for such complex, heterogeneous samples.
The valence-band electronic structure through the entire first Brillouin zone of the heterostructure, including the single-layer WS2 bands and the π-band dispersion of the underlying h-BN, is shown in Fig. 1d,e. The data are collected from a single spatial point where the WS2 features are most intense (white arrow in Fig. 1c). Custom electrostatic deflectors mounted in the photoelectron analyser enable full scans of k-space at exactly this position without any drift from sample motion. The Brillouin zone orientations and twist angle between the two materials are determined from the constant binding energy cuts shown in Fig. 1f–h. From the relative orientation of the hole pockets, we estimate a twist angle of (23 ± 1)°. The energy distribution curves (EDCs) in Fig. 1i,j track the VBM binding-energy positions of the two materials. The upper valence band of single-layer WS2 is located inside the bandgap of h-BN and the single-layer WS2 VBM is characterized by a spin–orbit splitting of 430 meV (see Fig. 1j), in agreement with theoretical predictions12 and previous experiments23,24. The clear electronic states and lack of band hybridization reveal a weak interlayer interaction between the two materials. Similar to work22 on graphene/h-BN, we expect these data to represent the intrinsic electronic structure of single-layer WS2 with negligible substrate influence.
The impact of electron doping on the electronic structure through in situ surface potassium deposition is shown in Fig. 2a,b (see Supplementary Fig. 2 for core level data on clean and potassium-dosed samples). Doping WS2/h-BN leads to the CBM being populated at the points of single-layer WS2, confirming the expected direct bandgap. A surprising change of the dispersion of the two spin–orbit-split valence bands VBA and VBB of WS2/h-BN around the point is observed, highlighted in the EDCs in Fig. 2c. The spin-splitting due to spin–orbit coupling ΔSO increases from 430 meV in the undoped case to 660 meV in the electron-doped case, as sketched in Fig. 2d. Such a large spin-splitting has not previously been observed in any single-layer material to our knowledge. In this case, the bandgap of single-layer WS2 is 1.65 eV and the CBM to VBB offset is 2.31 eV. These values are sketched in Fig. 2d and denoted as EA and EB, respectively, owing to the relation with the A and B exciton lines observed in optical experiments5.
A detailed evolution of the band extrema with increasing doping is shown in Fig. 3a–e. The dispersions around the VBM as determined from EDC line shape analysis (see Supplementary Fig. 6) are shown by dashed red curves and directly compared in Fig. 3f. We estimate the charge carrier density, N, at each dosing step from the CBM position (see Methods). These estimated doping levels are consistent with those achieved in similar experiments25 on bulk WSe2 and with the intensity of the potassium 3p core level (see Supplementary Fig. 3).
From EDC peak positions at (see Supplementary Fig. 7), we extract the VBM and CBM energies as a function of dosing (Fig. 3g). After the first dosing step (N = 1.7 × 1013 cm−2), the CBM becomes occupied, and VBA and VBB rigidly shift to higher binding energy. With further dosing, the CBM moves further down to higher binding energy, while VBA and VBB shift back towards lower binding energy, resulting in a continuous narrowing of the bandgap. In particular, the dispersion of VBA appears to renormalize with increased doping (see Fig. 3d–f), with a strong increase in ΔSO (Fig. 3g,h). This leads to a corresponding change in the relative energy separation between EA and EB (Fig. 3i), implying that the energies of the A and B exciton lines also separate. Data points with different marker shapes and colours in Fig. 3h,i stem from separate doping experiments on the three different flakes studied in Figs. 2 and 3 and in Supplementary Fig. 8. A reproducible trend is found across all flakes. Note also that in the carrier density range between 2 × 1012 cm−2 and 1.0 × 1013 cm−2, we find a more modest bandgap renormalization of (90 ± 30) meV, which is in excellent agreement with gated device measurements on a similar sample1. Our observations reveal that it is insufficient to consider only rigid band shifts, and that strong dispersion changes can result from doping of single-layer TMDs.
The surprising doping-induced changes in ΔSO are probably not directly related to the surface potential induced by the potassium deposition (through, for example, the Rashba interaction), which is not expected to affect ΔSO at for single-layer TMDs. This would introduce a splitting at , which we do not observe26,27. Furthermore, we can rule out any potassium-induced structural symmetry breaking in our heterostructure, as only minor rigid binding-energy shifts of the S 2p core levels of WS2 and of the underlying h-BN π-band are observed after complete doping (see Supplementary Figs. 4 and 5). The reproducible charge carrier dependence of the spectral changes demonstrated in Figs. 2 and 3 and Supplementary Fig. 8 suggests that these changes originate from Coulomb interactions around the band extrema of single-layer WS2 (refs 1, 2,14).
The linewidths of the VBA and CBM peaks exhibit a non-monotonic dependence on doping, which cannot be described by simple scattering on ionized potassium impurities (see further discussion in Supplementary Section 4). Specifically, the observation that VBA renormalization coincides with occupation of the CBM suggests that the renormalization is caused by new scattering channels available upon occupation of the conduction band. Previous works using surface potassium deposition for electron-doping of single-layer TMDs on conductive substrates16,17,18, where the Coulomb interactions are already strongly screened in the undoped case9, have shown no such changes in ΔSO. We believe that the reduced dielectric constant of the h-BN substrate plays a key role for these observations as it leads to reduced screening of the many-body interactions in the bare single-layer WS2.
An alkali-atom-induced renormalization of the valence band edge at K towards EF, observed in several bulk TMDs, has been attributed to the breaking of the degeneracy of the outermost layers by the doping-induced field25,28. This can either be a single-particle effect28 or a combination of single- and many-body effects25, the latter of which suggests a negative electronic compressibility (NEC), the motion of the chemical potential μ towards the VBM, i.e. dμ/dN < 0.
In contrast, we observe distinctive effects in single-layer WS2/h-BN, namely (1) a renormalization of ΔSO within the single layer, (2) an NEC in which |dμ/dN| is significantly larger than in the bulk25,28, and (3) a VBA slope that is discontinuous at k = (,) ± ~0.15 Å−1 in Fig. 2b at high doping. This leads to kinks in VBA, exemplified by the arrow in Fig. 3e, which develop continuously in strength with doping in Fig. 3a–e. Such kinks are common in ARPES when the created ‘photohole’ interacts strongly with well-defined (in energy and/or momentum) excitations15.
As noted above, the band renormalization coincides with the occupation of the conduction band, suggesting that such excitations are associated with electron–hole (e–h) pairs near EF in the conduction band, induced in response to the valence-band hole created during photoemission. In the undoped situation sketched in Fig. 4a, such interactions are not possible. At high carrier densities where the conduction band is occupied, an e–h excitation around (or ) can interact with the valence-band photohole, forming a positively charged, bound electron–hole–hole complex denoted as echchv, where (c, v) denotes charges in the (conduction band, valence band), respectively, as illustrated in Fig. 4b. Such a process would renormalize the bare band dispersion and lifetime of the valence-band states, broadening and shifting their spectra as observed.
These excitations may be compared to the X+ (X−) trions found in p- (n)-doped TMDs with configuration echvhv (ecechv). Such trions have been invoked to interpret additional spectral lines shifted in energy by ~20–60 meV with respect to the main A exciton line in optical absorption6 and luminescence6,29 measurements of single-layer TMDs. Our measurements show a relative shift of VBA by ~0.23 eV compared with VBB, which reflects the absolute binding-energy reduction ΔEehh of the photohole associated with the formation of the trion (see Fig. 4b). The order of magnitude of ΔEehh is compatible with the absolute trion binding energies that can be extracted from the optical experiments6,29, but the exact values are expected to depend on the dielectric environment of the sample and the doping. So far, corresponding trion features associated with B excitons have not been observed in optical experiments, which is fully consistent with the absence of renormalization of the VBB dispersion in the present ARPES data, as seen in Fig. 3a–e and as sketched in Fig. 4c. This lack of renormalization of the VBB dispersion might be attributed to additional decay channels of the VBB hole such as decay into the VBA band. The strong increase in ΔSO and the bandgap renormalization can therefore be viewed as direct consequences of forming trionic quasiparticles around the VBA and conduction-band extrema.
The assignment of trions in optical measurements of semiconducting TMDs is currently being debated, as a recent theoretical study points towards the possibility of interactions between e–h pairs and the remaining charge density forming other types of quasiparticles such as exciton-polarons30. Our ARPES measurements provide direct evidence for such multi-component excitations in single-layer TMDs and give access to both their energy and momentum dependence that is lacking from momentum-integrating transport, optical or tunnelling measurements. We envisage further theoretical and experimental studies to disentangle such many-body effects in the spectral function of single-layer TMDs. The charge-carrier-dependent electronic bandgap and spin-splitting that arise from these many-body effects will profoundly impact the charge-, spin- and valley-dependent dynamics and transport properties of devices, as well as the interpretation of excitonic effects.
Fabrication of WS2/h-BN heterostructures
The heterostructures were prepared by successively transferring few-layer h-BN (commercial crystal from HQ Graphene) and then single-layer WS2 onto 0.5 wt% Nb-doped rutile TiO2(100) (Shinkosha Co., Ltd). A thin film of polycarbonate (PC) was mounted onto polydimethylsiloxane (PDMS) on a glass slide to prepare a PC/PDMS stamp. This stamp was first used to pick up h-BN flakes from a SiO2 substrate and then dropped onto the TiO2 substrate under a microscope. The transferred h-BN flakes were cleaned of any polymer residue by annealing at 625 K in ultrahigh vacuum (UHV) for 1 h. Next, single-layer WS2 flakes were picked up from the SiO2 growth substrate and aligned to drop onto h-BN, such that a part of the flake made contact with the TiO2 substrate. The process was followed by another annealing step in UHV to clean off any remaining residues.
Experimental details of microARPES
The samples were transported through air to the Microscopic and Electronic Structure Observatory (MAESTRO) at the Advanced Light Source where they were inserted in the microARPES UHV end-station with a base pressure better than 5 × 10−11 mbar. The samples were given a mild anneal at 600 K prior to measurements, in order to desorb adsorbates from air. The synchrotron beam-spot size was on the order of 10 μm for the photon energies of 145 eV and 76.5 eV used to obtain the microARPES data. The data were collected using a hemispherical Scienta R4000 electron analyser equipped with custom-made deflectors that enable collecting ARPES spectra over a full Brillouin zone without moving the sample. Potassium-dosing experiments were carried out by evaporating potassium in situ from a SAES getter source mounted in the analysis chamber such that dosing is performed in an optimum sample measurement position without moving the sample. Core level data of undoped and potassium-dosed samples are presented in Supplementary Fig. 2 to document the cleanliness of the samples. The data in Figs. 1 and 2 are from the same sample, whereas the data in Fig. 3 and Supplementary Fig. 8 were obtained on fresh samples. The charge carrier density N is estimated using a simple model of a 2D parabolic band given by , where the factors gv = 2 and gs = 2 take spin- and valley-degeneracy into account, is the effective mass of the single-layer WS2 conduction band obtained from ref. 12, kB is Boltzmann’s constant, ħ is the reduced Planck constant, T is the sample temperature and E − EF is determined from the fitted CBM position. The total energy resolution and momentum resolution in the microARPES data were better than 20 meV and 0.01 Å−1, respectively. Measurements and dosing experiments were carried out at both 85 K and at 20 K, without any noticeable change in behaviour between the two temperatures.
The data that support the plots within this paper and other findings of this study are available from the corresponding author upon reasonable request.
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
We thank A. H. MacDonald for discussions. S.U. acknowledges financial support from the Danish Council for Independent Research, Natural Sciences, under the Sapere Aude programme (grant no. DFF-4090-00125) and from VILLUM FONDEN (grant no. 15375). R.J.K. is supported by a fellowship within the postdoctoral programme of the German Academic Exchange Service (DAAD). S.M. acknowledges support by the Swiss National Science Foundation (grant no. P2ELP2-155357). The work at Ohio State was primarily supported by NSF-MRSEC (grant DMR-1420451). Work at the US Naval Research Laboratory (NRL) was supported by core programmes and the NRL Nanoscience Institute, and by the Air Force Office of Scientific Research under contract number AOARD 14IOA018- 134141. This research used resources of the Advanced Light Source, which is a US Department of Energy Office of Science User Facility under contract no. DE-AC02-05CH11231.