For decades, semiconductor superlattices (SLs), periodically layered structures of two alternating semiconductors in the atomic thickness regime, have served as the material platform of various heterojunction devices in modern electronics, photonics and display technology1,2. Such prominent SL examples can be listed as III-V compound semiconductor SLs (i.e., GaAs/AlGaAs and GaInAs/AlInAs) for high electron mobility transistors3,4 and quantum cascade lasers5,6, GaN/AlGaN SLs for light emitting diodes7,8 and Si/Ge SLs for strained Si CMOS9. Therein, the constituent semiconductors are “covalent-bonded” across the heterointerfaces with the lattice-matching coherences10-12, where the two-dimensional (2D) charge carriers form diverse quantum well (QW) structures, depending on the degrees of interlayer coupling strengths13. Meanwhile van der Waals (vdW) semiconductors, which often stemmed from transition-metal dichalcogenides (TMDCs, MX2, where M and X represent transition-metal ions and chalcogen ions), naturally ensue the inherent 2D confinements within the unit monolayer (ML) across the chemical-bond free vdW gaps14-16. As a broad range of TMDC semiconductors are available as a ML crystal with diverse electronic structures, precise integration of each kind into vdW-SLs can generate another category of the QW structures for unexplored functionalities17-19. Despite extensive research on the bilayer stacks of dissimilar vdW-MLs to investigate new types of interlayer excitations, such as interlayer excitons20-23 and twist-angle dependent strong correlations24,25, little is known for the QW states of vdW-SLs 26,27, mainly due to fact that they are not synthetically available to date28. Otherwise, they can be prepared by manual transfer-based stacking with some atmospheric impurities, which may not be unavoidable for the scalable integrations29-31. In this work, we report direct growth of vdW-SLs, heteroepitaxially stacked with MoS2, WS2 and WSe2 MLs, by metalorganic chemical vapour depositions (MOCVD)32,33. We have achieved precise ML-by-ML sequential stacking with atomically clean and sharp heterointerfaces by kinetic control of heteronucleation in the near-equilibrium limit32,34,35. This ML-by-ML stacking epitaxy also enables to realize the tunable vdW SL electronic structures in ML precision. We identified several atomic stacking orders at the vdW heterointerfaces, and present scalable valley polarized optical excitations that only pertain to a series of 2D type II band alignments.
Heteroepitaxial growth of vdW semiconductor superlattices
Elemental variation in the MX2 SLs can be achieved either in M-alteration (MX2/M’X2) or X-alteration series (MX2/MX’2). We demonstrated both series in MoS2/WS2 and WS2/WSe2 SLs with time-lapse precursor modulations by MOCVD. The first example of MoS2/WS2 SLs was discussed in Fig. 1. The predetermined flow rates of Mo(CO)6 and W(CO)6 precursors for each MoS2 and WS2 ML were set to 4 sccm and 3 sccm on the (C2H5)2S background flow of 2 sccm in a given growth sequence. A series of cross-sectional high-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) images in Fig. 1a, taken in each growth step up to 7 ML stacks, demonstrate the precise ML-by-ML stacking of WS2 and MoS2. We have successfully achieved such vdW stacking epitaxy by the kinetics-controlled growth in the near-equilibrium limit, i.e., the lateral growth rate, ngrowth, in each ML was ~ 0.15 nm/min to guarantee the full lateral coverage, which is far slower than the cases of ~ 1500 nm/min in the usual thermal CVD growth33, by setting the lower growth temperatures (550 °C) and precursor partial pressures (~ 10-7 torr) of MO precursors. This optimized growth condition essentially suppresses the unwanted overgrowth and interlayer mixing36,37 - also see Supplementary Figs. 1-4 for the detailed growth optimizations. Here we stress that the basal planes of the first MLs with the highly preferred in-plane orientations are the prerequisite for the ML-by-ML growth. For that, we have grown the first ML films on the step-and-terrace terrains of c-sapphire substrates, which were then mechanically transferred on SiO2/Si substrates as the growth templates for the subsequent stacking growth. We fully discuss these features in Fig. 3. The layered structures of MoS2/WS2 SLs in greater details were presented in Fig. 1b, where the total 9 MLs (1:1 alternation of 5 WS2 MLs and 4 MoS2 MLs) were captured by the Z-contrast in HAADF-STEM image and energy dispersive X-ray spectroscopy (EDX) spectra. One can clearly identify the distinct intensity contrast between WS2 and MoS2 MLs, arising from the atomic number (Z) difference of ZMo = 42 and ZW = 74, where the brighter MLs are WS2 and the darker MLs are MoS2. In addition, the periodically alternating W-Lα and Mo-Kα peak intensity across each vdW gap, assures such SL modulations. Although the image in Fig. 1b cannot be atomically resolved with a fixed TEM zone axis in the entire area, due to finite in-plane orientation variants of each ML, when the zone axis is aligned to <11 0> of the lowest ML, the local atomic structures can be identified in the bright-field (BF) STEM images. For example, we observed either AC (translation) or AA' (180° rotation) stacking sequences between WS2 and MoS2 MLs with the > 90 % area coverage (Fig. 1c) – we call these stacking polytypes “coherent stacks” as classified in Supplementary Fig. 5. We also found local areas with some mixtures of random rotation stacks with the < 10 % area coverage (Fig. 1d) – we call these random orientation stacking polytypes “incoherent stacks”. Interestingly, we found different values of the interlayer distances for coherent and incoherent stacks to be 0.606 nm and 0.647 nm, arising from different vdW gap sizes, consistent with the calculations38: we also note that the value at the coherent stack, where individual atoms tend to reside at the thermodynamically stable coordinates, is notably smaller than the average value from mechanical stacking cases39. The synchrotron grazing incidence wide-angle X-ray diffraction (GI-WAXD) was employed to obtain larger-scale crystallographic coherence in both out-of-plane (i.e., qz) and in-plane (i.e., qxy) directions of our SLs, as in Fig. 1e. The 2D pattern of GIWAXS at the incidence angle of 0.12° in Fig. 1f shows a series of sharp Bragg spots in reciprocal space, indicating the highly ordered P63/mmc structures of our MoS2/WS2 SLs (9 ML stacks). For a reference, we compared GI-WAXD patterns from homoepitaxial MoS2 9L, which possesses smaller grain sizes of ~ 80 to 100 nm in Supplementary Fig. 6. The 1D line-cut profiles were analyzed in both out-of-plane (qz) and in-plane (qxy) directions, as shown in Fig. 1g,h. The qz line-cut profile at qxy = 2.3 Å-1 clearly shows 4 diffraction peaks from heteroepitaxial MoS2/WS2 SLs (purple) and homoepitaxial MoS2 9Ls (grey), verifying the highly ordered stacking textures – more pronounced peaks from MoS2/WS2 SLs indicate the higher degree of ordering, due to larger grain sizes. The interlayer distances are extracted from the (103) diffractions to be 0.616 nm, consistent with the value at the coherent stack regions measured from STEM images. Similarly we find it to be 0.617 nm from MoS2 9Ls - see also Supplementary Table 1 for the complete diffraction analyses. The vertical coherence length, Lc(103), which strictly quantifies the number of coherently repeating layers along the qz direction, can be also extracted to 1.87 nm by Scherrer equation – it is 1.49 nm in MoS2 9Ls. The in-plane diffraction peaks of (100), (110), and (200) are also distinct, and much pronounced in MoS2/WS2 SLs, suggesting the highly ordered in-plane textures with preferred orientations. Therein we extracted the (100) interplanar distance to be 0.273 nm from the (100) diffractions in the qxy line-cut profiles at qz = 0 Å-1 - see also Supplementary Table 2. Overall, we define the crystalline textures of our vdW SLs as largely coherent vdW vertical stacks, composed of in-plane oriented polycrystalline MLs.
Designer growth of vdW semiconductor superlattices with tunable periodicities
The established heteroepitaxial stacks of MoS2/WS2 SLs by a ML-by-ML mode enables to achieve designed SLs with arbitrary periodicities. Figure 2a,b demonstrate 1:2 MoS2/WS2 SLs and 2:2 MoS2/WS2 SLs. This growth tunability in our vdW SL heteroepitaxy is markedly contrasted from previous covalent-bonded SLs, which intrinsically suffer from the strict requirements of lattice matching heteroepitaxy upon stacking. We also show an example of X-alteration SLs in MX2, i.e., WSe2/WS2 SLs in Fig. 2c, where we employed (C2H5)2S and (C2H5)2Se precursors in each time-lapse on the continuous W(CO)6 background flow for the WS2 and WSe2 MLs (see also Supplementary Fig. 1 and 7 for the growth optimizations). Although the intensity contrast between WS2 and WSe2 MLs was less obvious in the Z-contrast in HAADF-STEM image, due to smaller Z-sensitivity, compared to MoS2/WS2 SLs, the periodic oscillation of S-Kα and Se-Kα peaks in the EDX spectra clearly validates SL modulations. Then, we have successfully established both M- and X-modulated SLs, heteroepitaxial WSe2/MoS2/WS2 trilayers, as shown in Fig. 2d. In addition, our heteroepitaxial growth can be extended to include graphene, which was intermittently inserted by an ex-situ dry-transfer method (Fig. 2E and Supplementary Fig. 8).
In-plane crystalline textures of vdW semiconductor superlattices
Heteroepitaxial evolution of in-plane crystal textures of our vdW SLs are investigated in WSe2/WS2/MoS2 trilayers, where the chemical compositions vary for both M and X, and the in-plane lattice constants also vary - they are known as 0.315 nm for bulk WS2 and MoS2, and 0.328 nm for bulk WSe240. Figure 3a-c are the schematic descriptions of each stacking during the successive growth, in which the initially formed multiple triangular facet crystals merge to form continuous MLs. Statistical variation in the in-plane crystal orientations can be verified in six-fold periodic clustering of (10 0) diffraction patterns with some degree of angular spreads (inset of Fig. 3d). The first bottom MoS2 ML was formed by multiple nucleation on the regular step and terrace terrains of c-sapphire substrates to initiate the preferred in-plane crystal orientations with a typical grain size of ~ 0.1 to 1 μm, which are either 0° or 60° rotated with respect to each other, as captured by a series of atomic force microscopy (AFM) images in Fig. 3d (see also Supplementary Figs. 9 and 10). This ML-by-ML growth proceeded upon the successive stacking growth for the second (WS2) (Fig. 3e) and third (WSe2) (Fig. 3f) MLs with smaller grains of 90 - 100 nm – see also in-plane TEM images in Supplementary Fig. 11 and grain size distribution in Supplementary Fig. 12. The larger lattice parameter of 0.331 nm in the third WSe2 ML was also verified from the (11 0) diffraction patterns, where we also measured those of underlying MoS2 and WS2 to be 0.322 nm (inset of Fig. 3f). Atomic scale images of such in-plane crystalline textures were directly captured by in-plane HAADF-STEM observations of partially covered bilayers, WS2/MoS2, as in Fig 3g-k. We verified that the top MLs are preferentially oriented with the basal planes of the bottom MLs, and the stacking polytypes are mostly either AA' (Fig. 3g and Supplementary Fig. 13) or AC coherent stacking (Fig. 3h). Such textures introduce grain boundaries (GBs), mainly identified as 0° or 60° GBs, which were interfaced between either AC-AC domains (Fig. 3k) or AC-AA' domains (Fig. 3j and Supplementary Fig. 14). We also observed random interlayer twists, showing Moiré interference patterns (Fig. 3i). According to first-principles calculations41, the formation energies of small angle (< 5°) and 60° GBs are relatively smaller than those of random angle GBs, suppressing the unwanted local overgrowth of the second MLs – for example, the second ML preferentially nucleates at the random angle GBs (other than 0° or 60° GBs), leading to non ML-by-ML growth. Whereas on the first MLs with 0° or 60° GBs, the nucleation is not locally concentrated leading to the ML-by-ML growth. We indeed observed such growth patterns, where on the smaller grained MoS2 ML templates with random GBs, the initial nuclei of the second WS2 are predominantly populated at such GBs (Supplementary Fig. 3). However, on the larger grained first ML with 0° or 60° GBs, the initial nuclei are uniformly distributed on the entire surfaces. It suggests that the preferred orientation growth in our work is critical to maintain the coherent ML-by-ML growth.
Valley-polarized interlayer excitations in type-II vdW superlattices
As for the electronic structures of our vdW SLs, we estimate them to be a series of type-II band alignments across the vdW gaps to a lowest order. Optical absorption spectra, collected from a series of SLs and each individual MLs in Fig. 4a, one can identify additive features of optical absorption with increasing ML stacks – the spectra of (MoS2/WS2)n SLs, where n is the bilayer stack numbers, are linear sums of WS2 and MoS2 MLs, according to Voigt fitting (Supplementary Fig. 15) – see also Supplementary Fig. 16 for a series of Raman scattering spectra, obtained from the same SL batches. One of the most distinctive electronic features that only pertain in TMDC ML heterostructures is the large and long-lived spin-valley polarization of charge carriers as observed earlier in (mechanically transferred) WS2/WSe2 and WSe2/MoS2 bilayers by pump-probe spectroscopy42. The ultrafast interlayer charge transfer process across the type-II alignment dramatically suppresses the exciton exchange interaction which is the major relaxation channel of valley polarization in TMDC MLs. We indeed observed scaling of such temporal population of the valley-polarized carriers, arising from a series of type-II band alignments in our MoS2/WS2 SLs. Figure 4b,c illustrate such valley-polarized excitations in real and momentum space with optical circular dichroism (CD), using time-resolved pump-probe spectroscopy. First, the pump pulses with the right-handed circular polarization (σ+) (blue arrows in Fig. 4b,c) creates the valley-polarized excitons on K valleys in MoS2 MLs. Then, the immediate interlayer hole transfer occurs across the type-II alignments from the MoS2 ML valence bands to that of WS2 MLs within the ultrafast timescale less than 50 fs43. Initially, we verified such interlayer charge separation from the substantially extended lifetime of the CD dynamics in our MoS2/WS2 bilayers, compared to that of MoS2 MLs (see Supplementary Fig. 17). Note that our SL series show shorter lifetime of valley polarization (~nsec), presumably due to higher defect densities, compared to mechanically exfoliated cases. Nevertheless, the valley-polarized electron selectively remains on the K valley of MoS2 MLs (see the upper panel of Fig. 4c). After the pump excitation, the delayed probe pulse (red arrows) was focused on the SLs, and by measuring the pump-induced differential reflectance ∆R with both σ+ and left-handed (σ-) circular polarization of the probe beam, we can evaluate CD = (∆Rσ+ - ∆Rσ-)/R0, where R0 is probe reflectance without the pump. Such valley-polarized electron gives rise to a helicity-dependent absorption difference (i.e., CD). Then, the transient CD response must scale to the amount of the residual valley-polarized electrons in MoS2 MLs. Figure 4d displays such time-resolved CD responses from a series of (MoS2/WS2)n SLs at 77 K, when the pump and probe photon energy were set to A-exciton resonance of MoS2 MLs with the pump fluence of 5.7 μJ/cm2. We observed the linearly proportional CD signals with increasing n in (MoS2/WS2)n SLs, as in Fig. 4d.