2D TMDs possess various polymorphic structures including 2H (trigonal prismatic), 1T (octahedral), 1T′ and Td phases, which are determined by the different chalcogen coordination geometries of metal atoms and stacking orders of atomic layers1. The polymorphic phases give rise to a variety of intriguing properties such as superconductivity2,3, ferroelectricity4, ferromagnetism5,6, Weyl semimetal7 and quantum spin Hall effect8. The physical and chemical properties of TMDs can be modulated by changing the structural phases. The so-called phase engineering of TMDs makes it possible to achieve the precise control of structural phases in a desired manner9-11. The phase engineering can be implemented through two main strategies of phase transformation and phase-selective synthesis. The phase transformation strategies including alkali metal intercalation12,13, alloying14, thermal treatment15,16, electric field17, mechanical strain18,19, and external irradiation20,21 have been employed to trigger the structural transformation to effectively obtain the target phases. The aforementioned chemical and physical methods might encounter the issue of impurities and defects introduced during the phase transformation process as well as the metastable phases recovering to their original structures when the external stimuli are removed22. In comparison, phase-selective synthesis has the ability to directly fabricate 2D TMDs with the specific phase and high phase purity. The strategies of precursor design, atmosphere regulation and temperature control have been utilized in chemical vapor deposition (CVD) growth to achieve the phase-selective synthesis23-26. As for the MBE growth, it is highly desirous to develop general and controllable strategies to fabricate 2D TMDs with excellent phase selectivity.
Heteroepitaxial templates have been employed to seed the growth of crystals with unconventional phases27-30. The heteroepitaxial template usually provides a constrained environment to guide the crystal growth with the formation of lattice-matched interfaces. Under the effect of heteroepitaxial templates, a stabilized nonequilibrium phase can be formed within a critical coherent thickness, for instance, the metastable cubic phase of aluminium nitride (AlN) was stabilized in AlN/TiN superlattice films27,31. The formation of a lattice-matched interface with a lower interfacial energy is the essential prerequisite for the stable existence of metastable phases. The heteroepitaxial templates play a prominent role in phase control of noble metal nanocrystals, for example, the unconventional 4H-phase Ag, Pd, Pt, Ir, Rh, Os, Ru and Cu were synthesized using 4H-Au nanoribbons as the epitaxial templates30,32. In two-dimensional TMD materials, due to the existence of atomically flat dangling-bond-free surfaces, in-plane heteroepitaxial template effects are considered to seed the growth of 2D TMDs for achieving the selective growth of metastable phase structures in MBE growth.
Selective growth of H-phase CrSe2 with the formation of lateral heterostructures
Atomically thin CrSe2 films are usually observed in the metallic 1T phase which is demonstrated by atomic-resolution transmission electron microscopy and other experiments6,33-35. As shown in the schematic illustration (Fig. 1a), the codeposition of Cr and Se atoms on a substrate of highly oriented pyrolytic graphite (HOPG) will produce the metallic T-phase CrSe2 islands. The T-phase CrSe2 has an octahedral coordination structure with D3d symmetry and degenerate and orbitals. The partially filled d orbitals give rise to the metallic electronic properties as illustrated in Fig. 1b. By utilizing the in-plane heteroepitaxial template of H-phase MoSe2 nanoribbons, the growth of H-phase CrSe2 monolayers can be guided with minimizing the interfacial energy (Fig. 1a). The atomically matched lattices guarantee the formation of lateral heterostructures with seamlessly connected interfaces. Unlike metallic T-phase CrSe2, the H-phase CrSe2 adopts the trigonal prismatic coordination structure with D3h symmetry, which splits the d orbitals into three groups , and with a sizeable bandgap (Fig. 1c).
In the STM image of the MBE-grown CrSe2 samples (Fig. 1d), the T-phase CrSe2 islands exhibit approximately hexagonal shapes with straight edges. A nonlayered growth behavior is unraveled by the existence of CrSe thin layer with a height of ~0.3 nm shown in Fig. 1d. The lattice constant is calibrated to be 3.4 ± 0.1 Å which is consistent with that of previously reported 1T-CrSe2 crystals (3.39 Å) synthesized through deintercalation of lithium from LiCrSe235. The metallic characteristic of the T-phase CrSe2 monolayer is unveiled in the differential conductance (dI/dV) spectrum (Fig. 1g). In order to achieve the phase-selective growth of 1H CrSe2, the in-plane epitaxial template of MoSe2 nanoribbons are prepared beforehand with the typical width of 10 − 50 nm at a growth temperature of about 550 °C (Fig. 1e). The distinctive Mo- and Se-edges at the opposite sides of nanoribbons can be distinguished by different edge features in the atom-resolved STM images (Supplementary Fig. 1). After the growth of MoSe2 nanoribbons, Cr and Se atoms are subsequently codeposited onto the same HOPG substrate. The diffusing atoms preferentially nucleate and aggregate at the active edges of MoSe2 nanoribbons. The H-phase CrSe2 rather than T-phase structure is formed with the lattice matching under the effect of the in-plane epitaxial template of MoSe2 nanoribbons. The as-grown H-phase CrSe2 segments are seamlessly fused to MoSe2 nanoribbons with the formation of lattice-matched lateral heterostructures (Fig. 1f). In the lateral heterostructures, the continuous H-phase interface structures have a lower interfacial energy compared with the 1H-1T interface structures as revealed by density functional theory (DFT) calculations (Supplementary Fig. 2). The chemical states of Cr and Se elements in two distinct phases of CrSe2 were investigated by X-ray photoelectron spectroscopy. The measured binding energies of Cr 2p and Se 3d electrons (Fig. 1h) show an energy shift of ~0.5 eV between H- and T-phase CrSe2 due to the different electronic structures. The electronic properties of H-phase CrSe2 are revealed by STS measurements which exhibit a semiconducting character with bandgap of 0.75 ± 0.05 eV. The increased peak intensity of local electronic states with the invariant bandgap can be observed in the dI/dV spectra taken at different tunneling currents (Fig. 1i). The semiconducting property of H-phase CrSe2 monolayer is also verified by DFT calculations. The electronic structure of H-phase CrSe2 manifests a direct bandgap of 0.72 eV with the valence band maximum and conduction band minimum derived mostly from the Cr 3d orbitals (Supplementary Fig. 2).
Structural characterization of MoSe2-CrSe2 interfaces
The atomically sharp interfaces of MoSe2-CrSe2 lateral heterostructures are revealed both by STM and nc-AFM. In the STM image of the MoSe2-CrSe2 lateral hetetrostructure (Fig. 2a), CrSe2 segments are seamlessly connected to the MoSe2 nanoribbons in between. The different STM image contrast between CrSe2 and MoSe2 primarily originates from the local electronic states rather than topographic features, which varies with the applied bias voltages (Supplementary Fig. 3). There is a continuous linear defect of mirror twin boundary (MTB) crossing through the heterostructure interface, reflecting the in-plane heteroepitaxial template effect. The occurrence of MTB characteristic defects in CrSe2 provides strong evidence for the claim of H-phase CrSe2. The atomic structures and electronic properties of MTBs in CrSe2 will be further elaborated in more detail later. The corresponding ball-and-stick model of the lateral heterostructure with a MTB line defect is exhibited in Fig. 2b. In the atomic row of MTB, four-fold rings shares a point at chalcogen site leading to each Se atom bound to four Cr atoms instead of three. The Mo- or Se-edge of the MoSe2 nanoribbon can be determined by the orientation of MTB due to the constraint of mirror symmetry. To detect the strain at the heterostructure interfaces, the moiré pattern in the MoSe2 region was studied since it can be used as a ‘magnifying glass’ to directly visualize the lattice-misfit strain at the heterostructure interfaces36. In the high-resolution STM image (Fig. 2c), the MoSe2 region exhibits an undistorted 3 × 3 moiré pattern, indicating that there is no obvious lattice-misfit strain at the heterostructure interface. The nc-AFM technique was also employed to characterize the interface structures as it can provide the surface topography without being disturbed by electronic states. The constant-height nc-AFM frequency shift images (Fig. 2d and Fig. 2e) taken at Se- and Mo-edges of MoSe2 nanoribbons respectively (labelled in Fig. 2a) both exhibit atomically sharp interface structures without any dislocations. The nc-AFM observations also revealed the atomically sharp MoSe2-CrSe2 interface structures at the adjacent Mo-edge and Se-edge with the included angle of 120° (Supplementary Fig. 4). This results imply that CrSe2 in the lateral heterostructure adopts the same H-phase structure as MoSe2. If we postulate that it were T-phase CrSe2 connected to the different edges of MoSe2 nanoribbons, dislocations and lattice misfit strain would emerge at the interfaces due to the different lattice symmetries and lattice constants21, however, they are not observed in our experiments. In the nc-AFM images, the MoSe2 regions have a brighter image contrast (higher frequency shift) compared with CrSe2 regions, which can be ascribed to the different Pauli repulsive forces at small tip-sample distances. We adopted the model of MoSe2-CrSe2 lateral heterostructure with continuous H-phase structures at the interfaces to simulate the STM and nc-AFM images. The observed STM and nc-AFM image contrast between the MoSe2 and CrSe2 regions are reproduced well by the DFT simulation (Fig. 2f).
Band alignments in the lateral and vertical heterostructures
STS measurements were carried out to explore the spatial evolution of electronic structures of both lateral and vertical heterostructures at the atomic scale, which are of fundamental importance for understanding the interface properties and evaluating their application potentials. The close-up STM image (Fig. 3a) displays the MoSe2-CrSe2 lateral heterostructures with 12 nm-width MoSe2 nanoribbon in between. The type-I band alignment with a straddling gap is found for the MoSe2-CrSe2 lateral heterostructures, which endows the lateral heterostructures with the potential for light-emitting applications and studying light-matter interaction. The band profile across the interface is directly visualized by the colour rendering of the dI/dV mapping (Fig. 3b). A bandgap of ~2.30 eV with Fermi level located 1.60 eV above the valence band maximum (n-type) is detected in the MoSe2 region away from the interface. The MoSe2 region exhibits upward band bending for both valence and conduction band near the interface due to the charge transfer from MoSe2 to CrSe2. The band profile at the interface shows a sharp transition into the CrSe2 region which has a smaller bandgap of ~0.75 eV. A slight downward band bending can be seen in the magnified band profile of CrSe2 region (Fig. 3c), which is attributed to electron injection from MoSe2. The valence and conduction band offset (VBO and CBO) in the MoSe2-CrSe2 lateral heterostructures are determined to be ~0.84 eV and ~0.78 eV, respectively. Besides the band bending, the effect of bandgap narrowing emerges at the heterostructure interface probably due to strain at the interface. As illustrated in Fig. 3d, the downward shift of the conduction band is larger than that of valence band, resulting in a bandgap narrowing of ~0.06 eV for CrSe2. The inconsistent upward band shifts in valence and conduction bands of MoSe2 also lead to a bandgap narrowing of ~0.10 eV.
Owing to layer-by-layer growth mode, the CrSe2/MoSe2 heterostructures are formed with the first-layer MoSe2 and second-layer CrSe2 in the bilayer structures (Fig. 3e). This is representative of a CrSe2/MoSe2 vertical heterostructure. The spatially resolved dI/dV spectra are acquired along the path with a step size of 8.5 Å labelled in Fig. 3e. The band offsets and edge states at the interface can be intuitively observed in the energy profile of the CrSe2/MoSe2 vertical heterostructure (Fig. 3f). The valence and conduction bands of the first-layer MoSe2 exhibits a joint upward band bending at the interface with a magnitude of ~0.14 eV and ~0.07 eV, respectively. Compared with the small band bending in the previously reported bilayer-monolayer MoSe2 and WSe2 homostructures37, the band shifts in the first-layer MoSe2 indicate a stronger interlayer interaction in the vertical CrSe2/MoSe2 heterostructures. Regarding the second-layer CrSe2 region, the valence and conduction bands are bended upward by ~0.28 and ~0.36 eV, respectively (magnified in Supplementary Fig. 5). The edge state with a narrow gap (marked by orange lines in Fig. 3f) emerges at the edge termination of second-layer CrSe2 due to the existence of dangling bonds. The STS mapping is further sliced into selected individual spectra obtained at the vicinity of interface (Fig. 3g). In the dI/dV spectra of second-layer CrSe2, the original characteristic peak around −0.6 eV turns to be a sloping bump with energy position ~0.18 eV lower than that of monolayer CrSe2 on the semimetallic HOPG substrate. The increase of bandgap in the second-layer CrSe2 can be attributed to the better electronic isolation provided by the underlying MoSe2 interlayer. At the heterostructure interface, the localized dangling-bond states give rise to the new spectral features with smaller bandgaps in the dI/dV curves. The edge states and band shifts in the vertical heterostructure are further verified by our DFT-calculated local density of states (LDOS) plots of the Se atoms on the surface (Fig. 3h). The calculated differential charge density reveals the charge accumulation at the edge termination of second-layer CrSe2 which induces the upward band bending of CrSe2 near the edge.
Electronic properties of MTBs in the H-phase CrSe2 monolayers
MTBs as quasi-one-dimensional metals provide an ideal platform for exploring the electronic behavior in the confined system. MTBs with a 4|4P structure have been extensively studied in H-phase molybdenum dichalcogenides such as MoSe2 and MoTe238-41, but rarely detected in other non-Mo-based TMD materials. Isolated MTBs with triangular shape (Fig. 4a) are obtained in the H-phase CrSe2 monolayers. The close-up high-resolution STM image of MTB (Fig. 4b) exhibits a typical feature of two bright parallel lines as previously observed in MoSe2 and MoTe2, but without an obvious spatial modulation period. The structure model of the MTB is shown in Fig. 4c and the corresponding atomic structure is revealed by the nc-AFM image (Fig. 4d), in which a surface feature that a row of Se atoms have a darker contrast than other Se atoms in the region can be observed. The darker contrast mainly results from the higher density of Cr atoms in the MTB which gives rise to stronger attractive forces. The dI/dV spectra (Fig. 4e) taken on the MTB with a length of ~12.8 nm exhibit a feature of gap opening at the Fermi level. The gap size increases with decreasing the length of MTBs, which is one of the signatures of Tomonaga-Luttinger liquid (TLL) behavior (Supplementary Fig. 6). In the magnified dI/dV spectrum (inset in Fig. 4e), four pronounced peaks located at 37.6 mV, 15 mV, −12 mV and −34.2 mV can be identified. The constant-height dI/dV conductance mappings (Fig. 4g) measured at the four different peak energies reveal a periodic charge density modulation at the period of ~3a (triple lattice constant of CrSe2). The modulation period is also observed in the low-bias constant-height STM image (Fig. 4f), which reflects the spatial distribution of low-energy electronic states. As shown in Fig. 4g, the spatial modulation of highest occupied state (HOS) and lowest unoccupied state (LUS) is in phase with a symmetric two-lobe feature in the MTB, but it is partially out-of-phase for the second occupied and unoccupied states. The observed charge density modulation may not be a consequence of charge density waves (CDW) but the quantum confinement of electrons in the finite-length MTBs. In the dI/dV spectra along the MTB (Fig. 4h), the periodic fluctuation of peak positions at the negative bias from −13.2 mV to −5.7 mV and at the positive bias from 11.3 mV to 15.4 mV for the HOS and LUS respectively can be discerned. And the discrete quantized states below HOS and above LUS are directly visualized in the color plot of dI/dV spectra along the MTB (Fig. 4i). The phenomena may be described by TLL theory that low-energy excitations are standing waves with discrete energies in a finite-length MTB system42. Moreover, the TLL behavior of spin-charge separation is observed in the Fourier transform of the 2D STS map (Fig. 4j), in which two linear dispersion branches corresponding to the spin and charge density excitations have different slopes (marked with blue and yellow lines)40.