Large-area synthesis of high-quality monolayer 1T'-WTe₂ flakes

In the surge of research into the properties of two-dimensional materials 'beyond graphene', transition metal dichalcogenides (TMDs) have emerged as a platform for novel scientific discoveries and exploratory device concepts [1–5]. TMDs, also known as MX₂ materials, in the monolayer form consist of a layer of metal atoms, such as molybdenum or tungsten, sandwiched between chalcogenide atoms, such as sulfur or selenium. The optical [6], electrical [7], mechanical [8] and tribological [9] properties of these MX₂ materials have been widely studied in recent years, with numerous proposed applications [10] including energy storage [11] and chemical sensors [12]. The majority of work to date has relied on mechanical micro-exfoliation to isolate flakes of high quality monolayer material but the yield for this method is typically very low, which makes comprehensive studies of monolayer MX₂ films difficult and tedious. Therefore there is an urgent need for the development of robust and reproducible methods to produce high quality, large-area monolayer MX₂ materials.

Monolayer MX₂ films can be found in different structural phases [13], which differ in the arrangement of the top layer of chalcogenide atoms with respect to the bottom layer (figure 1(a)). MX₂ materials frequently grow in the highly symmetric 1H phase where the top layer of chalcogenide atoms is aligned with the bottom layer. 1H-MoS₂, the most commonly studied 1H material to this point, is a semiconductor which can be incorporated into field effect transistor (FET) devices with high on/off ratio, reasonable values of carrier mobility, and a band gap that is tuned by the number of atomic layers [14, 15]. In the 1T phase, the top layer of chalcogenide atoms is shifted with respect to the bottom layer so that as viewed normal to the plane, the chalcogenide atoms form a hexagon around the metal atom. The 1T phase of MoS₂ has been accessed experimentally through chemical transformation induced by n-butylithium [16–18]. While 1H-MoS₂ is a semiconductor, 1T-MoS₂ is metallic, so lateral heterostructures of monolayer 1H/1T MoS₂ are attractive candidates for developing low contact resistance devices [17]. The 1T' phase is a distorted 1T phase where the chalcogenide atoms have distanced themselves from their hexagonal position around the metal atom. The 1T' phase can be easily recognized and differentiated from the 1T phase by the presence of isolated zigzag chains of chalcogenide atoms that run vertically in figure 1(a).

Zoom In Zoom Out Reset image size

The 1T' phase of TMDs has recently attracted great attention due to the prediction that they support topological electronic states [19]. For most of the MX₂ materials, it has been predicted that the 1H phase is the energetic ground state, and that transitions from the 1H to the 1T' phase can be induced by strain [20]. One exception is WTe₂, where the 1T' phase is predicted to be the ground state. Bulk crystals of 1T'-MoTe₂ and 1T'-WTe₂ have been grown, and exfoliation has been used to produce few-layer films, but monolayer films are rarely reported using this approach [21–23]. There are reports of a large non-saturating magnetoresistance [24], the appearance of weak antilocalization [23], and pressure induced superconductivity [25] in few-layer 1T'-WTe₂, which is relatively stable under ambient conditions for films of six or more layers. However, since it is difficult to obtain monolayer flakes with reproducible properties by exfoliation, and because both 1T'-MoTe₂ and 1T'-WTe₂ are highly unstable under ambient conditions, there are to date very few studies of the properties of monolayer flakes. Recently we reported growth of monolayer single-crystal flakes of 1T'-MoTe₂ by chemical vapour deposition (CVD) with good surface coverage, which enabled an extensive study of 1T'-MoTe₂ in the monolayer form [26]. This was followed with a study on the absorption dichroism in monolayer 1T'-MoTe₂ [27]. Recently a group has shown that growth of 1T'-WTe₂ in monolayer form is possible by CVD [28] but there remains a need for careful investigation of the properties of monolayer flakes as well as a deeper understanding of the process of degradation under ambient conditions.

Here we present a facile, reproducible approach for growth of large, monolayer single crystal 1T'-WTe₂ flakes by CVD with a surface coverage of ~20%. Monolayer 1T'-WTe₂ flakes degrade within minutes under ambient conditions, but we found that by minimizing the handling time and passivating the samples with CVD-grown graphene, we were able to characterize monolayer flakes of 1T'-WTe₂. The chemical composition of the films was determined using x-ray photoelectron spectroscopy (XPS), and atomic force microscopy (AFM) was used to confirm that the flakes were of monolayer height. Raman mapping was used to study the vibrational modes of monolayer and bilayer 1T'-WTe₂, with the results being in excellent agreement with theoretically predicted phonon frequencies. Aberration corrected scanning transmission electron microscopy (ACSTEM) was used to determine that the WTe₂ films were in the 1T' structural phase. Because the growth method produces a high density of monolayer 1T'-WTe₂ flakes compared to the exfoliation method, we were able to study the degradation of the material in air. We explored degradation mechanisms by using density functional theory calculations and concluded that the films oxidize quickly in ambient. We obtained the first electrical results on monolayer 1T'-WTe₂, finding a weak antilocalization effect at low temperature, which is consistent with the presence of strong spin–orbit coupling. This work creates a pathway towards the use of 1T'-WTe₂ for integrated device structures whose operation relies on control of topological states and phases.

The CVD process for monolayer 1T'-WTe₂ flakes relies on deposition of W-feedstock material and a growth promoter [26] onto the growth substrate, followed by exposure to tellurium metal vapour at a temperature of 650 °C (a furnace schematic is shown in figure 1(b)). We developed two different methods for achieving monolayer 1T'-WTe₂ flakes on a Si/SiO₂ substrate, which differ only in the method used to apply the W source material (figures 1(c) and (d)). The first approach used a 5 nm thermal evaporation of WO₃ (figure 1(c)), while in the second a large droplet of ammonia metatungstate was spread over a large fraction of the substrate (figure 1(d)). The second process was the most efficient, enabling dense growth of large (~50 µm) monolayer single crystal and polycrystalline 1T'-WTe₂ flakes across a 2 cm  ×  3 cm Si/SiO₂ substrate (figure 1(d)). Additional details are provided in the Methods section. Monolayer flakes grown by both methods showed identical physical and chemical properties. In the higher magnification optical images of figure 1(d), the top image shows regions of larger polycrystalline flakes that grew close to the initial source material, while monolayer single crystal flakes are seen further away from the initial ammonia metatungstate source. This suggests that the growth mechanism is similar to those for other TMDs such as 1T'-MoTe₂ [26] and 1H-MoS₂ [29, 30] where the ratio of Mo atoms to Te/S atoms is suitable for growth of monolayer single crystal flakes further from the Mo source.

XPS and energy dispersive x-ray spectroscopy (EDS) were used to determine the elemental and bond composition of the 1T'-WTe₂ flakes. Peaks in the x-ray photoelectron spectrum were observed at 573.0, 583.4, 243.7, 256.3 eV, corresponding to the Te 3d_5/2, Te 3d_3/2, W 4d_5/2, and W 4d_3/2 signals, respectively (figure 2(a)). These results are comparable to previous reports of XPS on exfoliated flakes of 1T'-WTe₂ [22]. On the basis of these data, the Te/W atomic ratio was found to be 2.04, very close to the expected value for 1T'-WTe₂. The full XPS spectrum and details as to how the stoichiometry was determined can be found in the supporting information and figure S1 (stacks.iop.org/TDM/4/021008/mmedia). The EDS results can be found in figure S2; they indicate a stoichiometry of 1:2, in good agreement with the XPS analysis. Atomic force microscopy (AFM) was performed on the flakes, revealing a height of roughly 1 nm (figure 2(b)) as expected for monolayers, and in agreement with previous reports on exfoliated 1T'-WTe₂ flakes [21, 31]. Flakes grown through this method are predominantly monolayer thickness, but occasionally small bilayer regions are found on the edges of the flakes as seen in figures 2(b) and (c).

Zoom In Zoom Out Reset image size

Raman mapping was performed on a 1T'-WTe₂ flake that was predominantly monolayer but with small bilayer regions on the edges (figure 2(c)) in order to study the effect of this thickness change on the phonon modes. The flake was encapsulated with large-area graphene to avoid degradation during the measurements. Two main Raman peaks were observed: the in-plane $A_{1}^{7}$ peak, and the $A_{1}^{9}$ peak, which contains both in-plane and out-of-plane vibrations [21]. For the monolayer region, the $A_{1}^{7}$ and $A_{1}^{9}$ peaks were observed at 163.2 cm⁻¹ and 214.7 cm⁻¹, respectively. For the bilayer region the $A_{1}^{7}$ peak was at 162.7 cm⁻¹, while the $A_{1}^{9}$ was shifted to 213.2 cm⁻¹ (figure 2(c)). All of these peak positions are red-shifted by ~1 cm⁻¹ compared to previously reported values [21, 23, 31, 32], but the separation between the $A_{1}^{7}$ and $A_{1}^{9}$ peaks is in good agreement with these earlier works. To explain this observation, we used density functional theory to compute the frequencies of the Raman-active phonon modes before and after introduction of a graphene capping layer. For uncapped monolayer 1T'-WTe₂ we calculate the position of the $A_{1}^{7}$ and $A_{1}^{9}$ peaks at 161.54 cm⁻¹ and 211.36 cm⁻¹ respectively. With the addition of graphene, the peaks show a small red shift to 159.95 cm⁻¹ and 208.35 cm⁻¹. These calculated values are in good agreement with the measurements, and the calculation also explains the slight red shift in our data compared to earlier reports, which we attribute to the presence of the graphene capping layer. Using the same approach, we also studied the evolution of the Raman-active phonon modes as a function of layer number. We find that the $A_{1}^{9}$ should redshift from 211.36 cm⁻¹ for monolayer material to 209.53 cm⁻¹ for bulk, in agreement with our measurements and previous DFT studies [21, 31].

Monolayer and few-layer 1T'-WTe₂ flakes are unstable under atmosphere as well as transmission electron microscopy (TEM) irradiation, so the 1T'-WTe₂ flakes were encapsulated between graphene monolayers (figure 3(a)) to reduce damage during transfer and TEM imaging [33, 34]. (Details regarding the fabrication of the heterostructure stack can be found in the method section.) The intensity in high-angle annular dark field (HAADF) images is proportional to the square of the atomic number (Z) so the two graphene (Z_C  =  6) layers surrounding the 1T'-WTe₂ monolayer (Z_W  =  74, Z_Te  =  52) are effectively invisible, making this graphene/MX₂/graphene structure ideal for further studies on unstable materials. To determine the lattice parameters and image the atomic structure of monolayer 1T'-WTe₂, selected-area electron diffraction (SAED) images and HAADF images were obtained using a JEOL 200CF (figure 3(c)) equipped with a CEOS corrector for the ACSTEM probe. The ACSTEM was operated at 60 kV to reduce beam-induced displacements of atoms from the lattice, although damage was observed at high magnifications after as little as one exposure. The symmetry of the graphene lattice is different from that of 1T'-WTe₂, so the two diffraction patterns are readily distinguished. Figure 3(b) is the SAED image of the graphene/1T'-WTe₂/graphene heterostructure, where the graphene diffraction spots are highlighted in blue and the 1T'-WTe₂ spots in pink. The SAED pattern for 1T'-WTe₂ matches the simulated SAED pattern in figure S3. Calibrating the SAED pattern using the graphene diffraction spots (${{\bar{a}}_{\left[1\,0\,\bar{1}\,0\right]}}=\frac{\sqrt{3}\times 2.46\,{\mathring{\rm A}}}{2}=4.692$ nm⁻¹) [35], we measure the lattice constants for 1T'-WTe₂ to be a  =  3.49 Å and b  =  6.32 Å, in excellent agreement with previous reports [36]. The single crystal domain size is measured to be on the order of 10–20 µm². Figure 3(c) is an HAADF image of the monolayer 1T'-WTe₂ that shows excellent agreement with the simulated HAADF image in figure 3(d).

Zoom In Zoom Out Reset image size

Bi-layer 1T'-WTe₂ is reported to be unstable under ambient conditions, as evidenced by a decay in the Raman signal intensity over the course of several hours [21]. If bi-layer samples decay in hours, one might expect that monolayer samples decay much more quickly. To date, however, careful study of the stability of monolayer 1T'-WTe₂ flakes has been problematic due to their rapid degradation and the low yield of the exfoliation method. Due to the relatively high coverage of monolayer 1T'-WTe₂ flakes that is characteristic of our CVD growth process, we were able to observe the decaying of the monolayer flake through multiple experimental approaches. We first studied degradation of the material through measurements of the image contrast in optical micrographs, defined as $\left[{{I}_{\text{FL}}}-{{I}_{\text{SUB}}}\right]/{{I}_{\text{SUB}}}$, where ${{I}_{\text{FL}}}$ and ${{I}_{\text{SUB}}}$ are the image intensity of the flake and the background oxidized silicon substrate, respectively (figure 4(a)). A sample of monolayer flakes was removed from the growth furnace and immediately transported to an optical microscope where it was left under ambient conditions for 3d (figure 4(a)). The optical contrast of the flake against the substrate was observed to decrease rapidly over time, suggesting that the flake underwent a chemical reaction with room air. After 30 min the optical contrast saturated at a low level, implying that the chemical reaction terminated.

Zoom In Zoom Out Reset image size

We studied the amplitude of the $A_{1}^{7}$ and $A_{1}^{9}$ peaks in the Raman spectrum of uncapped monolayer 1T'-WTe₂ flakes left exposed to air (figure 4(b)). We first performed a scan on an as-prepared monolayer 1T'-WTe₂ flake, and each scan at later times was conducted on a new, unmeasured flake in order to avoid the confounding influence of damage to the material caused by exposure to the Raman laser. Through this approach, we were able to study how exposure to room air affects the $A_{1}^{7}$ and $A_{1}^{9}$ Raman peaks of monolayer 1T'-WTe₂. The peak amplitudes decreased in a nearly identical manner, with a strong (~50%) reduction within a few minutes, and the peaks essentially disappeared after 30 min, in rough agreement with the optical contrast data of figure 4(a). To our knowledge this is first reported study of the decay of monolayer 1T'-WTe₂ flakes under ambient using Raman spectroscopy. The results clearly indicate that monolayer 1T'-WTe₂ flakes decay within tens of minutes when left unprotected under ambient conditions.

We discovered that degradation of monolayer 1T'-WTe₂ could be avoided by transferring a sheet of graphene onto the substrate immediately after growth. The graphene-capped samples were stable for days, which enabled Raman mapping measurements, as illustrated in figure 2(c). Figure 4(c) is an optical image taken 12 h after a sample of monolayer 1T'-WTe₂ flakes was partially capped with graphene. The top zoomed-in image shows that for monolayer flakes under graphene, the optical contrast against the oxidized silicon substrate remained very strong and was comparable to that observed for flakes immediately after growth. The bottom image is a portion of the sample where uncapped monolayer flakes were exposed to air for several hours. These flakes showed very low optical contrast, which is typical of degraded material, in agreement with figure 4(a). The middle image is of the region near the boundary of the graphene sheet. Some flakes in this region were half exposed to air, and the optical contrast of exposed regions was very different from that where the material was covered by the graphene overlayer. This image highlights the degradation of 1T'-WTe₂ flakes in air and the utility of graphene as a passivation layer. Based on these measurements, we hypothesized that the films were damaged by reacting with a component molecule of the air.

We explored possible degradation mechanisms by using density functional theory calculations to study the adsorption of H₂(g), O₂(g), and H₂O(g) on monolayer 1T'-WTe₂ as the first is present during synthesis and the latter two are ubiquitous. Figure 5(a) shows two perspectives of the 1T'-WTe₂ (0 0 1) surface. The thickness of the monolayer under vacuum is 4.22 Å. The Te atomic layer lies on top of the layer of W atoms, which can have one of two geometries: small and large W₃ triangles. In an oxygenated environment, 1T'-WTe₂ favors the dissociative adsorption of O₂(g), with the equilibrium structure at 0 K shown in figure 5(b). The adsorption process in the figure is highly exothermic (ΔE_ads  =  −3.60 eV/O₂) and the adsorption location shown is greatly favored over other locations. Therefore surface oxidation by O₂(g) is expected under experimental conditions of ambient O₂(g) pressure and temperature. When O₂(g) dissociates, one O atom binds directly atop a W atom in the sublayer. The other O atom displaces Te from a small W₃ site, thus forming a W₃O subunit. This site is favored over the large W₃ triangle for dissociative O₂ adsorption because it provides three-fold coordination for one of the O atoms. The displaced Te atom rises slightly off the surface (figure 5(c)), increasing the thickness of the monolayer from 4.22 Å to 5.12 Å. Therefore, this calculation suggests that 1T'-WTe₂ is highly susceptible to oxidation, which leads to structural degradation via displacement of Te. We find that oxygen chemisorption transfers intensity in the Raman T spectrum from the $A_{1}^{7}$ and $A_{1}^{9}$ peaks to new modes in the Raman spectrum associated with the vibration of W–O bonds. We calculated Raman frequencies of 322.82 cm⁻¹, 389.19 cm⁻¹ for in-plane oxygen vibrations, and a mode at 500.03 cm⁻¹ with predominantly oxygen out-of-plane vibrations.

Zoom In Zoom Out Reset image size

We also investigated the possibility that H₂(g), present during synthesis, could passivate the surface. We find, however, that H₂(g) prefers physisorption over dissociative chemisorption (figure S4(a)). This process is slightly exothermic with ΔE_ads  =  −0.07 eV/H₂, and the thickness of the monolayer is unaffected by H₂(g) adsorption. The surface greatly prefers oxidation to hydrogenation.

Finally, we considered the competition between O₂(g) and H₂O(g) as surface oxidizing agents (see supporting information). Unlike O₂(g), H₂O(g) prefers associative adsorption at Te sites. Figure S4(b) shows that H₂O(g) weakly interacts with the surface by forming van der Waals bonds between H and Te, with no effect on the thickness of the monolayer. Adsorption is spontaneous (ΔE_ads  =  −0.27 eV/H₂O) but the energetics are far less favorable than for dissociative O₂(g) adsorption. Therefore, O₂(g) is a much more potent surface oxidizer than H₂O(g). Based on these considerations, we conclude that 1T'-WTe₂ monolayer films degrade under ambient through oxidation, leading to the changes in optical contrast and the Raman spectrum that are discussed above.

First-principles calculations were performed to confirm the semimetallic nature of monolayer 1T'-WTe₂. As shown in figure S5, the Fermi level overlaps with the conduction and valance bands, forming electron and hole pockets near the Γ point. Nonetheless, since the bulk energy gap is well developed throughout the entire Brillouin zone, separating the conduction and valance bands, the time-reversal invariant Z₂ topological phase [38] can be evaluated using the all the states from 1st to nth bands, where n is the electron filling. We find the Z₂ topological invariant is nontrivial, as reported in a previous study [19]. The semimetallic nature of WTe₂ obstructs disentanglement of the helical edge states of the nontrivial topological phase and the bulk metallic states. Nonetheless, the quantum spin Hall effect in WTe₂ could potentially be observed by applying an external electric field or strain, which would induce a band gap and isolate the surface edges states from the bulk states [19].

Devices for transport measurements on single monolayer flakes were fabricated by electron beam lithography with minimal material degradation due to air exposure. Details of the fabrication process are provided in the Methods section. Care was taken to define the contact pattern so that the electrical leads to a particular monolayer flake would not be short-circuited by other 1T'-WTe₂ flakes on the substrate (figure 6(a)). The monolayer 1T'-WTe₂ material was kept passivated by a PMMA layer, with only brief exposure to air. After fabrication, we carried out low temperature magnetoresistance measurements under high vacuum. Once the electrical measurements were completed, Raman spectroscopy was performed on the flake to confirm its monolayer thickness.

Zoom In Zoom Out Reset image size

We monitored the sample resistance in a 4-probe measurement configuration as the sample was slowly warmed from 2 K to 300 K (figure 6(b)). The insert of the figure shows the reduced activation energy $W=-\text{d}\left(\ln R\right)/\text{d}\left(\ln T\right)~$ as a function of temperature revealing a positive slope at low temperature, indicating metallic behavior [39]. To our knowledge, these are the first measurements of monolayer 1T'-WTe₂ showing this property. The sample magnetoresistance showed a cusp-like weak antilocalization (WAL) feature that became more pronounced and narrower at low temperature. The WAL effect can be associated with either bulk or topological states and its appearance here indicates that spin–orbit coupling is sufficiently strong that the spin orbit scattering length is much less than the phase scattering length [40–43]. WAL has been observed in multilayer 1T'-WTe₂ [23] but until now has not been reported in monolayer devices, presumably because of their air sensitivity. The measured sample magnetoconductance was fitted with the Hikami–Larkin–Nagaoka (HLN) equation in the limit of strong spin orbit coupling, $\sigma (B)-\sigma (0)=-\alpha \frac{{{e}^{2}}}{\pi h}\left[\psi \left(\frac{1}{2}+\frac{{{B}_{\phi}}}{B}\right)-\ln \left(\frac{{{B}_{\phi}}}{B}\right)\right]\,$ [44], where ψ is the digamma function, and ${{B}_{\phi}}=\frac{\hslash}{4el_{\phi}^{2}}$ with l_ϕ the phase coherence length (figure 6(d)). Agreement for all temperatures was excellent with the inferred phase coherence length ranging from 40 nm at 2 K to 22 nm at 15 K. The parameter α indicates the number of conduction channels, and is of order unity in topological insulators [23, 40, 41, 45–47]. In our measurements we found nearly 100% magnetoresistance at 2 K but HLN analysis yielded $\alpha \ll 1$ due to the high channel resistance (far exceeding the resistance quantum), indicating diffusive transport and no contribution from ballistic channels. We note, however, that we assumed conduction in the flake was uniform across its area when converting conductance into conductivity, and that the value of α would be systematically underrepresented if conduction actually occurred through a path significantly less wide than the flake width. Through these electrical results we conclude that monolayer 1T'-WTe₂ possesses a strong spin orbit coupling effect, a first step toward realizing topological electronic states.

We have demonstrated a facile and reproducible growth method for achieving monolayer 1T'-WTe₂ flakes. Samples were carefully characterized using XPS, Raman spectroscopy, AFM and ACSTEM to confirm their atomic composition and structural configuration. We demonstrated the use of a graphene overlayer to passivate the material and avoid degradation under ambient. Degradation occurs rapidly (minutes) and is most likely driven by a reaction with oxygen gas, as we demonstrated computationally. Finally we presented the first electrical data on monolayer 1T'-WTe₂. A WAL effect appears at low temperature and is well fit by the HLN equation in the limit of strong spin orbit coupling. The approach presented here provides a pathway to further study of topological electronic states in this two-dimensional material.

7.1. Methods

This work was supported by NSF MRSEC DMR-1120901 and NSF EFRI 2-DARE 1542879. The authors acknowledge use of the Raman mapping system supported by NSF Major Research Instrumentation Grant DMR-0923245. WMP and MD acknowledge funding from NIH grant R21HG007856 and from NSF EFRI 2-DARE (EFRI-1542707). RWC and FS acknowledge support from NSF grant CMMI-1334241. HK and YWP acknowledge supports from the NRF-SSF 2014R1A2A1A12067266, the GRDC 2015K1A4A3047345, and the FPRD of BK21 from the NRF, Korea. The authors gratefully acknowledge use of the HR-TEM in the Krishna Singh Center for Nanotechnology at the University of Pennsylvania and the use of the AC-TEM facility at Lehigh University. ZL appreciates support from the Research Grant Council of Hong Kong SAR (Project number 16204815), Center for 1D/2D Quantum Materials and the Innovation and Technology Commission (ITC-CNERC14SC01). RBW, LZT, YK, and AMR acknowledge support from the US DOE through grant no. DE-FG02-07ER15920. Computational support was provided by HPCMO of US DoD and NERSC.

ATCJ directed the research. CHN proposed and designed the experiment, and carried out 1T'-WTe₂ growth and measurements of the degradation of the Raman spectrum over time. FS performed the XPS measurement (under the supervision of RWC). YRZ and CEK performed the optical measurements over time. Electrical measurements were performed by MN and JMK. Graphene growth for passivation was conducted by ZG, supervised by ZL Devices for electrical measurements were prepared by CHN and HK (with supervision from YWP). CHN, WMP, and ZG performed AFM and Raman characterization; WMP and CHN performed the TEM experiments (under the supervision of M.D.); and WMP performed the SAED and QSTEM image simulations. RBW computed the energetics of degradation, while LZT performed the Raman simulations and YK calculated the electronic band structure (under the supervision of AMR). CHN and ATCJ wrote the manuscript, with input and approval from all the authors.