Nucleation, growth, and stability of WSe₂ thin films deposited on HOPG examined using in situ, real-time synchrotron x-ray radiation

The exfoliation and characterization of graphene have motivated research into other two-dimensional (2D) materials, notably transitional metal dichalcogenides (TMDs), due to their fascinating physical and chemical properties.^1–3 Monolayers of TMDs are particularly of interest because they exhibit a large direct bandgap, a property that is not intrinsic to graphene monolayers. These 2D materials have a stoichiometry of MX₂, where each layer consists of a plane of the transition metal (M) atoms (Mo, W, etc.), sandwiched between two planes of chalcogen (X) atoms (S, Se, Te, etc.).^4,5 Bonding within each layer is covalent, whereas the layers are bound to each other via much weaker van der Waals interactions. These properties of TMDs make them promising for use in flexible electronics and next-generation semiconducting devices.^6–8 These structural features may also act to distinguish the thin film growth of these materials when compared to higher symmetry semiconductors such as Si and GaAs.

The predominant method of producing thin films of TMDs for device demonstration is mechanical exfoliation.⁹ However, this top-down approach only produces small flakes and is not scalable. Recently, conventional vapor phase growth techniques have been explored as a way to overcome the shortcomings of exfoliation. Prior studies have investigated the growth of TMDs by vapor deposition using powders (e.g., WO₃, S, and Se) as sources,^10–13 atomic layer deposition (ALD),^14–18 as well as molecular beam epitaxy MBE.^19–24 However, the lack of uniformity, small grain sizes, and the limited control of film thickness are persistent challenges. Metal-organic chemical vapor deposition (MOCVD) is one traditional method that has shown considerable promise since the vapor-phase molecular reactants can be introduced to the surface of the substrate precisely and reproducibly. Large area, polycrystalline growth has been demonstrated using MOCVD;^25–30 unfortunately, it was achieved under reaction conditions that resulted in an extremely slow growth rate.²⁵ To date, studies of the early stages of growth have focused on controlling the nucleation density,^24,27,31 possibly by providing specific sites to initiate growth.^32,33 Other work has sought to build connections between the morphology of the TMD thin films and the method and conditions used for growth.^34,35 However, nearly all of these studies have lacked the capability of monitoring growth in real-time from the initial stages where the nuclei are formed to the eventual coalescence and formation of thicker continuous thin films. Although there continue to be numerous studies using MOCVD,³⁰ there have been relatively few studies employing the related technique of metal-organic molecular beam epitaxy (MOMBE),³⁶ which can in principle provide a much more precise delivery of the thin film precursors to the substrate surface.

As indicated above, the growth of thin films of TMDs by molecular beam epitaxy (MBE) using elemental sources has been demonstrated.^19–24 This method of deposition is quite useful since it occurs in ultrahigh vacuum (UHV), which allows for the use of in situ surface sensitive analytical techniques such as reflection high energy electron diffraction (RHEED). In principle, using MBE, one can deposit multiple layers of TMDs without exposure to air, which enables the growth of heterostructures involving different TMDs bound to each other by van der Waals interactions.³⁷ However, the low volatility of the elementary refractory metal sources makes the growth of TMDs by MBE challenging. This has led to the use of other approaches. As indicated above, the use of organometallic precursors as sources for Mo and W has achieved considerable success using MOCVD techniques.^25–38 Combining the chemistries used in MOCVD with the UHV environment of MBE could provide a potentially useful platform to study the growth of thin films of TMDs.

In this work, we report the growth of thin films of WSe₂ on highly ordered pyrolytic graphite (HOPG) substrates in UHV using tungsten hexacarbonyl [W(CO)₆] and elemental selenium as the sources for W and Se, respectively. W(CO)₆ has been used previously as a source of W concerning the growth of both WS₂^29,39 and WSe₂^25–28,40 via MOCVD and related techniques occurring at pressures much higher than UHV. Concerning the growth of other compound solid-state materials, such as III-V semiconductors, it is recognized that the ratio of the incident flux of the species that incorporate into the thin film can affect the stoichiometry, morphology, and quality of the thin film being grown.⁴¹ Here, since growth is conducted in UHV where transport of the depositing species is ballistic, we have exquisite control of the two incident fluxes of W and Se. In this work, we will examine the growth of thin films of WSe₂ under conditions where both species are incident on the substrate surface, and two extreme examples of the effect of the incident flux ratio: one where the flux of one species is completely extinguished for a period, and another where the sequence is reversed. We make use of in situ real-time x-ray techniques at the Cornell High Energy Synchrotron Source (CHESS) to determine thin film composition using x-ray fluorescence (XRF). We also make use of ex situ characterization techniques such as atomic force microscopy (AFM), x-ray diffraction (XRD), and x-ray photoelectron spectroscopy (XPS) to determine other thin film properties. We focus here on examining the mode of thin film nucleation, including the role of the substrate, as well as the stability of the deposited thin film in the absence of one of the depositing species.

The experiments were conducted in a custom-built UHV chamber fitted with Be windows, detailed in previous work,⁴² in the G3 hutch located at the Cornell High Energy Synchrotron Source (CHESS). This system has been employed in several in situ and real-time examinations of the nucleation and growth of crystalline thin film materials.^43–48 We exposed HOPG substrates to tungsten hexacarbonyl [W(CO)₆] via a supersonic molecular beam and elemental selenium [Se_n] via a thermal effusion source. A supersonic beam of W(CO)₆ was achieved by flowing a carrier gas, He, through a stainless-steel vessel containing W(CO)₆ powder (99.99%, Sigma Aldrich), and then, expanding the gas mixture into a UHV source chamber (base pressure ∼2 × 10⁻⁹ Torr) through a 150 μm diameter nozzle. The beam then passed through a skimmer, a differentially pumped antechamber, and finally through an aperture plate which defines the beam that strikes the substrate in the main chamber. The vessel containing the W(CO)₆ was heated to 45 °C and the He flow was 40 SCCM for all the experiments. The thermal effusive source (CreaTec Fischer & Co. GmbH) was comprised of a 10 cm³ pyrolytic BN crucible which contained the elemental Se (99.99%, Sigma Aldrich). This source was mounted directly onto our main chamber (45° angle of incidence to the substrate surface normal) and the temperature of the evaporator was held at 110 °C for the experiments (this produces a flux of Se that is dominated by molecular clusters).⁴⁹ Remotely controlled shutters allow for the independent gating of the incident flux of species from both the supersonic molecular beam and effusive sources. The substrates used in the experiment were 1 × 1 cm² grade 1 HOPG obtained from SPI supplies. Before insertion into the UHV chamber, the HOPG substrates were prepared by cleaving the topmost layer.⁵⁰ The experiments were performed at three different substrate temperatures, T_s = 400, 470, and 540 °C.

The growth of thin films was monitored in situ and in real-time using x-ray fluorescence (XRF). This technique can be used to measure directly the elemental composition of our thin films grown on the HOPG substrates, by making the appropriate corrections for the x-ray fluorescence experimental sensitivities as described in the supplementary material.⁷⁷ The energy of the incident x-ray beam was 13.45 keV, and measurements of the fluorescence during thin-film deposition were taken using a confocal XRF detector (CII Nano Technology USA Inc.). The sample area exposed to the x-ray beam is smaller than both the area exposed to the molecular beam and the HOPG samples themselves (see the supplementary material⁷⁷ for a description of this geometry). Following deposition, the samples were removed from UHV and were characterized ex situ with a Bruker Innova AFM (operated in tapping mode) and by SEM using a Tescan Mira FESEM at an acceleration voltage of 10 kV. Ex situ x-ray diffraction (XRD) was performed using a Rigaku SmartLab x-ray diffractometer. X-ray photoelectron spectroscopy (XPS) data were collected using an Omicron Sphera U5 concentric hemispherical electron energy analyzer, operated at constant pass energy of 50 eV. These data were analyzed using casaxps software (version 2.3.15). Non-monochromatic MgKα X-rays (1253.6 eV excitation energy) were produced using an Omicron DAR 400 twin anode source operated at 300 W (15 kV anode potential × 20 mA emission current).

We first consider the growth of thin films of WSe₂ using continuous exposure to both sources. In these experiments, the HOPG substrate was heated to the appropriate temperature and was exposed simultaneously to the supersonic molecular beam containing W(CO)₆ and the flux of Se_n emanating from the effusion cell. Fig. 1(a) presents a plot of the intensity from XRF of W and Se as a function of time. These intensities are “uncorrected” raw integrated intensities. Here, the fluxes of the two incident species were initiated at t = 10 s, and the substrate temperature was T_s = 470 °C. First, we observe there is a period where the accumulation of W and Se is essentially zero, and this period of incipient growth (or “nucleation delay”) lasts for ∼200–300 s. Deposition of both W and Se may be occurring during this time but at a much smaller rate—the intensity from XRF is simply too small to detect these events. After this initial period, we observe a strong increase in the XRF intensity for both W and Se, and the signal increase is eventually nearly linear. A more quantitative analysis (see the supplementary material,⁷⁷ of these two regimes, gives a nucleation delay of ∼249 ± 30 s for both W and Se. Using the results displayed in Fig. 1(a) and making the appropriate corrections for the x-ray fluorescence experimental sensitivities, we can compute the elemental ratio, Se/W, of the deposited thin film. These results are plotted in Fig. 1(b). After an initial period (up to ∼1000 s), where the statistical fluctuations are largely due to small signal strength, the ratio approaches a constant value of 1.78 ± 0.05. This indicates that the thin film is deficient in Se compared to the expectations for WSe₂.

Following growth, ex situ x-ray diffraction was performed on this sample as shown in Fig. 1(c), which is a plot of XRD intensity as a function of the out-of-plane scattering vector from 0–5 Å⁻¹. We observed diffraction features in the q-space that relate to both the HOPG substrate and the WSe₂ thin film. We saw the WSe₂ (002) and (006) diffraction peaks and also observed the HOPG (002) and (004) diffraction peaks, which overlap with the WSe₂ thin film (004) and (008) peaks, respectively. We utilized the full width at half maximum (FWHM) of the WSe₂ (002) peak to calculate the out-of-plane grain size of the thin film using the Debye–Scherrer equation and a crystal factor (K) = 0.9.⁵¹ The grain size of this thin film was ∼33.3 nm.

Next, we consider the growth of thin films of WSe₂ under an extreme case, where the flux of the W(CO)₆ is gated (with sub 1 s resolution). As in the case considered above in Fig. 1, neither species is incident on the substrate surface for the first 10 s. After the first 10 s, no W(CO)₆ is incident for 720 s, and then the substrate is subsequently exposed to W(CO)₆ for 720 s. This is repeated two more times for the incident flux of W(CO)₆. Whereas the flux of the Se species (Se_n) is initiated at t = 10 s and remains incident on the substrate surface until both sources are extinguished at 4330 s. Following the termination of the incident fluxes of both species, the XRF intensities for Se and W continued to be collected until t = 4800 s. Fig. 2(a) displays a plot of W and Se XRF intensities as a function of time. This experiment was conducted at a substrate temperature of T_s = 400 °C. We can make several observations from these data. First, not unexpectedly, we observe the constant intensity of the W XRF feature when there is no incident flux of W(CO)₆. Second, we also observe that the intensity of the Se XRF feature is also constant in the absence of an incident flux of W(CO)₆. Thus, for these conditions, growth only occurs when both species are incident, while the thin film is apparently stable in the absence of the incident flux of W(CO)₆—neither the decomposition of the thin film nor the deposition of additional Se is observed. Growth of multilayer Se is extremely unlikely in this case due to its high vapor pressure for these conditions (>1 Torr). As with the results we displayed above in Fig. 1, we can compute the elemental ratio, Se/W, using the XRF intensities in Fig. 2(a) after making the appropriate corrections. This result is plotted as a function of time in Fig. 2(b). After initial fluctuations, we observe that the ratio approaches a constant value of 1.77 ± 0.01.

One additional issue we can address regarding the results shown in Figs. 2(a) and 2(b) concerns the nucleation delay we observed in the data shown in Fig. 1(a). In analyzing the data shown in Fig. 2(a), it is apparent that there are at least three events to consider: the first introduction of the W species, and the second and third time this species is introduced. These differ of course since there is no pre-existing thin film of WSe₂ for the former, whereas a thin film is present for the latter two cases. Again, employing a more quantitative analysis (see the supplementary material⁷⁷), we estimate a nucleation delay for the first exposure to W(CO)₆ of ∼227 ± 16 s for W and ∼233 ± 8 s for Se, identical within experimental uncertainties. In contrast, for the second and third exposures to W(CO)₆, we find values of ∼−11 ± 11 s and −6 ± 17 s for W, and ∼1 ± 6 s and −8 ± 8 s for Se, i.e., all essentially zero, accounting for experimental uncertainties. These results highlight the importance of the presence of a pre-existing WSe₂ thin film concerning the nucleation and growth phenomena.

We also investigated the effect of substrate temperature in conjunction with the ratio of incident fluxes, on the growth of WSe₂ thin films. Figs. 2(c) and 2(e) display the XRF intensities from W and Se as a function of time at higher substrate temperatures of T_s = 470 and 540 °C, respectively. The sequence of exposures to Se_n and W(CO)₆ was identical to that considered in Fig. 2(a). As may be seen from Fig. 2(c) for growth at T_s = 470 °C, the results are quite similar to those observed at T_s = 400 °C: we observe a nucleation delay associated with the first exposure to W(CO)₆ (∼203 ± 18 s for W and ∼196 ± 11 s for Se), and minimal delays associated with the second and third exposures (∼−1 ± 7 and −12 ± 9 s for W, and ∼1 ± 4 and −6 ± 6 s for Se, all essentially zero). The elemental ratio Se/W for growth at T_s = 470 °C approaches a value of 1.74 ± 0.01 at long times, slightly smaller than that observed at 400 °C, as may be seen in Fig. 2(d). At T_s = 540 °C, in contrast, we observe significantly different behavior. Concerning the intensities from XRF, the results are mostly comparable to those at the two lower temperatures—i.e., growth only for those times when both species are incident, and no growth when the W(CO)₆ incident flux is extinguished. A possible exception is what appears to be an increase in the Se signal following the second period of exposure to W(CO)₆. What is clear, as may be seen in Fig. 2(f), is the formation of a sub-stoichiometric thin film for these conditions. For this thin film grown at T_s = 540 °C, we find the elemental ratio approaches a value of 0.77 ± 0.01 at the end of the deposition. It is possible, therefore, that excess W (e.g., with respect to WSe₂) at the surface can react with incident Se_n, resulting in the deposition of Se in the absence of an incident flux of W(CO)₆.

We have also considered the growth of thin films of WSe₂ where the sequence of the incident species is reversed: the Se_n species is initially completely extinguished for 720 s (plus the initial 10 s delay for both species), then the Se_n species is subsequently introduced for 720 s. Given the difference in volatility of the pure components, we suspect that this mode of growth could lead to the deposition of non-stoichiometric, W-rich thin films. In Fig. 3(a), we present a plot of the W and Se XRF intensities as a function of time, at a substrate temperature of T_s = 400 °C. First, for the first period of exposure, i.e., only W(CO)₆, followed by both W(CO)₆ and Se_n, the behaviors of the two intensities are not significantly different from the reverse case we considered in Figs. 2(a) and 2(b). There is minimal growth until both species are incident, and analysis of the nucleation delay after the Se_n is first incident gives values of ∼153 ± 10 s for W and ∼129 ± 6 s for Se. In addition, the nucleation delays for the second and third exposures to Se_n are minimal, namely, ∼27 ± 3 and −3 ± 5 for Se, all essentially zero, similar to the behavior observed in Fig. 2.

Following the extinguishing of the Se_n incident flux, however, we observe strikingly different behavior compared to growth at the same substrate temperature but with the gating sequence reversed. Indeed, we see that the signal from the W XRF continues to increase, whereas the Se XRF intensity decreases. These changes in the uptake of W and Se result in an oscillation in the composition of the thin film, as may be seen in Fig. 3(b). At the end of the first Se_n exposure, the elemental ratio Se/W reaches a value of ∼1.7, whereas dropping back to a value of ∼0.8 before the second Se_n exposure is initiated. This oscillation in the elemental ratio continues for the second and third exposures to Se_n, whereas the ratio approaches a constant value of 1.20 ± 0.01 at the end of growth. Significantly, at the end of the experiment when both incident fluxes have been terminated, the intensities remain constant (and the elemental ratio) over the period examined (∼500 s), thus the thin film is stable in the absence of the incident flux of both W(CO)₆ and Se_n. Given these observations, we have also analyzed if a delay exists in the sudden change that occurs in the Se uptake (i.e., growth versus depletion) as the Se_n incident flux is extinguished. Analysis shows that the delay for the first and second occurrences of this event are ∼9 ± 3 s and 1 ± 4 s, based on the intensity of the Se XRF signal, i.e., basically zero.

We again looked at the effect of substrate temperature in conjunction with the ratio of incident flux for the case where we modulate the incident flux of Se_n. Figs. 3(c) and 3(e) display the intensities of W and Se from XRF as a function of time, for substrate temperatures T_s = 470 °C and T_s = 540 °C, respectively. At T_s = 470 °C, there is minimal growth until both species are incident, and analysis of the nucleation delay after the Se_n is the first incident, gives values of ∼230 ± 17 s for W and ∼229 ± 13 s for Se. Unlike the reaction conditions that we have considered to this point [e.g., displayed in Figs. 1(a), 2(a), 2(c), 2(e), and 3(a)], we observe a decidedly non-zero delay in the increase in the Se signal from XRF after the Se_n incident flux is re-introduced for the second and third periods; namely, we find values of ∼133 ± 8 and ∼116 ± 7 s, respectively. Similar to what we observe at T_s = 400 °C, we also observe sudden changes in the intensity of the W and Se signals as the incident flux of Se_n is extinguished—a continuous increase in the W intensity, and a decrease in the Se intensity. Oscillations in the elemental ratio are also observed at T_s = 470 °C, as may be seen in Fig. 3(d). Analysis of the delay associated with the termination of the incident flux of Se_n at the end of the first and second periods gives ∼11 ± 5 s and −1 ± −4 s based on the intensity of the Se XRF signal. Again, there is essentially no delay in this response.

At the highest substrate temperature examined, T_s = 540 °C, we observe what appears to be severely degraded growth of WSe₂, and mostly W is being deposited, as may be seen in Figs. 3(e) and 3(f). Following the first exposure to Se_n, the elemental ratio Se/W may have reached a value of ∼0.5, whereas after the termination of the incident flux of Se_n, the ratio drops to ∼0.2. There is a hint of oscillations in the elemental ratio, but it is not nearly as strong at this substrate temperature. At the end of the deposition, the ratio approaches a value of 0.13 ± 0.01, indicating the formation of a very W-rich thin film.

We have conducted post-deposition, ex situ surface analysis using x-ray photoelectron spectroscopy. These measurements can serve to confirm the stoichiometry of the thin films observed using XRF, but they can also provide insight into the stability of the thin films when exposed to ambient air. Many dichalcogenides are susceptible to oxidation, losing the chalcogen in an exchange process with oxygen. Figure 4 plots XP spectra of the W(4f) and Se(3p) regions for four experiments: those that gated W(CO)₆ at T_s = 400 and 540 °C [(a) and (b)], and those that gated Se_n at T_s = 400 and 540 °C [(c) and (d)]. We begin with the former two experiments. We observe the expected doublet features for both W(4f_7/2,4f_5/2) and Se(3p_3/2,3p_1/2). For the main W(4f_7/2) peak displayed in Fig. 4(a), we see that it shifts to lower energy at the higher substrate temperature of 540 °C and the peak also broadens, indicating a mixture of oxidation states involving W(metal) and WSe_x. We also observe a small shoulder on the high binding energy side, indicating the presence of WO_x. This oxide feature appears at similar binding energies at both 400 and 540 °C, unlike the lower energy feature. Finally, we will demonstrate below that in areas where the thin film covers the substrate, the film is thick enough that the photoelectron intensity should be saturated due to the finite escape depth. Thus, the small intensity for the thin film grown at 540 °C indicates that its coverage of the substrate must be smaller than that observed at 400 °C. Moving on to the results for the Se(3p) feature, we find similar binding energies for the thin films grown at 400 and 540 °C, suggesting similar chemical environments for the Se in these films. We also observe a much smaller intensity for the thin film grown at 540 °C, which can be due to the smaller coverage, and also a difference in the stoichiometry of the thin films.

For the thin films grown by gating of the Se_n incident flux, we observe spectra that are quite different from those grown with the gating of the W(CO)₆ incident flux. For example, as may be seen in Fig. 4(c), the shape (e.g., FWHM) of the low binding energy peaks for the W(4f) feature are similar for the two substrate temperatures. However, the binding energy observed for growth at 540 °C is consistent with metallic W. A shoulder for WO_x is observed at both substrate temperatures, whereas the FWHM is larger for growth at 540 °C. Concerning the Se(3p) feature, displayed in Fig. 4(d), the peaks are similar to those observed for growth with gating of the W(CO)₆ incident flux. As may be seen, the intensity of the Se(3p) peak is greatly reduced for growth at 540 °C.

The XP spectra shown in Fig. 4, and ones not shown (e.g., growth conducted at 470 °C) were all analyzed to determine binding energies of the individual features, and the shape of the peaks (i.e., FWHM). The smooth curves plotted in Fig. 4 represent the results of these fits. First, concerning the W(4f) feature, we expect as many as three oxidation states associated with W(metal), WSe₂, and WO₃. Based on prior work, the binding energies expected for these species are 31.4, 32.3–32.7, and 35.7.^52–54 Based on the significant overlap of the former two species, we fit the W(4f) feature to two species (i.e., W/WSe₂ and WO_x), fitting the doublet (4f_7/2, 4f_5/2) in each case. Thus, an unexpectedly large FWHM will indicate multiple oxidation states are present. Fig. 5 presents (a) the binding energy and (b) FWHM for growth at all conditions where the incident flux of either W(CO)₆ or Se_n was gated. As may be seen in Fig. 5(a), for five of the conditions examined, the features could be fit well with two peaks (doublets) representing the binding energies expected for WSe₂ and WO₃. The one obvious outlier is the growth at T_s = 540 °C and Se_n gating, where the fit of the W(4f) peak indicates the presence of a significant amount of metallic W.

Concerning the FWHM, we see for growth at 400 and 470 °C for both W(CO)₆ and Se_n gating, the width of the peaks are approximately constant, and that associated with the WO_x feature is consistently larger (this effect is seen in other systems, even for oxides that represent a single oxidation state, e.g., Si/SiO₂).⁵⁵ These results indicate the co-existence of near stoichiometric WSe₂ and WO₃ for these conditions. For growth at 540 °C, however, the peaks for both features are significantly broadened, except for the outlier identified above, the low-binding energy feature for growth with a gated incident flux of Se_n. Thus, these results indicate the formation of a mixture of oxidation states for the WO_x phase, and also the W/WSe₂ phase for growth using a gated incident flux of W(CO)₆.

Fig. 5(c) displays the results of the fits to the Se(3p) feature. As may be seen, the binding energy for the Se(3p_3/2) peak is quite similar in all cases, and the mean and standard deviation are given by 161.2 ± 0.13 eV. Values for the Se(3p_3/2) have been reported to lay between 160.8 and 161.9 eV for elemental (metallic) Se,⁵⁶ whereas a most recent study reports 161.3 for Se in WSe₂.⁵⁴ This value is in excellent agreement with our observations. The peak widths are also quite similar, except again for the previously identified outlier for growth at T_s = 540 °C and Se_n gating.

To conclude our presentation of the results from XPS, we now consider the stoichiometry of the deposited thin films. We note that XPS is a surface-sensitive technique, providing compositional information concerning the top few nm, unlike XRF that probes the entire thickness of the thin films produced here. Figure 6(a) plots the elemental ratio found from XPS and XRF as a function of the substrate temperature for the experiments where the incident flux of W(CO)₆ is gated. For XPS, we plot two elemental ratios: one where we include only the low binding energy W(4f) features and another where we consider the entire W(4f) peak. For T_s = 400 and 470 °C, we see that the Se/W_{low BE} ratio is close to the stoichiometric value of 2, dropping to a value of ∼1.1 at T_s = 540 °C. Concerning the ratio found using the entire W(4f) peak, we find that these values track those found from in situ XRF (within ± 0.1). This result suggests that air exposure does not result in significant loss of Se, or significant thin film reorganization involving, for example, surface segregation. It also indicates that the thin film composition is relatively uniform over the entire thickness of the WSe₂ thin film.

Fig. 6(b) plots the elemental ratio found from XPS and XRF as a function of the substrate temperature for the experiments where the incident flux of Se_n is gated. Again, for T_s = 400 and 470 °C, we see that the Se/W_{low BE} ratio is close to the stoichiometric value of 2, dropping to a value of ∼0.3 at T_s = 540 °C. Concerning the ratio found using the entire W(4f) peak, we find that these values do not track those found from in situ XRF, and they are consistently higher by an amount ∼0.5 at T_s = 400 and 470 °C. This result suggests that the near-surface layers of the thin film are rich in Se, where a W-rich phase is covered by a near stoichiometric WSe₂ thin film.

We have performed post-deposition, ex situ structural characterization using x-ray diffraction. In Fig. 7, we plot the scattered x-ray intensity as a function of the out-of-plane scattering vector, q_z, for the series of experiments we considered above in Figs. 2–6. In Fig. 7(a), we consider the experiments where W(CO₆) was gated at T_s = 400, 470, and 540 °C, and in Fig. 7(b), those where Se_n was gated at T_s = 400, 470, and 540 °C. To highlight diffraction features from the thin films and not the HOPG substrate, we masked the region where the (002) diffraction peak from HOPG would appear. First, for the experiments where W(CO₆) was gated, we observe diffraction features that correspond to the (002) and (006) peaks of WSe₂. It is also apparent that diffraction peaks become weaker and broader at the highest substrate temperature examined. The positions of these (002) peaks are found to be at q_z = 0.965, 0.966, and 0.966 Å⁻¹ for T_s = 400, 470, and 540 °C, respectively, similar to values reported previously (e.g., 0.969 Å⁻¹).⁵⁷ A similar agreement is found from the analysis of the (006) peak. Next, in Fig. 7(b), we consider the experiments where the Se_n incident flux was gated. At T_s = 400 and 470 °C, we observe the (002) peak for WSe₂, however, the higher-order (006) peaks are less distinct than those observed for the thin films deposited with W(CO)₆ gating. Most apparent, at T_s = 540 °C, we only observe one diffraction feature at q_z = 2.81 Å⁻¹, which corresponds to the (211) diffraction peak from elemental W.⁵⁸ This result is entirely consistent with the thin film stoichiometry indicated by the results from XRF [Fig. 3(f)] and XPS [Fig. 6(b)], which shows that this thin film is predominantly comprised of tungsten.

The results from XRD shown in Fig. 7 have been analyzed to determine the peak positions and the peak shapes (FWHM). The former can be used to calculate the d-spacing [we used both the (002) and (006) peaks for this purpose], whereas the latter can be used to calculate the out-of-plane grain sizes of the thin films [only the (002) peaks were used here]. We fit these diffraction features to a pseudo-Voigt function and used the Scherrer equation to calculate a grain size.⁵¹ Figure 8 presents both the calculated d-spacing (open symbols, left ordinate) and the crystalline grain size (filled symbols, right ordinate) as a function of the substrate temperature for the 5 cases where a WSe₂ thin film was deposited. As may be seen, for four of these five cases (excluding T_s = 470 °C and Se_n gating), we observe a very similar d-spacing for the WSe₂ thin films of ∼6.504 ± 0.003 Å (cf. to values of 6.47–6.50 reported previously).^59,60 From our analysis, at T_s = 470 °C and with Se_n gating, the d-spacing increases to a value of ∼6.526 Å (0.34%). The calculated grain sizes for growth with W(CO)₆ gating exhibit a strong dependence on the substrate temperature at which the thin films were deposited. The grain size decreases with increasing temperature, from a value of ∼19 nm at T_s = 400 °C, to a value of ∼8 nm T_s = 540 °C. We note that a similar analysis of the thin film grown with coincident fluxes of W(CO)₆ and Se_n, considered above in Fig. 1, gave a grain size of ∼28 nm. For growth using Se_n gating, we see that smaller grains are formed in both cases (∼6 nm).

We have investigated the film morphology using ex situ scanning electron microscopy (SEM) and AFM. We consider the former first. Fig. 9 displays SE micrographs (6.5 × 6.5 μm²) for the series of six experiments we have considered above in Figs. 2–8. We also display a micrograph for the bare HOPG substrate. First, as expected, the bare HOPG substrate [Fig. 9(g)] appears flat and featureless. Examining the micrographs for the thin films, we observe long, one-dimensional features, with spikelike features on their edges, similar to a pipe/tube cleaner. In both cases [W(CO)₆ and Se_n gating], the coverages of the substrate by the thin films decrease with increasing substrate temperature. We also observe that in many cases, the 1D features are approximately parallel to each other. A likely cause for these features could be the preferential nucleation of the thin films at the step edges present on the HOPG surfaces. At minimum, we do not observe isolated triangular crystalline grains that have been observed in other work involving WSe₂ thin films, particularly for thin films consisting of a single or a few monolayers.²⁵ Finally, the micrograph for the thin film that is predominantly metallic W is distinct from those representing WSe₂—the 1D features lack the spikelike edges and exhibit more regular polyhedron shapes.

We have also investigated the thin-film morphology using ex situ AFM. Figure 10 displays AF micrographs (15 × 15 μm²) for the series of six experiments we have considered above in Figs. 2–9. These results are in general agreement with those we considered above in Fig. 9 using SEM—one-dimensional features, often parallel to each other, with spikelike edges. Unlike in SEM, the step edges on HOPG are apparent in the AF micrograph [Fig. 10(g)], as are a number of straight step edges (the RMS roughness is 8.89 nm). As each HOPG sample was unique (due to the cleaving procedure), we hesitate to speculate more as to the density and orientation of the step edges. We note that previous work examining MBE of WSe₂ on HOPG has observed features from AFM very similar to what we report here.²⁴ Concerning the thin films, we have conducted two sets of analyses. First, we have used these micrographs to estimate the amount of thin film that was deposited in each case. We used the images to calculate the volume deposited and then calculated an apparent thickness as if this volume were spread over the entire area analyzed by AFM. For the images corresponding to W(CO)₆ gating [Figs. 10(a)–10(c)], we find mean thin film thicknesses of 20.7, 20.3, and 1.8 nm at T_s = 400, 470, and 540 °C. For Se_n gating [Figs. 10(d)–10(f)], we find mean thin film thicknesses of 24.8, 9.6, and 3.3 nm at T_s = 400, 470, and 540 °C. In the supplementary material,⁷⁷ we consider the correlation between these thicknesses and the XRF signal from W. After accounting for differences in the density of W in WSe₂ and W metal (and corresponding differences in volume), we find a good correlation between the mean thickness estimated from AFM and the W XRF intensity, particularly for the thicker, thin films. In Fig. 11, we plot representative line scans for the lowest and highest substrate temperatures examined for (a) W(CO)₆ and (b) Se_n gating. We also display the mean thickness that is determined by the volume integration. As may be seen, the features are much taller at the lower substrate temperature, reflecting the thicker (mean) thin film deposited at the lowest substrate temperature. Also, the deviation of the line scan from the mean thickness is apparent at the lower substrate temperature, reflecting the lower coverage of the substrate for these conditions. We can also use the images to generate normalized height distributions (using selected areas of the images) for the features we observe, and these results are shown in Fig. 11 again for the lowest and highest substrate temperatures examined for (c) W(CO)₆ and (d) Se_n gating. These results also show that the features are in general much taller for thin films deposited at the lowest substrate temperature.

We have examined the deposition of thin films of WSe₂ on a HOPG substrate using W(CO)₆, delivered via a supersonic molecular beam, and Se_n, delivered via an effusion cell, as the reactant sources. We have employed x-ray fluorescence using a synchrotron source to examine thin-film nucleation and growth in situ and in real-time. The deposited thin films have also been characterized ex situ using x-ray photoelectron spectroscopy, x-ray diffraction, scanning electron microscopy, and atomic force microscopy. An emphasis of this work has been on an examination of the effects of the ratio of the incident fluxes of the W and Se containing species, and the substrate temperature.

First, we have shown that we can deposit near-stoichiometric crystalline thin films of WSe₂ in UHV using W(CO)₆ as the thin film precursor and elemental Se_n as the co-reactant. This is significant since transition metal carbonyls are known to be relatively unreactive at thermal incident kinetic energies.⁶¹ The thin films also possess little C contamination but are susceptible to post-deposition oxidation, via exchange with Se in the thin films. Analysis of the deposited thin films from XRD indicates rather small crystalline grains, on the order of ∼5–30 nm, which is also the approximate (local) thickness of the thin films. The thin films do not uniformly coat the HOPG substrate, rather one-dimensional features are formed that are rough and spikelike along their edges.

Moving on to the emphasis of our work, we observed significant differences when growth is conducted where one of the sources is extinguished. In general, extinguishing the incident flux of W(CO)₆ (or both W and Se sources) does not result in changes in the thin film thickness or stoichiometry (and by implication the microstructure). In contrast, extinguishing the Se_n incident flux results in additional deposition of W, and loss of Se, and a change in (overall) stoichiometry. Some of these observations are consistent with expectations. For example, concerning the growth of a compound thin film material, it is generally desirable to use a higher incident flux of the component species that possesses a higher vapor pressure (as a pure elemental species). If one does not do so, the growth of an elemental thin film (and/or rich in that component) of the species possessing the lower vapor pressure could occur. In the case of WSe₂, the refractory transition metal W has a vapor pressure that is lower than that of the chalcogen Se by multiple orders of magnitude. For example, the vapor pressure of Se is calculated to be ∼3–80 Torr over the substrate temperatures (T_s = 400–540 °C) considered here. Likewise, the vapor pressure of the precursor W(CO)₆ is high, reaching a value of 760 Torr at ∼176 °C. In contrast, W (metal) is known to possess the lowest vapor pressure of any transition metal. For example, to reach a vapor pressure for W representative of the UHV regime (1 × 10⁻¹⁰ Torr), a temperature of ∼1900 °C is required.⁶² These arguments of course are most relevant concerning the growth of the pure elements or compounds, but they do provide guidance as to the likely behavior. One must also consider the binding of species to the substrate and eventually the deposited thin film.

For all conditions considered here, we observed a nucleation delay—a finite period passed before we could detect the growth of a thin film, independent of its exact composition. Concerning this issue, largely due to the lack of in situ real-time probes, characterization of this stage of growth of TMDs has not been examined as thoroughly as other aspects of growth. Our use of in situ real-time XRF, as well as precise control of the incident fluxes of the two reactant species, has enabled our ability to directly measure the presence and extent of the nucleation delay. In all cases, the nucleation delay was within 140–250 s. For the measured growth rates, this period would result in the deposition of ∼1–2 nm, or 2–3 monolayers of WSe₂.

To frame our discussion of the nucleation regime, we consider the scenarios we display in Fig. 12. First, given the number of observations and considerations of the surface chemistry, we can state with certainty what the first crucial step is: chemisorption of the W(CO)₆ thin film precursor. This is supported by the observation that W can be deposited in the absence of an incident flux of Se, whereas the inverse is not true. The binding of a transition metal is also expected to be much stronger than that of the nonmetal main group element, Se. Concerning the W(CO)₆ thin film precursor itself, there have been a few studies examining its reactivity. On a W surface, dissociative chemisorption of W(CO)₆ is relatively facile.⁶³ On other surfaces, such as dielectrics and polymers, the reactivity is much lower.⁶⁴ In our case, delivery of the W(CO)₆ precursor in a seeded supersonic molecular beam results in dissociative adsorption with high probability.^61,65

Thus, we are left with the possible interactions displayed in Fig. 12(a). One could argue that the sites presented by steps on the HOPG surface should be the most reactive, given dangling bonds/low coordination number surface atoms. We do not exclude the possibility that dissociative chemisorption could occur on the basal planes of the HOPG surface, however, with reduced probability. Once chemisorbed, for example, via loss of a single ligand, W(CO)₆(g) → W(CO)₅(a) + CO(g), the W fragment is free to diffuse along the surface, while CO is expected to thermally desorb immediately from the HOPG surface. Continued loss of the CO ligand is likely, and one would expect that the binding energy of the fragment to the surface will increase with increasing ligand loss, whereas the barrier to diffusion may increase concomitantly. Some W fragments may thermally desorb if they do not find a sufficiently strong binding site. Eventually, however, some of these fragments will make their way to the strongest biding sites that are at HOPG step sites.

Figure 12(b) displays additional steps in the nucleation of a WSe₂ thin film. As deposition of W fragments continues, some of these will react with each other to form a cluster. It is possible that a critical nucleus exists, which requires the formation of a W cluster of sufficient size to continue to grow (and also not thermally desorb). Given the high reactivity of W(CO)₆ with W metal, the cluster may continue to grow mostly via direct impact. There is a question as to when the Se_n(g) cluster may begin to react with the surface. First, the experiments we have conducted with gating of the W incident flux indicates that the Se_n(g) clusters, if they do dissociate on a clean HOPG surface, immediately desorb and do not build up a finite coverage. On the other hand, if a W-rich thin film exists on the surface, dissociative adsorption of Se_n essentially occurs immediately (cf. Fig. 3). In addition, in the case of near-stoichiometric WSe₂ thin films, re-introduction of an incident flux of W(CO)₆ (and deposition of W) results in almost an immediate uptake of Se (cf. Fig. 2). Likewise, termination of the incident flux of W(CO)₆ results in almost an immediate cessation of the uptake of Se. These two latter observations indicate that rather small clusters of W may be required to result in the dissociative adsorption of Se_n(g), and importantly, irreversible adsorption. How small—2, 3, 10 etc. atoms are—we cannot determine from our experiments.

In Figs. 12(c) and 12(d), we consider additional steps in the formation of a WeS₂ thin film. A WSe₂ cluster is formed after sufficient reaction of Se_n(g) with a cluster of W(a), and this cluster may also possess a critical size. This cluster likely nucleates near a (lower) step edge where there is a high density of reactive sites. This cluster then grows laterally, as the incorporation of W (followed by Se) is expected to be most facile at the edges of the WSe₂ crystal. Formation of multilayers of WSe₂ will first require the nucleation of a W cluster on the top of an existing layer of WSe₂ [cf. Fig. 12(c)], and then, the growth of this second layer of WSe₂, or possibly nucleation of a WSe₂ layer on an “upper” layer of HOPG, which then grows laterally to cover a lower layer. This latter possibility might require homogeneous nucleation of WSe₂ on a basal plane of HOPG and/or the existence of steps of a single unit cell in height to facilitate overgrowth [cf. Fig. 12(d)].

Once one or more layers of WSe₂ are formed, additional growth will depend on the interlayer transport in addition to direct nucleation of new layers on top of pre-existing layers. Figure 13 displays how these transport events, coupled with the addition of species due to dissociative adsorption, can contribute to the evolution of the thin film morphology. Concerning the latter, the addition of species (W first) on top of each layer depends on the product of the incident flux (F) and the probability of irreversible dissociative adsorption (S_i), the latter of which depends on the chemical identity and microstructure of each layer (i). In addition to differences in the reactivity of step versus terrace sites, there may be significant differences in the reactivity of the basal planes of HOPG (layer 0), and WSe₂ (layers 1, 2, etc.). For example, if the reaction probability on HOPG (S₀) were much higher than that on WSe₂ (S₁, S₂, etc.) then the growth of the first layer should have proceeded to cover the substrate (perhaps inhibited only by steps in the HOPG) before a second layer formed. Obviously, we did not observe this (cf. Figs. 9–11), as the multilayer WSe₂ thin films do not cover the substrate.

The apparently rapid formation of the 2nd, 3rd, and additional layers could be the result of two or more effects: much higher reaction probabilities on WSe₂ versus HOPG and/or decreased “downward” interlayer transport. For example, referring to Fig. 13, for species depositing on top of the first WSe₂ layer (SF₁), they can experience intralayer transport (horizontal arrows, blue on-line), react with another species in this layer, perhaps nucleating a cluster for a second layer, or be captured by a step site of a pre-existing second layer. Alternatively, the species could hop down to a lower layer (“downward” interlayer transport, downward pointing arrows, red on-line), which could facilitate the growth of the underlying first layer. If the latter dominates, we should observe layer-by-layer (LbL) growth where the first layer completes before the second layer nucleates. We do not observe LbL growth here, which suggests that there may be a significant (Ehrlich–Schwoebel) ^66,67 barrier to downward interlayer transport, although one calculation indicates this barrier is small on WSe₂ for W atoms.³⁵ We concede that an alternative explanation could explain some of these observations concerning the evolution of the thin film morphology, namely, “upward” interlayer transport (upward pointing arrows, green on-line). Here, if the surface energy of the deposited thin film is significantly higher than that of the substrate, there is a thermodynamic driving force for thin film reorganization and “de-wetting.”^44,48 Calculations of the interfacial binding energy (essentially twice the surface energy) using DFT methods that account for van der Waals interactions indicate that these energies lay in a relatively small range of ∼16–21 meV Å⁻² for essentially all TMDs and graphite.⁶⁸ This would argue against de-wetting as a mechanism responsible for the morphologies that we observe here.

One final issue we will discuss concerns the effects of extinguishing either source and its effect on the rate of growth and the stoichiometry for the reaction conditions we have considered here. First, in the absence of an incident flux of W(CO)₆, absolutely no growth occurs—W and WSe₂ for obvious reasons, but also Se. In the absence of an incident flux of Se_n, however, W can be deposited, and as we have discussed, Se in the thin film can be removed. Thin films of both W and WSe₂ are stable in vacuum at the substrate temperatures we have considered here, but we have found upon exposure to air results in oxidation, and a loss of Se, particularly for thin films that are rich in the W component. Many of these observations are in line with expectations given the vapor pressures of the pure components, except for the behavior of the thin films when the incident flux of Se_n is extinguished and W(CO)₆ is not.

In Fig. 14, we consider a schematic representation of the evolution of the thin films we considered in Figs. 3(c), 3(d), 9(e), and 10(e), where the incident flux of W(CO)₆ is constant, but the incident flux of Se_n is modulated, and the substrate temperature is 470 °C. In this figure, we assume that growth proceeds from a step edge, shown proceeding in both directions, to match the spikelike edges seen in Figs. 9(e) and 10(e). We designate regions covered by WSe₂ in a copper/bronze tone and those covered by W metal in blue. The areas are chosen to represent the amount of material deposited with W as a constituent and do not reflect density differences between W and WSe₂. Fig. 14(a) shows that after the first exposure to both W(CO)₆ and Se_n, a thin film is formed that is close to stoichiometric (WSe_x, x ∼ 1.81) in composition. As the Se_n incident flux is extinguished, the stoichiometry begins to change immediately, as W continues to grow, whereas the amount of Se in the thin film is reduced. At the end of this cycle of exposure to only W(CO)₆, the stoichiometry has dropped to a value of x ∼ 0.67, as illustrated in Fig. 14(b). Re-introduction of Se_n, as shown in Fig. 14(c), results in renewed growth of WSe₂, whereas W metal also continues to grow, resulting in an overall stoichiometry of x ∼ 1.14. Finally, as the Se_n is extinguished for a second time, W continues to grow, whereas the amount of Se in the thin film is reduced, as we display in Fig. 14(d). At the end of this final cycle of exposure to only W(CO)₆, the stoichiometry has dropped to a value of x ∼ 0.71. Of course, the situation is likely not as simple as we display in Fig. 14, where we indicate that growth tends to expand laterally, as opposed to some combination of lateral and vertical growth. Nevertheless, the oscillation in stoichiometry, including the removal of Se from the thin film, is inescapable.

What is the nature of the mechanism that leads to the removal of Se? First, in all cases, when both incident fluxes are terminated (cf. Figs. 2 and 3), neither the loss of Se or W is observed within experimental uncertainties. Thus, whatever thin films are formed, they are stable in UHV at the growth temperature. When the Se_n incident flux is terminated, W(CO)₆ remains incident on the surface, whereas Se is lost while W continues to deposit. We propose that the most likely explanation for our observations is a reaction between the W(CO)₆ precursor and Se in the thin film. This could involve the following reactions:

Dissociative adsorption of the W(CO)₆ thin film precursor,

additional decomposition reactions,

thermal desorption of adsorbed CO,

and the reaction of adsorbed CO with Se,

First, is SeCO, carbonyl selenide, a plausible reaction gas phase product? SeCO possesses chemical similarities to other Group 16 chalcogen carbonyls (linear triatomic species), including carbonyl sulfide (SCO), and OCO or CO₂. SeCO possesses a very high vapor pressure (∼1 atm at −22 °C)⁶⁹ but is a relatively unstable molecule.⁷⁰ Thus, under UHV conditions, once SeCO desorbs into the gas phase, it probably decomposes on the stainless steel walls of the chamber, depositing Se, and releasing CO into the gas phase. In any event, reaction,⁴ effectively a ligand exchange reaction is the most likely event that leads to the loss of Se in the thin film and should be considered in any chemical model for the growth of Se containing thin films of TMDs.⁷¹ 

Figure 15(a) displays the various events that lead to the etching of Se by the incident W(CO)₆, and we considered similar events concerning the nucleation stage of growth in Fig. 12(a). These events include dissociative adsorption of W(CO)₆ (on terrace and step sites), and diffusion of adsorbed W(CO)_6−n(a) fragments, and possibly CO(a). Concerning the Se atoms present on the surface, we consider Se located in defect-free basal planes of WSe₂, the least likely candidates to be etched by CO. Se located at step and kink sites (or other defect sites), however, are much more susceptible to removal from the surface to the gas phase in the form of SeCO(g). These atoms are bound to fewer nearest neighbor W atoms as compared to the basal planes and should be more reactive.^{21,35,72–74} Based on our observations from SEM and AFM (cf. Figs. 9 and 10), the WSe₂ thin films we have deposited here on a HOPG surface possess a high density of step and kink sites making them very susceptible to Se etching by W(CO)₆. As to the exact nature of the intermediate and/or transition state, that is responsible for SeCO formation, it may involve multiple atoms and species—i.e., not simply diffusion of an adsorbed CO molecule to an appropriate Se site. For example, a four-center transition state, such as we show in Fig. 15(b), could be involved concerning the interaction between an adsorbed W(CO)_m(a) fragment and a step site, leading to the formation of SeCO(g). Multi-nuclear complexes containing W, Se, and CO ligands can be formed in solution,⁷⁵ and these complexes possess both W–W bonds, W–Se bonds, and W–CO bonds. What we cannot completely exclude at this point is whether a sufficiently high incident flux of CO(g) could produce similar effects. Arguing against a mechanism involving CO(g) would be its weak binding with the WSe₂ surface. Additional work may be required to completely resolve the mechanism responsible for the removal of Se from the thin film.

Finally, do our observations extend to growth via MBE or MOCVD? In the work we have reported here, we use a gas phase source for W [W(CO)₆], delivered via a supersonic molecular beam, and elemental Se (Se_n) delivered via an effusive Knudsen cell. The former has been employed in the MOCVD growth of WS₂ and WSe₂, whereas the latter has been employed in studies using MBE. Concerning MBE growth, we make the following comments. Growth from both elemental W and Se sources will be similar to what we observe here except for our observations when the Se source is extinguished. We would predict deposition of W, but no change in the content of Se, as the removal of Se in the form of volatile SeCO will not occur. Behavior arising from the extinguishing of the incident flux of W would be similar to what we observe here, i.e., no change in the thin film thickness or composition.

For MOCVD growth, first, depending on the pressure and temperature, gas phase reactions may occur that are totally absent in our case. For MOCVD, the Se can be supplied in the form of H₂Se, Se(CH₃)₂, or Se(C₂H₅)₂, for example. First, if the W source is extinguished at the substrate temperatures examined here, the deposited WSe₂ thin films should be stable, with neither a loss of Se or a gain of Se due to reactions of the Se precursors. The latter is controlled by the vapor pressure of Se(s), which is high for the reaction conditions we have examined here. Extinguishing the flow of the Se precursors could result in the removal of Se by reactions with W(CO)₆, as we have proposed here. Could gas-phase CO produce a comparable effect? To answer this question, one would have to conduct high-level quantum chemistry calculations of possible transition states between surface Se(a) and either CO(a) or W(CO)_n(a) (the latter, a four-center transition state, as shown in Fig. 15). A recent study⁷⁶ has observed the effects of gas-phase CO on the growth of MoS₂ from Mo(CO)₆ and H₂S. The possible formation of SCO was not discussed in this work, nor were direct measurements of thin film stoichiometry collected in situ and in real-time as we have reported here.

In this work, we have examined the nucleation and growth of WSe₂ thin films on HOPG using synchrotron x-ray radiation, specifically in situ x-ray fluorescence to monitor deposition in real-time. We employed W(CO)₆ as the W source, delivered via a supersonic molecular beam, and Se_n delivered via an effusion cell. For a variety of reaction conditions, we were able to deposit near stoichiometric WSe₂ thin films on HOPG, which exhibited strong diffraction features indicating the growth of polycrystalline thin films. Given the ballistic nature of the reactant sources employed in this study, we were able to examine the effects of extinguishing either source (W or Se)—both before and during growth—and examine the effects in real-time. We found that when the W(CO)₆ source was extinguished, no growth occurred (i.e., from the incident Se_n), and the deposited thin film was stable in the absence of the incident flux of this W species. For substrate temperatures at or below 470 °C with W(CO)₆ gated, near stoichiometric WSe₂ thin films were formed, whereas at 540 °C the film consisted of a mixture of WSe₂ (observed in XRD) and W (implicated from XRF and XPS). We found substantially different results when the Se_n incident flux was gated. For all substrate temperatures, we observed an oscillation in the overall stoichiometry of the deposited thin film, with Se being removed in the absence of an incident flux of Se_n, but with W(CO)₆ incident. As was the case when W(CO)₆ was gated, the thin film formed at 540 °C was much more W-rich as compared to those formed at the lower substrate temperatures. From ex situ XPS, the stoichiometry (Se:W) of the thin films found using XRF was confirmed, whereas all thin films exhibited features due to WO₃, in addition to those assigned to WSe₂ and W, indicating partial oxidation of the thin film when exposed to air. From SEM and AFM, we found that for all cases examined, the thin films were not continuous, i.e., covering the HOPG substrate, but consisted of filamentlike features with spikelike edges. We speculate that these may form along step edges on the HOPG substrate, first with the formation of small W clusters before eventually forming WSe₂. Finally, the striking observation of the removal of Se from the thin film in the absence of a flux of Se_n, but in the presence of a flux of W(CO)₆ is best explained by a ligand exchange reaction, where CO bound to either the HOPG surface or a W(CO)_n fragment reacts with Se to form a volatile SeCO species.

We would like to thank Arthur R. Woll, Howard Joress, Jiun-Ruey Chen, Taewon Suh, A. Sundar, A. Chaney, and H. H. Lien for their invaluable technical contributions. This work is based on the research conducted at the Cornell High Energy Synchrotron Source (CHESS), which was supported by the National Science Foundation under NSF award DMR-1332208. H.J.B. acknowledges support from the Alfred. P. Sloan Foundation.

The data that support the findings of this study are available from the corresponding author on reasonable request.