Thermal stability of monolayer WS₂ in BEOL conditions

The substrates used for CVD growth of monolayer WS₂ were dice cut from c-axis, HEMCOR single-crystal, sapphire α-Al₂O₃ (0001) wafers supplied by Alfa Aesar (Germany). Before WS₂ growth, sapphire substrates were cleaned via sonication in acetone, isopropanol, and de-ionized (DI) water, then immersed in piranha solution (1:3, H₂O₂:H₂SO₄) for 15 min and finally washed in DI water. Subsequently, the dice were etched in hydrogen atmosphere as described in [58] to remove polishing scratches and reveal atomic steps.

The etched sapphire substrate was then directly loaded in the CVD reactor for the growth of monolayer WS₂ via LP-CVD. Tungsten trioxide (WO₃, powder, Sigma Aldrich, 99.995%) and sulfur (S, pellets, Sigma Aldrich, 99.998%) were used as solid precursors. The process was performed within a 2.5 inches horizontal hot-wall furnace (Lenton PTF): the furnace comprises a central hot zone (growth zone), where a crucible loaded with WO₃ powder was placed 3 cm away from the growth substrate, and an inlet zone, where the S powder was positioned and separately heated by a resistive belt at 120 °C. Temperature in the growth zone was concurrently ramped-up to 930 °C, at a chamber pressure of ∼5 × 10⁻² mbar. During the growth Argon was adopted as carrier gas, as described in [37, 59].

Medium-vacuum thermal annealing of WS₂ was performed in a hot-wall CVD quartz tube reactor, while high-vacuum annealing was carried out in an ultra-high-vacuum (UHV) chamber used for XPS measurements. In the CVD reactor, the annealings were performed at a temperature of 300 °C, a base pressure of 1 × 10⁻³ mbar and a temperature ramp-up rate of 5 °C min⁻¹. The medium-vacuum annealing was carried out either in a continuative manner up to 10 h or by extracting and characterizing the sample after each annealing step. The high-vacuum annealing was performed at 300 °C, at a base pressure of 2 × 10⁻⁸ mbar for 10 h, consecutively.

Raman and PL analyses were performed using a Renishaw inVia system equipped with a 532 nm green laser, a 100× objective lens (0.89 NA) and a spot size of ∼1 μm. All Raman and PL experiments were carried out in standard laboratory conditions (temperature of 22 °C, 30% humidity) and at atmospheric pressure. Before and after each analysis, the samples were kept in dark and in low vacuum (1 mbar). Different laser powers were used to evaluate the influence of laser exposure on the thermal stability of monolayer WS₂. All single spectra were acquired at the center of the same crystal and, unless otherwise stated, the laser power was set at 730 μW or 73 μW and the exposure time was 10 s. The fluence was calculated to be 9.4 mJ μm⁻² and 0.9 mJ μm⁻², respectively. SEM analysis was carried out using an in-lens detector, with an accelerating voltage of 2 keV and an electron current of 12 pA (specimen current <0.5 pA) in order to minimize the charging of the insulating substrate and the damaging due to the electron beam irradiation. Low energy electron diffraction (LEED) measurements were performed in UHV with a SPECS Er-LEED optics. The electron beam energy was kept above 150 eV to avoid charging from the sample. XPS was performed within a PHI-instruments 5800 station equipped with a monochromatized Al K_α X-ray source. The X-ray spot size on the sample was few-hundred micrometers. The energy resolution was set to 100 meV for high-resolution spectra. The BE scale was calibrated by setting the adventitious carbon C 1s peak at 284.8 eV. The W 4f and S 2p peaks were fitted by means of an approximated Voigt function (Gaussian/Lorentzian product form, mixing = 0.5, chosen from CasaXPS line-shapes database).

Monolayer WS₂ was grown on hydrogen-etched α-Al₂O₃ (0001) [58] using LP-CVD following the procedure described in the Experimental section. Figure 1(a) shows a typical secondary electron (SE) micrograph of the synthesized WS₂ single-crystals with a lateral average size of 5 μm. The darker SE contrast, obtained with the in-lens detector, is due to the attenuation of the secondary electrons emerging from the substrate beneath the layer of WS₂ [60]. The symmetry of the Al₂O₃ (0001) surface imposes a registry on which WS₂ crystals are observed to grow with rotation of 30° ± 60° with respect to the crystal lattice of the underlying α-Al₂O₃. This mutual orientation is clearly visible in the SEM image in figure 1(a) and the percentage of misoriented crystals is estimated to be 30% over a dataset of 161 single-crystals. An LEED pattern measured over an area of about 1 mm² is shown in figure 1(b) and further confirms the epitaxial alignment of the synthesized WS₂ crystals with respect to the substrate. The pattern perfectly matches the superimposed model (blue and red circles), which represents WS₂ R30 (blue circles) over the Al₂O₃ (0001) surface (red circles). No pattern of the Al-rich (√31 × √31) ± 9 surface reconstruction is instead visible after WS₂ growth [58]. The displayed LEED pattern also exhibits some diffused intensity, which cannot be ascribed neither to the substrate, nor to the WS₂. The diffused intensity is in fact most likely due to unreacted amorphous WO₃ or to some transient chemical compound, as also suggested by XPS results (see figure 4). Minor rotational disorder, due to the presence of few misoriented WS₂ single-crystals over the large area analyzed, is also visualized in the LEED pattern as a very faint continuous ring of intensity, crossing the main WS₂ reflections, as suggested also by SEM results (figure 1(a)).

Zoom In Zoom Out Reset image size

Raman mapping (figure 1(c)) and spectroscopy (figure 1(d)) confirm that WS₂ is monolayer with a high degree of homogeneity. The benchmarking for WS₂ monolayer is that the ratio of the 2LA(M) + E_2g(Γ) and A_1g(Γ) Raman modes is higher than 2.2, as previously demonstrated in [61]. It is worth noting that the increase of the 2LA(M) + E_2g(Γ)/A_1g(Γ) ratio on the edge of the crystal, visible in figure 1(c), is an intrinsic artifact of high-resolution Raman mapping. When the laser reaches the edge of the crystal, there is a concurrent decrease of all mode intensities and the intensity of the A_1g(Γ) goes close to noise level, causing an increase of the 2LA(M) + E_2g(Γ)/A_1g(Γ) ratio. PL measurements (figures 1(e) and (f)) also confirm that the synthesized WS₂ is monolayer: the room-temperature PL spectrum, reported in figure 1(e), shows an intense emission at (630.7 ± 0.1) nm (1.96 eV), related to the exciton of monolayer WS₂. The PL peak is slightly red-shifted if compared to the PL peak position typically observed for WS₂ on sapphire (∼620 nm) [57, 62–64]. Moreover, this peak has a strong asymmetry, showing a tail on the high wavelength side. These results are likely due to the presence of intra-gap states related to sulfur vacancies, as reported by Carozo et al [65]. The PL map (figure 1(f)) shows an increase of intensity at the edges of the crystal, while the center shows homogeneous PL signal, as it was also previously reported [13, 66]. The statistic distributions of 2LA(M) + E_2g(Γ)/A_1g(Γ) ratio, A_1g(Γ) Raman shift, PL intensity and PL position are reported in figure S1 (available online at stacks.iop.org/JPMATER/4/024002/mmedia).

To investigate the stability of WS₂ in typical BEOL conditions, freshly grown WS₂ single-crystals were initially annealed at a temperature of 300 °C and medium-vacuum (pressure of 10⁻³ mbar). These conditions were chosen to be as close as possible to those at which 2D materials are usually exposed during device fabrication steps [50, 51], i.e. dielectric deposition, metal evaporation or post-deposition annealing. To further mimic typical device fabrication flows, in which the sample is subjected to multiple heatings during the different processing steps, the annealing (cumulatively 10 h long) was interrupted several times (i.e. after 30 min, 1 h, 3 h and 10 h) and the sample characterized after each step by means of SEM, Raman and PL spectroscopies, which have been suggested as inline metrology for TMD-Si BEOL flow [27] (details in the Experimental section). This intermediate characterization, indeed, is often carried out to monitor the quality of the material and of the assembling device during its fabrication. In figure 2 the characterization of WS₂ after sequential annealing steps is reported. From SEM micrographs (figures 2(a)–(c)) a change of shape of the crystal is visible already after 3 h of annealing (i.e. 30' + 30' + 2 h), suggesting that the irregular edges of the as-grown WS₂ (figure 2(a)) are more reactive than the center of the single-crystal, in agreement with previous results [47, 67]. After 10 h annealing (i.e. 30' + 30' + 2 h + 7 h) (figure 2(c)) small triangular holes are also visible both on the edges and in inner areas of the crystal (white triangles), indicating loss of WS₂ material. Additional SEM micrographs taken on the same sample after the fifth and sixth annealing steps, for a total of 15 and 20 h, are reported in figures S2(a) and (b). Triangular holes, already visible on the edge of the crystal after 10 h annealing, become increasingly larger (i.e. lateral size up to 500 nm) and denser on the entire crystal, with a decrease of the overall area of WS₂ monolayer from ∼81 μm² (as grown) down to ∼74 μm² (20 h annealing).

Zoom In Zoom Out Reset image size

Figure 2(d) shows the evolution of the Raman spectrum after multiple annealing steps from 30 min up to 10 h, carried out through four sequential steps. The annealing in these conditions mainly affects the intensity of the 2LA(M) + E_2g(Γ) Raman mode: as shown in figure 2(d), after a first slight increase (30 min), the 2LA(M) + E_2g(Γ) intensity slowly decreases for the subsequent steps. Raman spectra taken after additional annealing steps of 15 and 20 cumulative hours are reported in figure S2(c), in which a large decrease of the 2LA(M) + E_2g(Γ) peak intensity is visible, down to a minimum of 52% of the starting value (i.e. after 20 h). The decrease of the intensity of the 2LA(M) + E_2g(Γ) mode for annealings longer than 1 h is related to the formation of holes in the WS₂ crystal (figures 2(a)–(c)) and therefore to a loss of the overall quantity of material analyzed under the laser spot. The PL spectra for each annealing step are reported in figure 2(e). The PL peak related to the WS₂ A-exciton before annealing lies at (630.7 ± 0.1) nm (1.96 eV) with a full width at half maximum (FWHM) of (14 ± 1) nm. Upon annealing, the PL intensity rapidly decreases and becomes negligible after 15 h at 300 °C (figure S2(d)). After the first annealing step, the PL blue shifts down to (623 ± 0.2) nm (1.99 eV), while with increasing annealing time, it red-shifts up to (644.2 ± 0.2) nm (1.93 eV) with a concurrent broadening (FWHM (29 ± 1) nm). The first blue-shift of PL can be attributed to residual contaminations deposited on the growth substrate desorbing during the first annealing step. The subsequent broadening and red shift of the PL peak is likely due to a higher population of point defects such as sulfur vacancies (e.g. at the edge of the triangular holes), a hypothesis that is supported by several works [65, 67, 68]. The PL peak broadening is also combined with an increase of peak asymmetry. This asymmetry is due to the presence of an additional luminescence peak at about 650 nm, which is related to sulfur vacancy bound excitons [65, 66].

From these preliminary results, one might conclude that monolayer WS₂ is not stable at 300 °C in medium-vacuum and that its properties rapidly degrade after the first few annealing steps. However, this large instability at relatively low temperature and pressure is in contrast with both the melting and the dissociation temperatures of bulk WS₂, which are around 1250 °C [69] and 1040 °C [70], respectively. Also, such degradation is more dramatic than that reported by Rong et al [46], who observed that after 20 min of annealing at 380 °C in air, only highly defective areas of polycrystalline WS₂ start to degrade. The fast degradation of monolayer single-crystal WS₂ reported in figure 2 might then have been accelerated by external events and can have different origins: (a) the generation of defects upon exposure to laser and/or electron beam during the intermediate characterization analyses; (b) the presence of a relatively high concentration of residual oxide species in the chamber during the annealing at the vacuum level employed; (c) the exposure to air after each annealing step.

To shed light on this, each of the possible degradation causes was systematically studied. First, the role of intermediate characterization after each annealing step was evaluated by analyzing two different areas, namely 1 and 2, of a fresh WS₂ sample using SEM, Raman and PL spectroscopies, respectively. In figure 3(a) (area 1) two WS₂ crystals (black arrows) were imaged using only SEM with e-beam current 12 pA (specimen current <0.5 pA); otherwise the sample was left in dark. The WS₂ crystal reported in figure 3(c) (area 2, white arrow) was additionally characterized using Raman and PL spectroscopies with laser power set to 730 μW (fluence 9.4 mJ μm⁻²), as in figure 2. The values of both e-beam current for SEM imaging and laser fluence for Raman and PL spectroscopies were chosen to be compatible with typical values used to characterize TMDs on insulating substrate, i.e. sapphire. Higher e-beam current would lead to charging of the sample during SEM analysis, while higher laser fluence can damage the exposed material. The sample was then annealed at 300 °C and 10⁻³ mbar for 10 h in a continuative manner. While no difference is visible in the morphology of the crystal in area 1 before and after annealing (figures 3(a) and (b)), a large morphological degradation is observed for the WS₂ crystal in area 2, which was exposed to a laser power of 730 μW (white arrow, figures 3(c) and (d)). Moreover, both Raman and PL signals of the analyzed crystal in area 2 largely decrease after the continuative annealing at 300 °C (figures 3(e) and (f)). These results suggest that the values of e-beam current typically used for SEM on sapphire preserve the intrinsic stability of WS₂ at 300 °C and medium-vacuum (10⁻³ mbar), even if heated for 10 continuative hours. Conversely, the previous exposure of WS₂ single-crystal to a focused laser (∼1 μm) with fluence 9.4 mJ μm⁻² appears to trigger the instability of WS₂, which then degrades upon annealing.

Zoom In Zoom Out Reset image size

To further investigate the relation between laser fluence and thermal instability of WS₂, an additional area (area 3) was characterized using Raman and PL spectroscopies with laser power set to 73 μW (fluence 0.9 mJ μm⁻²) prior the annealing at 300 °C and 10⁻³ mbar for 10 h (figures 3(g)–(j)). Panels 3(g)–(j) report the SEM micrographs and the Raman and PL spectra of WS₂ exposed to laser power of 73 μW (fluence 0.9 mJ μm⁻²), before and after annealing: no clear difference in the morphology nor in Raman or PL signal is visible. These results indicate that WS₂ is intrinsically stable at 300 °C in medium-vacuum as long as the degradation is not triggered by the exposure to laser with a fluence higher than a certain threshold (9.4 mJ μm⁻²). On the contrary, below this threshold, the thermal stability is preserved.

It is interesting to note that, even when the degradation of WS₂ is triggered by exposing the single-crystal to a relatively high-power laser, this phenomenon is localized only to the exposed area, as shown in figure 3(d). After the annealing, the previously exposed crystal (white arrow) is nearly destroyed, while a small WS₂ crystal next to it (figure 3(d), up left corner) appears to be fully preserved. These results also rule out that oxidative species in the chamber might be responsible for the degradation observed in figure 2 (hypothesis (b)), as both Raman and PL of WS₂ (figures 3(i) and (j)) are unaffected after a 10 h annealing in the chamber. It is also worth mentioning that the different levels of degradation for WS₂, exposed to a laser power of 730 μW (fluence 9.4 mJ μm⁻²) and annealed for 10 h in a multiple (figure 2(c)) and continuative (figure 3(d)) manner, are expected to be related only to the different mapping methods adopted in the two experiments (see figure S3 for further details).

In figure 4, the role of the pressure in the chamber is taken under consideration by annealing WS₂ at 10⁻⁸ mbar (high-vacuum) for 10 h: again, no large difference after the annealing is observed if the sample is only analyzed by means of SEM with e-beam current set to 12 pA (panels (a) and (b)), as typically used for insulating substrates, and laser fluence of 0.9 mJ μm⁻² (panels (c) and (d)). Both Raman and PL spectra, reported in panels (c) and (d) respectively, confirm that using low laser fluence is safe for WS₂: both structural and optical properties of WS₂ are seemingly preserved after the annealing in high-vacuum. Indeed, although the PL intensity in panel (d) slightly decreases after annealing, the FWHM of the WS₂ A-exciton peak before and after the thermal treatment remains ∼12 nm with no redshift, indicating that the optical properties of WS₂ are not considerably degraded by the treatment. As the annealing was performed in a chamber adopted for XPS analysis, it was also possible to perform chemical characterization of the as-grown sample and monitor the effect that such a long annealing (i.e. 10 h) has on WS₂ chemical composition. The plots in figure 4(e) show the high-resolution spectra, after background subtraction, acquired within the BE window of tungsten W 4f and sulfur S 2p core electrons, before and after the continuative high-vacuum annealing, respectively. In order to study the chemistry of the sample surface, a deconvolution of the experimental curves was performed. The tungsten spectrum of the as-grown sample can be fitted as the superposition of three doublets, with the W 4f_7/2 components at (32.1 ± 0.1) eV (yellow), (32.9 ± 0.1) eV (light-blue) and (36.0 ± 0.1) eV (pink), respectively. The W 5p_3/2 peaks are shifted 5.8 eV above the corresponding W 4f_7/2. Both the yellow and light-blue W 4f doublets at lower BE have their counterpart in the sulfur spectrum, which shows two doublets with S 2p_3/2 components at (161.7 ± 0.1) eV and (162.5 ± 0.1) eV. The yellow and light-blue peaks are ascribed to atoms belonging to WS₂ crystals, while the pink doublet was ascribed to residuals of unreacted WO₃ precursor. In particular, the energy position of the light-blue subcomponent is in perfect agreement with the one reported in literature for the hexagonal WS₂ semiconducting phase (2H), while the lower BE of about 0.8 eV related to the yellow minor subcomponent can be a fingerprint of the 1T-WS₂ metallic phase, sulfur vacancies or other kinds of defects due to CVD growth process [45, 71, 72]. Interestingly, the spectra acquired after 10 h continuative annealing at 300 °C in high-vacuum show almost unaltered yellow and light-blue components. The only significant change in the W spectrum is related to the WO₃ doublet (due to unreacted material on the sapphire surface) [37], which shows a decreased contribution to the W 4f envelope in comparison to the pre-annealing data. Moreover, a new W-related doublet (green) appears at (33.6 ± 0.1) eV, which does not have a counterpart in the S 2p spectrum acquired after the annealing. The green component can be ascribed to a chemical evolution into lower valence oxides of the unreacted WO₃ [73] as a consequence of the high-vacuum annealing, which, however, does not have any sizeable effect on the thermal stability of WS₂ crystals.

The absence of modifications in the surface chemistry of WS₂ crystals, confirmed by XPS results, is also particularly important since desorption of sulfur in a BEOL could cause line contamination.

To investigate whether the exposure of the sample to air after each annealing step (listed as cause (c)) might contribute to the fast degradation reported in figure 2, a fresh sample was annealed at 300 °C in medium-vacuum in two subsequent steps of 10 h each and analyzed using the safe conditions found in figure 3. No clear difference is observed from the SEM, Raman and PL analyses on the same crystal (figure S4), confirming that, if no oxidation is triggered by a relatively high-power laser, WS₂ is intrinsically stable at 300 °C and medium-vacuum even when exposing the sample to air in-between multiple annealings.

Finally, the thermal stability of WS₂ at high temperatures was also investigated by annealing a fresh sample of WS₂ at 600 °C and 10⁻³ mbar. Figure S5 shows that at this annealing temperature WS₂ becomes unstable and completely degrades after 1 h. The XPS spectra of a sample after 2 h annealing under the abovementioned conditions show a complete absence of WS₂ related components and that traces of tungsten oxides are the only remaining species (figure S5(e)).

The results reported in this work indicate that monolayer single-crystal WS₂ is intrinsically stable at 300 °C, a typical temperature adopted in BEOL fabrications steps, even for 10 h. However, care must be taken in order to preserve its thermal stability and the appropriate conditions for handling the sample between each step must be carefully chosen. A schematic summary of the conditions investigated in this work is reported in figure 5. If a base pressure of 10⁻³ mbar (or lower) is used when annealing at 300 °C, WS₂ does not degrade: the vacuum level is low enough to avoid fast oxidation of WS₂ at a relatively high temperature.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The inline monitoring of the sample in-between annealing steps needs to be carefully performed. In fact, monolayer WS₂ remains stable at 300 °C and 10⁻³ mbar as long as the degradation is not triggered by exposure to a focused light with relatively high power, in agreement with previous reports [56, 57]. The exposure to laser above a certain threshold either generates or activates defects, i.e. sulfur vacancies, which then rapidly react at high temperature with oxidative species that are still present in the annealing chamber at a pressure of 10⁻³ mbar. However, if the laser fluence is lowered down to ⩽0.9 mJ μm⁻² (laser power 73 μW), this photo-induced process is not activated. Atkin et al [56] reported a fluence threshold of defect photo-activation for WS₂ between 2 and 20 mJ μm⁻² at room temperature [56]. They suggest that if the laser exposure is energetic enough the top sulfur layer of WS₂ is affected, generating localized defects (i.e. contaminant adsorption), and becomes more susceptible to deterioration; below the threshold, instead, the reaction is not activated. Our experiments indicate that carrying out analysis with a light fluence of 0.9 mJ μm⁻² is safe for monolayer WS₂, even when annealing at 300 °C and in medium-vacuum. Moreover, multiple exposures to the laser with light fluence of 0.9 mJ μm⁻² and different combinations of laser power and exposure time (i.e. laser power 730 μW, exposure time 1 s, fluence 0.9 mJ μm⁻²) were taken into consideration and no major difference was observed.

On the contrary, SEM analysis with an e-beam current of 12 pA does not induce degradation of WS₂ single-crystals. A relation between e-beam current and defect activation of WS₂ might be expected, similarly to the laser fluence. In perspective of BEOL integration on a Si platform, particular care should be adopted and the effect of higher electron currents on monolayer WS₂ should be investigated.

Finally, the effect of ambient exposure in-between processing steps was investigated by annealing WS₂ at 300 °C in two subsequent steps of 10 h each. It was found that WS₂ is compatible with sequential fabrication steps at this temperature and in medium-vacuum.