**1. Introduction**

In recent years, layered materials such as two-dimensional (2D) transition metal dichalcogenides (TMDs) have attracted a high level of interest due to their features, which make them appealing for potential applications in gas sensing [1,2], photo-electro-catalytic hydrogen evolution [3,4] optical and electronic devices [5,6] and energy storage [7,8].

Few-layer 2D TMDs can be produced through di fferent methodologies like mechanical and liquid phase exfoliation of bulk crystals, classified as top-down routes, or via direct bottom-up routes like chemical vapor deposition [9,10]. High-yield liquid exfoliation methods comprising ion intercalation [11,12] and ultrasonic cleavage [13,14] have also been widely employed to exfoliate bulk-layered materials. Besides liquid phase exfoliation, low energy ball milling as a newly explored high-yield mechanical exfoliation method has been utilized for scalable production of mono and few-layer graphene [15,16] and TMDs nano-sheets [17,18]. More recently, enhanced mixed methods comprising assisted grinding and sonication have been shown to produce higher concentrations of TMDs nano-sheets and a reduced amount of defects [19].

Regarding gas sensing applications, 2D single layer MoS2 Field E ffect Transistor [20,21] Pd-doped WS2 films [22] have been shown to be alternative substitutes for traditional metal oxides sensors. Moreover, considering that the possibility to find practical applications of those materials is generally dependent on the reproducibility of the preparation with respect to both microstructure (i.e., number of layers, lateral size, surface area, etc.) and chemical composition (i.e., defects concentration and surface oxidation), the need to find a practical, high-reproducible, and easy way to exfoliate TMDs is always under investigation. We already presented the fabrication of chemo-resistive thin films gas sensor, by drop casting suspensions of few flakes graphene oxide [23], phosphorene [24] and more recently TMDs utilizing both liquid-exfoliated MoS2 [25] and commercially exfoliated suspensions of WS2 [26]. The aim of this work is to apply a low energy ball milling and sonication method to achieve a reproducible and high-yield exfoliation methodology and to test the gas sensing performances of the obtained material. Starting from commercial WS2 powders, we have firstly performed the exfoliation process based on the grinding and sonication method and investigated the morphology of few flakes WS2. Secondly, we have determined that with exposing the material to mild air annealing at 180 ◦C, a controlled partial oxidation of WS2 flakes to amorphous WO3 is achievable. Lastly, we investigated the sensing responses to NO2, H2 and humidity of drop casted exfoliated WS2 chemo-resistive thin films, discussing the likely gas-response mechanism.

#### **2. Materials and Methods**

*WS2 exfoliation*: According to the flow sheet shown in Supplementary Figure S1, 2 g of WS2 commercial powder (Sigma–Aldrich 243639-50G, St. Louis, MO, USA) were dispersed in 4 mL Acetonitrile (ACN—VWR 83639.320, Radnor, PA, USA) with 30 g Zirconium Oxide balls (D = 3 mm), and ball milled in a planetary milling machine (Fritsch—Planetary Micro Mill Pulverisette 7, Idar-Oberstein, Germany) at 400 rpm, for 2 h in ambient air.

To evaporate ACN residuals after milling, the collected slurry was left overnight at 23 ± 2 ◦C temperature and 40% ± 3% Relative Humidity (RH) (ATP DT-625 High Accuracy Thermo-hygrometer, Ashby-de-la-Zouch, UK). After ACN evaporation, 0.05 g of the dried powder was dispersed in 100 mL of pure ethanol (99.94% VWR 20821.330, Radnor, PA, USA) and probe sonicated (Sonics VC 505 ultrasonic processor, Newtown, CT, USA) at 250 W for 90 min in a thermostat bath to prevent temperature rise (T 25 ◦C). Finally, the solution was centrifuged at 2500 rpm for 40 min in a refrigerated (20 ± 2 ◦C) micro-centrifuge (Eppendorf 5417R, Hamburg, Germany) and the supernatant collected.

*Microstructural and chemical characterization*: Air tapping mode Atomic Force Microscopy (AFM) was performed with a Veeco Digital D5000 system. Using silicon tips with spring constant of 3 N·m<sup>−</sup><sup>1</sup> and resonance frequencies between 51 and 94 kHz. Samples for AFM investigations were prepared via spinning (at 2000 rpm for 30 s) 10 μL of centrifuged WS2/Ethanol solution on a Si3N4 substrate. The substrates have been previously cleaned in a piranha base solution (3:1:3 mixture of ammonium hydroxide NH4OH with hydrogen peroxide and milli-Q water) to enhance their wettability.

Exfoliated flakes were investigated using High Resolution Transmission Electron Microscopy HRTEM—JEOL 2100 Field Emission electron microscope (Tokyo, Japan) operated at 200 kV. Samples prepared by drop casting the WS2/Eth solution on Si3N4 substrate were investigated by X-Ray Photoelectron Spectroscopy (XPS) using a PHI 1257 spectrometer (Perkin Elmer, Norwalk, CT, USA) equipped with a monochromatic Al K α source (hν = 1486.6 eV) with a pass energy of 11.75 eV (93.9 eV survey), corresponding to an overall experimental resolution of 0.25 eV. Raman spectra were acquired using a Micro Raman Spectrometer (μRS) (LABRAM spectrometer, λ = 633 nm, 1 μm spatial resolution, and ≈2 cm<sup>−</sup><sup>1</sup> spectral resolution, Horiba-Jobin Yvon, Kyoto, Japan) equipped with a confocal optical microscope (100 × MPLAN objective with 0.9 numerical aperture and 0.15 mm work distance). 10 μL of the WS2/Eth solution was deposited on a 270 nm SiO2 substrate.

*Gas sensing measurements*: Electrical properties were determined by a volt-amperometric technique (AGILENT 34970A), as reported in Supplementary Figure S2, utilizing WS2 thin films prepared by multiple drop casting and air annealing at 180 ◦C for 1 h the centrifuged WS2/Eth suspension on Si3N4

substrates provided with 30 μm-spaced Pt interdigitated electrodes on the front side and a Pt resistor acting as a heater on the back side. Different gas concentrations in the range 1 ppm–250 ppm H2 and 40 ppb–5 ppm NO2 were obtained by mixing certified H2, and NO2 mixtures with dry air carrier, by means of an MKS147 multi gas mass controller. Different relative humidity (RH) air streams in the 10–80% RH range were obtained by mixing dry with saturated water-vapor air. The following definitions apply to discuss the gas response properties: base line resistance (BLR): the resistance in dry air at equilibrium before gas exposure, relative response (RR): the ratio (RA/RG) where RA represents the resistance in air and RG the one in gas at equilibrium for a given gas concentration, and sensor sensitivity (S): is the slope of the calibration curve in the sensitivity plot.

#### **3. Results and Discussion**

#### *3.1. Microstructural Properties of Exfoliated WS2*

The microstructure of exfoliated WS2 obtained by the combined ball milling and sonication process is characterized. In our case, acetonitrile (ACN) as the milling solvent, with surface tension of 29.5 mJ m −2 [27], has been selected as a trade-off between surface tension and moderate boiling point, enabling complete removal of the solvent at room temperature after grinding.

Regarding the influence of the grinding time, particles' size distribution of the starting WS2 powder (blue plot), determined by Dynamic Light Scattering (DLS) technique (see experimental section) and shown in Figure 1a, downshifts towards smaller average sizes after 72 h grinding (red plot) and slightly further after 90 min sonication (green plot). The particle size distribution of the WS2 starting powder displays an average particle size of ≈ 8 μm (blue plot) whereas the 72 h ball milled shows a bimodal distribution (red plot), with larger aggregates centered at ≈ 20 μm and smaller ones at ≈0.7 μm. After 90 min sonication, the bimodal distribution of the grinded powder disappears (green plot) and the average particle dimension places at ≈ 0.6 μm.

It may be concluded that grinding has an effective influence to reduce the particle size, while sonication, beside its effectiveness to suppress the bimodal distribution (presumably by separating agglomerated WS2 particles), shows only minor effects to further decrease the particle size of the grinded powder, confirming the dominant role of the milling step. Figure 1b shows the AFM image of the 72 h ball milled and 90 min sonicated WS2 sample. The inset of this image depicts a rough thickness profile along the selected line, with an average height from the substrate of 2 nm. Low and high magnification TEM images illustrated in Figure 1c,d, show that long term ball milling for 72 h results in a fragmented structure, which was eventually revealed to be amorphous by fast Fourier electron diffraction measurements. These features can be explained considering the two main forces induced by ball milling. The primary force is the shear force provided by rolling of balls on the surface of layers, which causes the removal and the exfoliation of surface layers. The secondary force is the vertical impact from the balls which combined with longer grinding times can fragment the larger exfoliated sheets into smaller ones, eventually collapsing of the crystal structure [15,28].

With the aim to minimize the fragmentation effect, we have reduced the ball milling duration time from 72 to 2 h, maintaining the sonication time at 90 min. Decreasing the milling time to 2 h, the flake's fragmentation sharply decreases, enforcing the formation of well-defined terraced structures comprising stacked WS2 flakes, as shown in Figure 1e,f.

Figure 2 shows the main microstructural features of the 2 h ball milled and 90 min sonicated WS2 powders. The AFM image shown in Figure 2a depicts the formation of a well-shaped 2D-flake with a large flat surface of 1 μm length. The corresponding thickness profile drawn in Figure 2b highlights a clear formation of a stacked structure comprising a 3 nm thick basal plane, 6 nm thick secondary plane and a third one at the top. Considering that the slight step on top of the profile, 0.6 nm high, corresponds to 1 layer thickness WS2 [29], it is shown that the first step is made of 5 layers and the second one of 10 layers, respectively. The 2D character of the actual stacked structure shown in Figure 2b, as defined by the "aspect ratio" (i.e., the ratio of lateral dimension to the thickness), is

high, with an associated value of 250, attesting to the successful optimization of the grinding time for exfoliation. The low-resolution TEM image depicted in Figure 2c also illustrates a well-shaped exfoliated flake with edge angles of 120◦, confirming the preservation of the typical crystalline WS2 hexagonal geometry.

**Figure 1.** Comparison of the effect of long-time ball milling on WS2 exfoliation: (**a**) Comparison of the particle size distribution of the starting WS2 commercial powder (blue), 72 h ball milled (red) and 72 h ball milled and 90 min sonicated (green), (**b**) AFM picture of 72 h ball milled and 90 min sonicated and associated thickness profile along the white line, (**c**) TEM picture of the 72 h ball milled and 90 min sonicated WS2, (**d**) Comparison of the particle size distribution in case of 2 h ball milling, (**e**) AFM picture of 2 h ball milled and 90 min sonicated and associated thickness profile, (**f**) TEM picture of the flakes obtained by 2 h ball milling and 90 min sonication of WS2.

Figure 3 shows HRTEM images of both edges and surfaces of the flakes. Figure 3b (the magnification of Figure 3a), exhibits two layers with associated interlayer distances of ≈ 0.63 nm which is in good agreemen<sup>t</sup> with the AFM thickness measurements illustrated in Figure 2b. This interlayer displacement could also be observed at the flake's edges depicted in Figure 3d, where almost 11 layers can be clearly counted on the 7 nm thick edge. The atoms arrangemen<sup>t</sup> displayed in Figure 3b exhibits the hexagonal atomic structure, with lattice spacing of 0.27 nm and 0.25 nm, that are characteristics of (100) and (101) crystal planes of 2H-WS2 flakes, respectively [30,31]. Moreover, the Fast Fourier Transforms (FFTs) shown as the inset of Figure 3d, further confirms the hexagonal crystalline structure of the flake.

**Figure 2.** (**a**) AFM picture of the 2 h ball milled and 90 min sonicated WS2, (**b**) thickness profile of the stacked flake along the white line of Figure (**a**), (**c**) low-resolution TEM picture of an exfoliated flake.

**Figure 3.** (**<sup>a</sup>**,**<sup>c</sup>**) TEM of 2 h ball milled and 90 min sonicated WS2, (**b**) HRTEM corresponding to the circled area shown in figure (**a**) with highlighted the interlayer distance (0.63 nm) and lattice spacing (0.27 nm and 0.25 nm), corresponding to (100) and (101) planes of WS2 respectively, (**d**) HRTEM of the edge of the flake corresponding to the circled area shown in figure (**c**) with highlighted the 7 nm thick edge corresponding to 11 layers. The inset shows the Selected Area Electron Diffraction (SAED) of the flake.

To give a statistical insight of the reproducibility of the preparation, four different suspensions were prepared after 2 h milling and 90 min sonication and the corresponding centrifuged suspensions collected, and spin coated on Si3N4 substrates (see experimental section). Figure 4a–d shows the AFM images of each prepared sample covering an area of 10 × 10 μm<sup>2</sup> and corresponding to a total population of ≈ 220 flakes. Overall, flakes' thickness follows a log-normal distribution as shown in Figure 4e, indicating that almost 30% of the flakes are ≤3.0 nm thick (i.e., ≈ 5 layers) and that about

75% are ≤6 nm (i.e., ≈ 10 layers). Moreover, as displayed in Figure 4f, average flake lateral dimensions are approximately ≈ 110 nm, yielding a surface coverage of ≈ 6%, as shown in Figure 4g. The overall calculated "aspect ratio" is 27.5, which is comparable to the ones previously reported for MoS2 and WS2, given the same preparation methodology [27,32]. The reduced standard deviations shown in Figure 4f,g attest to the high reproducibility of the exfoliation process.

**Figure 4.** (**<sup>a</sup>**–**d**) AFM images of WS2 exfoliated corresponding to four different samples prepared under the same conditions (i.e., 2 h ball milling and 90 min sonication). Statistical analysis corresponding to thickness distribution (**e**), Lateral dimensions (**f**) and surface area coverage (**g**).

#### *3.2. Chemical Composition of the Exfoliated WS2*

Chemical issues related to both the evolution of point defects and oxidation phenomena of TMDs highlight important challenges associated with the practical utilization of TMDs monolayers in electronic and optoelectronic devices. Sulphur vacancy is one of the most typical point defects in 2D MoS2 and WS2 monolayers [33], eventually leading to active sites for gas adsorption. On the other hand, spontaneous oxidation of metal sulphides into their metal oxide counterparts [34], may result in poor reproducibility of the gas-sensing response over the long-term, as discussed in our previous publication [26].

Figure 5, panel (a), reports the W 4f core level XPS spectra of the pristine commercial WS2 powder (PWD). The two doublets corresponding to 4f7/2 peaks are assigned, according to the literature, to WS2

and WO3, respectively [35,36]. It turns out that the pristine powder is already oxidized, with a WO3 content of ≈ 18%. This phenomenon, is not surprising considering that spontaneous oxidation of MoS2 at room temperature after 6–12 months has been already reported in the literature, demonstrating the occurrence of ageing phenomena of MoS2 [25] and WS2 monolayers in ambient air [26]. Figure 5b shows that after exfoliation, the WO3 content decreases to ≈ 16%. This could be explained considering that by grinding and sonicating WS2 powders, newly not-yet-oxidized surfaces are formed, resulting in a smaller content of WO3, as evidenced by XPS measurements. Furthermore, as discussed in the next paragraph, given the optimum operating temperature for gas sensing at 150 ◦C, exfoliated WS2 suspensions, were therefore drop-casted and pre-annealed at 180 ◦C for 1 h in air to stabilize the oxidation levels. As shown in Figure 5c, after annealing, the WO3 content increased to ≈ 58%.

**Figure 5.** X-Ray Photoemission Spectroscopy (XPS) spectra of W 4f core level acquired respectively on (**a**) pristine WS2 commercial powder (WS2 PWD), (**b**) Exfoliated WS2 by ball milling and sonication at 25 ◦C, (**c**) WS2 exfoliated and post-annealed at 180 ◦C. All the components and their relative atomic percentages are labelled in the figure.

Regarding sulphur vacancies formation, considering 1% the detection limit of the XPS measurement, we found no clear evidence of defects of sulfur vacancies-related components (typically at binding energies of ≈ 36.1 in the W 4f core level XPS spectra in Figure 5). The analysis of the S 2p core level XPS spectra, reported in Supplementary Figure S3, is in line with the analysis reported for the W 4f.

Figure 6 displays the Raman spectra of bulk powder, exfoliated flakes and 180 ◦C annealed film. Peaks located at 350 and 419 cm<sup>−</sup><sup>1</sup> refer to crystalline WS2. The peak at 520 cm<sup>−</sup><sup>1</sup> corresponds to the substrate (i.e., crystalline SiO2). Raman spectra reveal that neither crystalline nor amorphous WO3 are formed. According to the literature [37], the displacement of crystalline WO3 is excluded considering that no peaks corresponding to the dashed lines at 719 and 807 cm<sup>−</sup><sup>1</sup> are observed. No signals of amorphous WO3 were shown, as attested by the absence of a broad peak between 600 and 900 cm<sup>−</sup>1, attributed to the W-O stretching vibration of amorphous WO3. The lack from the Raman spectra of any WO3 signal, as opposed to XPS, can be explained considering that Raman spectroscopy penetrates more deeply inside the material, suggesting that the overall amount of WO3 throughout the whole flakes is negligible.

**Figure 6.** Raman spectra of WS2 bulk powder, WS2 as-exfoliated and WS2 flakes post-annealed at 180 ◦C.

In order to have a better understanding where the as-formed WO3 is located, firstly we have to consider that both XPS and Raman techniques give information on the chemical bonding of the elements, secondly that XPS information comes from photoelectrons escaping maximum up to 10 nm below the material surface, lastly that Raman spectroscopy, compared to XPS, is a "bulk" technique (given the negligible attenuation of visible light at the length scale of microns). It turns out that the WS2/WO3 percentage content measured by XPS and shown in Figure 4 (i.e., ≈ 58%), represents the average chemical compositions of a portion of the material confined within at last 10 nm from the material surface. This region, for simplicity, can be referred as a "surface layer", which represents the reacting surface to interfering gases. Regarding the crystallographic nature of the "surface layer", grazing incidence XRD diffraction carried on exfoliated and 200 ◦C annealed WS2 flakes (shown in Supplementary Figure S4) revealed the occurrence of peaks belonging only to WS2, thus excluding the formation of any crystalline WO3 in the 180 ◦C annealed film. It may be concluded that the "surface layer" is a composite structure comprising amorphous WO3 and pristine crystalline WS2. These experimental results are in line with that previously discussed in the literature [38], that the oxidation of bulk TMDs provides two parallel steps. Oxygen atoms rapidly exchange with surface sulphur forming an amorphous oxide layer, whilst WS2 interlayer channels provide a path for inward-oxygen and backward-sulphur diffusion, resulting in the formation of amorphous WO3, which propagates over time inside the TMD flakes.

#### *3.3. Gas Sensing Response*

It has been reported that TMDs gas sensors operating at room temperature have shown remarkable limitations, largely related to irreversible desorption of the gas molecules, displaying incomplete recovery of the baseline at 25 ◦C [39,40]. The selection of the best operating temperature (OT), which it corresponds the complete base line recovery, within reasonable response times, and acceptable gas relative responses (RR), is limited in TMDs by the intensifying of the oxidation processes with increasing the OT, as previously discussed. Baseline recovery and response times depend on the adsorption-desorption kinetic of gases with the sensor surface, which eventually improves with increasing the OT. RR is mostly related to microstructure (i.e., surface area, grain size), concentration of surface defects (i.e., oxygen or sulphur vacancies) and structure of the reacting surface (crystalline or amorphous). To find the best OT exfoliated WS2 drop casted on Si3N4, substrates were previously air annealed at 180 ◦C to stabilize the WO3 content up to ≈ 58%, as shown in Figure 5c. Afterwards, different OT in the 50–150 ◦C temperature range where tested, resulting in 150 ◦C as the best OT, as shown in Supplementary Figure S5.

The Scanning Electron Microscopy (SEM) picture of the sensing device shown in Figure 7 highlights a homogeneous distribution of annealed flakes, enabling current percolation paths between adjacent flakes, covering an area of 1.4 × 0.6 mm<sup>2</sup> over 30 μm spaced Pt interdigitated electrodes.

**Figure 7.** SEM image of sensor obtained by drop casting exfoliated WS2 and annealing at 180 ◦C on Si3N4 substrate provided with Pt finger-type electrodes (30 microns apart).

Figure 8a shows the normalized dynamic resistance changes, at 150 ◦C OT of few-layers WS2 thin films to NO2 and H2 in the 100 ppb–5 ppm and 1 ppm–250 ppm gas concentration ranges, respectively. WS2 films respond as *n*-type semiconductors with decreasing/increasing resistance upon exposure to H2/NO2, respectively. Degassing with dry air at 150 ◦C OT, the baseline resistance (BLR), as indicated by the black dotted line in the figure, is almost recovered. WS2 flakes are more sensitive to NO2 than H2 gas, with associated low detection limits (LDL) of 200 ppb and 5 ppm respectively, confirming what has been previously reported in the literature.

**Figure 8.** Electrical responses of the exfoliated WS2 post-annealed at 180 ◦C, at 150 ◦C operating temperature in dry air. (**a**) Comparison of the normalized dynamic response to NO2 (100 ppb–5 ppm) and H2 (1–250 ppm), (**b**) NO2 cross-sensitivity to H2: first panel, the response to 120 ppm H2 in dry air, second panel, response to 120 ppm H2 with 600 ppb NO2, third panel, response to 120 ppm H2 (as to first panel) for comparison, (**c**) Reproducibility and baseline recovery by exposing the film to both pulse and cumulative H2 concentrations in the range 40–100 ppm. H2 concentrations are highlighted in the figure by grey shadowed rectangular plots.

Cross-sensitivity, which represent the ability of WS2 to detect H2 in the presence of NO2 interfering gas, has been shown in Figure 8b. Panel (1) of Figure 8b shows the WS2 response to H2 alone in dry air carrier at 150 ◦C OT. The "cross-sensitivity" produced by interfering NO2 to the measure of H2 is displayed in panel (2) of Figure 8b. By exposing to 600 ppb NO2, sensor resistance initially increases, yielding at equilibrium the resistance value RNO2. As soon as 120 ppm H2 is introduced, the resistance decreases, yielding the equilibrium value shown as R(H2+NO2). The cross-sensitivity e ffect is displayed in the picture as ( ΔR), to indicate the gap between the electrical resistance to 120 ppm H2 alone (i.e., RH2) and that in the presence of 120 ppm H2 and 600 ppb NO2 (i.e., R(H2+NO2)). These results imply that the response to 120 ppm H2 is a ffected by the presence of a small amount (i.e., 600 ppb), of NO2 interfering gas, confirming the stronger a ffinity of WS2 to detect NO2 as respect to H2.

Reproducibility of the electrical response to pulse (on/off) and cumulative modes H2 gas adsorption is shown in Figure 8c, demonstrating acceptable response characteristics to H2. Under pulsed conditions, the baseline resistance (BLR) fully regains its initial value after completion of each desorption cycle in dry air. Under cumulative stepwise adsorption/desorption mode, the H2 gas resistance increases/decreases steadily, matching almost the same H2 resistance values obtained under pulsed conditions (black lines at saturation correspond to 40, 80 and 100 ppm H2).

Selectivity tests to both oxidizing and reducing gases carried out at 150 ◦C operating temperature with respect to 5 ppm NO2, exhibit satisfactory WS2 selectivity to both 5 ppm H2 and NH3 gases and to 250 ppm ethanol and acetone, as shown in Supplementary Figure S6. These data were demonstrated to be in line with previous research on WS2 nanoflakes synthesized by electrospinning [41].

Long-term stability properties of the electrical response of both baseline and saturation resistances to 800 ppb NO2 over a period of 12 months (corresponding to approximately 5 months of cumulative operations at 150 ◦C operating temperature) were also recorded. Figure 9a shows baseline resistances (lower curve) and saturation resistances corresponding to 800 ppb NO2 (upper curve), randomly collected over a period of 52 weeks. Average resistances with associated standard deviations are calculated over a set of 5 consecutive measurements. Relative responses (RR) taken over the investigated period are also highlighted in the figure. No remarkable fluctuations of both baseline and resistances at saturation were detectable, attesting good long-term stability of the electrical properties of the WS2 films. To validate the electrical responses shown in Figure 8a, we also investigated the oxidation state of the sensor surface, measuring the WO3 content before and after 52 weeks conditioning. Figure 9b compares the XPS W 4f signal of the as-exfoliated WS2 180 ◦C annealed film before (lower curve) and after 52 weeks long-term conditioning to di fferent gases and 150 ◦C OT (upper curve). Beside an increase of the signal noise after long-term conditionings, no substantial increase of the WO3 content was detected. These observations imply that exfoliated WS2 films, previously stabilized at 180 ◦C, can satisfactorily respond to di fferent gases under sustained conditions at 150 ◦C operating temperature.

Finally, the influence of humidity on NO2 and H2 gas response at 150 ◦C OT is also reported. Figure 10 shows the dynamic response of the films exposed to air with increasing amounts of humidity, in the 10–80% relative humidity (RH) range. The inset of Figure 10a displays the related sensitivity plot. Considering that at 150 ◦C operating temperature physisorbed water is reasonably evaporated, it is unlikely that the decrease of the resistance is induced by a protonic surface charge-transfer mechanism, as reported for WS2 humidity sensors operating at room temperature [42]. More reasonably, water vapor at 150 ◦C OT behaves like a reducing gas inducing a steady resistance decrease in WS2 *n*-type semiconductor, which does not saturate with increasing the RH, as shown in the inset of Figure 10a.

**Figure 9.** WS2 exfoliated and post-annealed at 180 ◦C. (**a**) Long-term stability properties of the electrical resistances of the baseline (lower curve) and 800 ppb NO2 over a period of 12 months (equivalent to approximately 5 months of continuous operation at 150 ◦C operating temperature). Average resistance values with associated standard deviations are calculated over a set of 5 consecutive measurements. (**b**) Comparison of the XPS signals of the as-exfoliated WS2 annealed at 180 ◦C (lower curve) and the same sample after one-year conditioning to various gases and 150 ◦C operating temperature.

In order to evaluate the influence of humidity to NO2 and H2 gases response, cross-sensitivity tests have been performed. Figure 10b, c shows the dynamic responses to different NO2 and H2 concentrations, measured in humid air at 40% RH. As soon as water vapor is introduced, a downshift of the baseline from BLRDRY to BLRH2O is shown. NO2 and H2 dynamic responses in dry and 40% RH are almost similar, as shown in Figures 8a and 10b,c. Comparison of the NO2 and H2 gases sensitivities in dry and 40% RH air are shown in Figure 10d. Notably, no significant differences are displayed as respect to slopes of the calibrating curves (i.e., sensitivity) and relative responses values (i.e., Ra/Rg), attesting that both NO2 and H2 measurements are not affected by the presence of moisture.

These results, indeed, demonstrate the possibility to produce efficient and reproducible gas sensors, able to detect NO2 and H2 with no significant cross-sensitivity effects induced by humid air in the 10–80% RH range and 150 ◦C operating temperature.

**Figure 10.** Electrical responses of the exfoliated WS2 post-annealed at 180 ◦C (150 ◦C operating temperature) to different Relative Humidity (RH) conditions. (**a**) Normalized dynamic response to humidity (10–80% RH). The inset depicts the corresponding sensitivity plot. (**b**) Dynamic response to increasing NO2 concentrations in air with 40% RH, (**c**) Dynamic response to increasing H2 concentrations in in air with 40% RH, (**d**) Comparison of the sensitivity plots to NO2 and H2 in dry air and 40% RH, respectively.

#### *3.4. Gas Response Mechanism*

As previously discussed, by annealing the exfoliated WS2 at 180 ◦C a "surface layer" containing ≈58% of amorphous WO3, penetrating at last 10 nm from the surface, is formed. It turns out that both structure and chemical composition of the "Surface layer", comprising crystalline WS2 and amorphous WO3, strongly influence the gas response mechanism.

HRTEM investigations of the pre-annealed 180 ◦C sample shown in Figure 11, display the occurrence of a complex surface patchwork comprising amorphous WO3 regions (located inside the green square of Figure 11b), possibly rearranging as not-connected, amorphous, isolated-clusters which are eventually embedded in crystalline WS2 phase (located inside the red square of Figure 11b).

To investigate the contribution of single WO3 on to the electrical properties of WS2/WO3 pre-annealed composite, we have prepared a fully-oxidized WS2, containing ≈ 99% amorphous WO3 and tested to NO2. This sample has been prepared by the same exfoliation method by grinding and drying WS2 powders at 25 ◦C, but setting the sonicating temperature at 60 ◦C, instead of 25 ◦C for 90 min, and finally pre-annealing in dry air at 180 ◦C.

Figure 12a,b shows the XPS and grazing incidence XRD patterns of the fully oxidized film, attesting that the chemical composition of the "surface layer" is ≈ 99% WO3 (Figure 12a) and that the as-formed WO3 is amorphous, as highlighted by the absence of WO3 peaks inside the inset of the XRD pattern of Figure 12b. It turns out that the fully oxidized region possibly covers the whole surface of the flakes, not extending to the core, which maintain the crystalline structure of pristine exfoliated WS2 (as attested by the presence of WS2 peaks in the XRD pattern of Figure 12b). The electrical response of the fully oxidized amorphous WO3 to 800 ppb NO2 and different operating temperatures in dry air is

shown in Figure 12c. According to Figure 12c, amorphous WO3 is not responsive to NO2 gas in dry air, in the operating temperature range 75–150 ◦C.

**Figure 11.** HRTEM images of the WS2 film pre-annealed at 180 ◦C. (**b**) Magnification of the yellow area of Figure 11 (**a**) displaying the presence of ordered structures (i.e., inside the red square) referred to crystalline WS2 and the presence of disordered ones (i.e., inside the green square) attributed to amorphous WO3. Related Selected Area Electron Diffraction (SAED) patterns are shown in (**c**) and (**d**), highlighting the occurrence of sharper reflections (**d**) associated to crystalline WS2.

Literature reports discussing the gas sensing properties of amorphous WO3 are very scarce. Some authors found no gas response to NOx of amorphous WO3 deposited by sputtering [43] whereas others demonstrated negligible NO2 response using photochemically-produced amorphous WO3 [44]. In most cases, the NO2 gas response of amorphous WO3 is smaller with respect to crystalline WO3, frequently associated with baseline drift phenomena, with few exceptions, mostly related to the preparation conditions. In our case, we demonstrated that the interaction of NO2 with amorphous 99% WO3 has no effects altogether.

Having shown in Figure 8a the substantial NO2 and H2 gas response of the WO3/WS2 composite, we conclude that it is crystalline WS2 which primarily respond to NO2 gas. The predominant gas sensing role played by crystalline WS2 with respect to amorphous WO3, is also supported by the decrease of the electrical resistance with increasing relative Humidity (RH), as shown in Figure 10a. Considering that WO3 interacts with humidity by increasing the resistance, due to WO3 lattice oxidation induced by humidity, as reported in the literature [45], the resistance decrease displayed in Figure 10a rules out any significative contribution of WO3 to the overall humidity response. Moreover, the hypothesis that WS2 is likely to be the responding material is also supported by our previous research, which demonstrated that humidity decreases the sensor resistance in MoS2-based exfoliated [25].

Discussing the contribution of crystalline WS2 to the overall electrical resistance, it was recently demonstrated by first-principles calculations on single MoS2 sulphur-defective layer that O2 irreversibly chemisorbs on sulphur vacancies [46] and that the "heal" of these defects by substitutional O atoms is thermodynamically favorable [47]. Furthermore, in case of direct NO2 molecules interaction with sulphur vacancies, a dissociative adsorption of NO2, leading to O atoms passivating the vacancies, and NO molecules physisorbed on the MoS2 surface, was also proposed [48,49]. Given these premises, we hypothesize that both O2 and NO2 suppress sulphur vacancies, supporting a gas response mechanism based only on physisorption of NO2 and H2 molecules on passivated (i.e., defect-free) WS2 surface. This hypothesis is sustained by theoretical studies on the adsorption of NO2, H2, O2, H2O, NH3 and CO gases on defect-free single layer MoS2 and WS2 [50,51]. According to this physisorption model, the size and sign of the resistance changes, when exposing few-flakes of WS2 to oxidizing (NO2) and

reducing (H2) gases, depend on the number of exchanged carriers (i.e., electrons) and their direction. NO2 being more electronegative than H2 induces a large electron withdrawal, whereas H2 results in weak electron injection, explaining the increase/decrease of electrical resistance in *n*-type WS2, as well as the smaller detection limit measured for NO2 (200 ppb) as compared to the one found for H2 (i.e., 5 ppm).

**Figure 12.** Chemical composition, crystalline structure and microstructural features of a fully oxidized WO3 thin film. (**a**) W 4f core level XPS spectra, (**b**) XRD grazing incidence spectra. Top right inset shows the close up of the 2θ region characteristic of crystalline WO3 (corresponding peaks of crystalline WO3, according to ICDS 98-001-7003, are highlighted by dashed green lines), (**c**) electrical response of the fully oxidized WO3 amorphous film to NO2 and different OTs.

Lastly, a question to be resolved is why both NO2 and H2 sensitivities are not affected by the presence of moisture, as shown in Figure 10d. This behavior may sugges<sup>t</sup> that water vapor adsorbs on to the WS2 surface with a different and non-competitive mechanism with respect to NO2 and H2 gases. Clearly this interaction is a complex issue, and is ye<sup>t</sup> to be clarified based on specific theoretical and experimental studies.
