Metal Oxide Nanoparticle-Decorated Few Layer Graphene Nanoflake Chemoresistors for the Detection of Aromatic Volatile Organic Compounds

Benzene, toluene, and xylene, commonly known as BTX, are hazardous aromatic organic vapors with high toxicity towards living organisms. Many techniques are being developed to provide the community with portable, cost effective, and high performance BTX sensing devices in order to effectively monitor the quality of air. In this paper, we study the effect of decorating graphene with tin oxide (SnO2) or tungsten oxide (WO3) nanoparticles on its performance as a chemoresistive material for detecting BTX vapors. Transmission electron microscopy and environmental scanning electron microscopy are used as morphological characterization techniques. SnO2-decorated graphene displayed high sensitivity towards benzene, toluene, and xylene with the lowest tested concentrations of 2 ppm, 1.5 ppm, and 0.2 ppm, respectively. In addition, we found that, by employing these nanomaterials, the observed response could provide a unique double signal confirmation to identify the presence of benzene vapors for monitoring occupational exposure in the textiles, painting, and adhesives industries or in fuel stations.


Introduction
Industrial and human developments have been steadily increasing the production and the emission towards the environment of toxic and hazardous gases. Among the various types of pollutants, aromatic volatile organic compound vapors like benzene, toluene, and xylenes are widely used in numerous fields, most notably by the petrochemical and textile industries. According to the World Health Organization (WHO), a long-duration exposure to such compounds, especially to benzene, leads to life-threatening diseases [1]. As those vapors are harmful to both human health and the environment, the European parliament has stated via a council directive that the limit value for exposure to benzene [2] is 1.6 ppb. In addition, the US Occupational Safety and Health Administration has set permissible exposure limits (PLE), eight-hour time weighted average (TWA) concentrations for toluene and xylene, which are 10 ppm and 100 ppm, respectively [3,4]. Even though the human olfactory system can identify the characteristic odor of aromatic volatile compounds, the olfactory threshold in humans is well above the exposure limits described above. Furthermore, the olfactory system fails at correctly identifying the type of aromatic compound that may be present in a given environment. Taken together,

Materials
The graphene used was purchased from Strem Chemicals Inc. (Newburyport, MA, USA), with reference no. 06-0235. This material consists of graphene nanoplatelets with an average thickness of 5-10 nanometers (3.35A between sheets) and sub-micron size (a few hundreds of nanometers). In addition, its surface area is 750 m 2 /g. Commercially available nanopowders of tungsten oxide (WO 3 ) and tin oxide (SnO 2 ) were purchased from Sigma Aldrich (Saint Louis, MO, USA). The carrier gas was nitrogen U (N 2 purer than 99.995%) and was bought from Air Liquide (Tunis, Tunisia). All these products were of analytical grade and used without further purification.

Sensing Layer Preparation
The graphene to be used as a sensing material was prepared by chemical impregnation. In the first step, a graphene suspension was produced with ethanol (10 mg of graphene were added to 10 mL of ethanol). Then the suspension was ultrasonicated (Bandelin electronic GmbH, Berlin, Germany) for one hour in order to have a homogenous dispersion of the graphene platelets in the ethanol solution, avoiding the agglomeration of graphene. In the second step, other independent dispersions in ethanol with metal oxide nanoparticles, tin oxide (SnO 2 ) and tungsten (WO 3 ), were prepared (0.5 mg of SnO 2 or WO 3 nanoparticles were dispersed in 10 mL of ethanol). Subsequently, the suspension with the de-agglomerated graphene was placed on a hot plate while agitated vigorously at 50 • and then a metal oxide NP suspension was added drop-by-drop. The stirring at 50 • was kept up to half an hour after the last drop of the NP suspension was incorporated. This was done to favor the homogeneous attachment of the metal oxide nanoparticles onto the graphene surface. The loading with metal oxide NPs of graphene was adjusted to 5% in weight. In the final step, the dissolution of graphene decorated with metal oxide nanoparticles was deposited on the interdigitated gold electrodes (alumina substrate) by the airbrushing technique at a temperature between 80 • C and 90 • C degrees (see Figure 1). The room-temperature resistance of the deposited graphene, SnO 2 /graphene, and WO 3 /graphene layers was typically about 1 kΩ, 11 kΩ, and 6 kΩ, respectively. For ensuring that a high reproducibility of the deposition method is achieved, the deposition times are controlled and the device resistance is monitored during deposition to ensure that the same typical resistances mentioned above are achieved. The volume of graphene nanoflakes used on the electrodes was optimized on the basis of our previous experience in the use of carbon nanomaterials [41]. A too thick graphene coating over the electrodes results in low baseline resistance and low responsiveness. In addition, the desorption of gas molecules from a thick layer during the recovery phases becomes more difficult. Conversely, the use of a too low quantity of graphene could result in non-homogeneous coatings, dramatically reducing responsiveness, and increasing noise levels. The loading level of metal oxide nanoparticles to graphene nanoflakes was optimized in part. Our previous experience with carbon nanotube sensors indicates that the optimal levels of metal or metal oxide NP loading is in the units of % in weight [42]. Here we used a loading level of 5% in weight to ensure that graphene nanoflakes could be decorated with tin or tungsten oxide NPs avoiding the occurrence of agglomerations of the metal oxide on the graphene surface.

Sensing Layer Preparation
The graphene to be used as a sensing material was prepared by chemical impregnation. In the first step, a graphene suspension was produced with ethanol (10 mg of graphene were added to 10 mL of ethanol). Then the suspension was ultrasonicated (Bandelin electronic GmbH, Berlin, Germany) for one hour in order to have a homogenous dispersion of the graphene platelets in the ethanol solution, avoiding the agglomeration of graphene. In the second step, other independent dispersions in ethanol with metal oxide nanoparticles, tin oxide (SnO2) and tungsten (WO3), were prepared (0.5 mg of SnO2 or WO3 nanoparticles were dispersed in 10 mL of ethanol). Subsequently, the suspension with the de-agglomerated graphene was placed on a hot plate while agitated vigorously at 50° and then a metal oxide NP suspension was added drop-by-drop. The stirring at 50° was kept up to half an hour after the last drop of the NP suspension was incorporated. This was done to favor the homogeneous attachment of the metal oxide nanoparticles onto the graphene surface. The loading with metal oxide NPs of graphene was adjusted to 5% in weight. In the final step, the dissolution of graphene decorated with metal oxide nanoparticles was deposited on the interdigitated gold electrodes (alumina substrate) by the airbrushing technique at a temperature between 80 °C and 90 °C degrees (see Figure 1). The room-temperature resistance of the deposited graphene, SnO2/graphene, and WO3/graphene layers was typically about 1 kΩ, 11 kΩ, and 6 kΩ, respectively. For ensuring that a high reproducibility of the deposition method is achieved, the deposition times are controlled and the device resistance is monitored during deposition to ensure that the same typical resistances mentioned above are achieved. The volume of graphene nanoflakes used on the electrodes was optimized on the basis of our previous experience in the use of carbon nanomaterials [41]. A too thick graphene coating over the electrodes results in low baseline resistance and low responsiveness. In addition, the desorption of gas molecules from a thick layer during the recovery phases becomes more difficult. Conversely, the use of a too low quantity of graphene could result in non-homogeneous coatings, dramatically reducing responsiveness, and increasing noise levels. The loading level of metal oxide nanoparticles to graphene nanoflakes was optimized in part. Our previous experience with carbon nanotube sensors indicates that the optimal levels of metal or metal oxide NP loading is in the units of % in weight [42]. Here we used a loading level of 5% in weight to ensure that graphene nanoflakes could be decorated with tin or tungsten oxide NPs avoiding the occurrence of agglomerations of the metal oxide on the graphene surface. Sensor configuration: A film consisting of a random organization of few layer graphene flakes, decorated with metal oxide nanoparticle coats, and interdigitated electrodes screen-printed onto an alumina substrate. The resistance of the film is measured between the two contact pads printed on the substrate. The substrate includes a printed platinum heating meander on its backside that can be used to raise the operating temperature of the sensor above the ambient temperature. Sensor configuration: A film consisting of a random organization of few layer graphene flakes, decorated with metal oxide nanoparticle coats, and interdigitated electrodes screen-printed onto an alumina substrate. The resistance of the film is measured between the two contact pads printed on the substrate. The substrate includes a printed platinum heating meander on its backside that can be used to raise the operating temperature of the sensor above the ambient temperature.

Characterization Techniques
The tested graphene sensors were characterized by transmission electron microscopy (TEM) and an Environmental Scanning Electron Microscope (ESEM). The TEM morphology investigation was undertaken with a JEOL JEM 2100F (Tokyo, Japan) TEM, at 200 kV operating voltage. The Environmental Scanning Electron Microscope was realized by Carl Zeiss AG-ULTRA 55 (Oberkochen, Germany) ESEM.

Vapor Sensing Experimental Setup
The graphene, SnO 2 /graphene, and WO 3 /graphene sensors were tested for the detection of BTX VOCs. The sensor chamber (35 cm 3 in volume) can house up to six sensors at a time and measurements were performed on two replicate devices for the three different sensitive nanomaterials studied, thus ensuring that the differences in response recorded were consistent. The vapors were produced by a dilution bench consisting of a chemical solvent vaporization cell and two flowmeters (with a flow rate of 200 sccm). Sensor resistance was measured via an Agilent HP 34972A Multimeter (Santa Clara, CA, USA) at a fixed operating frequency of 50 Hz. Due to the system configuration, the lowest achievable vapor concentrations were 2, 0.4, and 0.2 ppm for benzene, toluene, and xylene, respectively.
The target vapors were injected in the sensor chamber using nitrogen as carrier gas. The response of the sensors was calculated according to the normalized resistance variation, presented in Equation (1).
where R 0 and R g are the resistance under the carrier gas and the aromatic VOC, respectively.

TEM Characterization
A transmission electron microscopy (TEM) analysis, performed in order to characterize the pristine graphene layers, is displayed in Figure 2.

Characterization Techniques
The tested graphene sensors were characterized by transmission electron microscopy (TEM) and an Environmental Scanning Electron Microscope (ESEM). The TEM morphology investigation was undertaken with a JEOL JEM 2100F (Tokyo, Japan) TEM, at 200 kV operating voltage. The Environmental Scanning Electron Microscope was realized by Carl Zeiss AG-ULTRA 55 (Oberkochen, Germany) ESEM.

Vapor Sensing Experimental Setup
The graphene, SnO2/graphene, and WO3/graphene sensors were tested for the detection of BTX VOCs. The sensor chamber (35 cm 3 in volume) can house up to six sensors at a time and measurements were performed on two replicate devices for the three different sensitive nanomaterials studied, thus ensuring that the differences in response recorded were consistent. The vapors were produced by a dilution bench consisting of a chemical solvent vaporization cell and two flowmeters (with a flow rate of 200 sccm). Sensor resistance was measured via an Agilent HP 34972A Multimeter (Santa Clara, CA, USA) at a fixed operating frequency of 50 Hz. Due to the system configuration, the lowest achievable vapor concentrations were 2, 0.4, and 0.2 ppm for benzene, toluene, and xylene, respectively.
The target vapors were injected in the sensor chamber using nitrogen as carrier gas. The response of the sensors was calculated according to the normalized resistance variation, presented in Equation (1).
where R0 and Rg are the resistance under the carrier gas and the aromatic VOC, respectively.

TEM Characterization
A transmission electron microscopy (TEM) analysis, performed in order to characterize the pristine graphene layers, is displayed in Figure 2.  Figure 2 shows a typical example of the 2-D graphene nanoplatelets used. They consist of fewlayer graphene sheets (four layers for the sample in Figure 2) with a diameter of about a few hundreds of nanometers. This platelet shape offers a very high surface area (more than 100 m 2 /g) for its decoration with metal oxide nanoparticles and interaction with the chemical environment.  Figure 2 shows a typical example of the 2-D graphene nanoplatelets used. They consist of few-layer graphene sheets (four layers for the sample in Figure 2) with a diameter of about a few hundreds of nanometers. This platelet shape offers a very high surface area (more than 100 m 2 /g) for its decoration with metal oxide nanoparticles and interaction with the chemical environment. Additionally, graphene has a small content of oxygen functional groups (less than 8% wt.) derived from their synthesis. Nevertheless, these oxygen groups can easily interact with metal oxide nanoparticles, enhancing the decoration of graphene.

ESEM Characterization
Additionally, ESEM images were obtained for pristine and graphene decorated with WO 3 and SnO 2 . Figure 3a,b shows the ESEM images of graphene once it was deposited onto the substrate to be employed as a gas sensor at two different scales. In addition, in Figure 3c,e, the graphene decorated with WO 3 and SnO 2 are observed. In Figure 3d,f, a back-scattered electron detector (BSE) was used showing the graphene (black background) decorated with the metal oxide nanoparticles, which correspond to the bright spots. These images reveal a quite homogeneous distribution of WO 3 and SnO 2 metal oxide nanoparticles on the graphene surface.
Sensors 2020, 20, x FOR PEER REVIEW 5 of 16 Additionally, graphene has a small content of oxygen functional groups (less than 8% wt.) derived from their synthesis. Nevertheless, these oxygen groups can easily interact with metal oxide nanoparticles, enhancing the decoration of graphene.

ESEM Characterization
Additionally, ESEM images were obtained for pristine and graphene decorated with WO3 and SnO2. Figure 3a,b shows the ESEM images of graphene once it was deposited onto the substrate to be employed as a gas sensor at two different scales. In addition, in Figure 3c,e, the graphene decorated with WO3 and SnO2 are observed. In Figure 3d,f, a back-scattered electron detector (BSE) was used showing the graphene (black background) decorated with the metal oxide nanoparticles, which correspond to the bright spots. These images reveal a quite homogeneous distribution of WO3 and SnO2 metal oxide nanoparticles on the graphene surface.

Graphene Sensing Response to Aromatic VOCs
The vapors sensing performances at room temperature of the pristine graphene-based sensor towards the aromatic vapors benzene, toluene, and xylene are presented in Figure 4.
The observed response of the graphene-based sensor shows an increase of the sensor resistance with the increase in the injected vapor concentration either for benzene, toluene, or xylene. A good repeatability was observed (Figure 4b,c) at room temperature. The mechanisms behind the responses registered are probably based on electrostatic interactions. Thereby, even though π-π interactions cannot be ruled out, the room-temperature desorption of gas molecules indicates that the formation of H-bonds is probably the most predominant interaction.
A baseline drift was observed for the higher concentration tested. Due to the highly stable and reversible responses recorded for the lowest concentrations tested, this effect can be attributed to VOC molecules remaining adsorbed at the graphene surface at the end of a cleaning phase, preventing a complete baseline recovery. In fact, a short heating pulse of a few seconds at 70 °C is enough for completely desorbing all vapor molecules and fully regaining the baseline (see the inset in Figure 4b). This heating was obtained by applying an external voltage (one volt) to a heating element printed on the backside of the sensor substrate and helps desorbing vapor molecules from the graphene surface. The same behavior has been observed in a previous research in which carbon nanotubes were used [43]. This strategy of heating was applied only once to show its effect, but all the measurements shown in Figure 4 were performed at room temperature and no heating was used to help regain the baseline. The lowest concentration tested were 2 ppm, 0.4 ppm, and 0.2 ppm for benzene, toluene, and xylene, respectively, which, combined to the very low noise levels observed in these responses (0.2% at room temperature), indicates that the lower detection limit is in a range of about 200 ppb for benzene and toluene and about 50 ppb for xylene. This estimation of the limit of

Graphene Sensing Response to Aromatic VOCs
The vapors sensing performances at room temperature of the pristine graphene-based sensor towards the aromatic vapors benzene, toluene, and xylene are presented in Figure 4.
The observed response of the graphene-based sensor shows an increase of the sensor resistance with the increase in the injected vapor concentration either for benzene, toluene, or xylene. A good repeatability was observed (Figure 4b,c) at room temperature. The mechanisms behind the responses registered are probably based on electrostatic interactions. Thereby, even though π-π interactions cannot be ruled out, the room-temperature desorption of gas molecules indicates that the formation of H-bonds is probably the most predominant interaction.
A baseline drift was observed for the higher concentration tested. Due to the highly stable and reversible responses recorded for the lowest concentrations tested, this effect can be attributed to VOC molecules remaining adsorbed at the graphene surface at the end of a cleaning phase, preventing a complete baseline recovery. In fact, a short heating pulse of a few seconds at 70 • C is enough for completely desorbing all vapor molecules and fully regaining the baseline (see the inset in Figure 4b). This heating was obtained by applying an external voltage (one volt) to a heating element printed on the backside of the sensor substrate and helps desorbing vapor molecules from the graphene surface. The same behavior has been observed in a previous research in which carbon nanotubes were used [43]. This strategy of heating was applied only once to show its effect, but all the measurements shown in Figure 4 were performed at room temperature and no heating was used to help regain the baseline. The lowest concentration tested were 2 ppm, 0.4 ppm, and 0.2 ppm for benzene, toluene, and xylene, respectively, which, combined to the very low noise levels observed in these responses (0.2% at room temperature), indicates that the lower detection limit is in a range of about 200 ppb for benzene and toluene and about 50 ppb for xylene. This estimation of the limit of detection (LOD) considers that a meaningful response signal should be at least three times higher than the noise level experienced. The calibration curves presented in Figure 4d reveal a better sensitivity (i.e., slope of the calibrations curves) for xylene compared to benzene and toluene. The error bars in the calibration curves shown were calculated as the mean +/− the standard deviation over replicate measurements performed with two replicate bare graphene sensors.
It is noteworthy to stress that the best graphene sensor performance, either from its sensitivity or its baseline stability, was observed when operated at room temperature. Additional measurements performed at operating temperatures above the ambient indicate that the response decreases when graphene is heated. This decrease in responsiveness is probably due to the fact that heating promotes desorption of the weakly physisorbed VOC molecules from the graphene surface. Such results are presented in the Supplementary Materials file, Figure S1 and Table S1.
Sensors 2020, 20, x FOR PEER REVIEW 7 of 16 detection (LOD) considers that a meaningful response signal should be at least three times higher than the noise level experienced. The calibration curves presented in Figure 4d reveal a better sensitivity (i.e., slope of the calibrations curves) for xylene compared to benzene and toluene. The error bars in the calibration curves shown were calculated as the mean +/− the standard deviation over replicate measurements performed with two replicate bare graphene sensors. It is noteworthy to stress that the best graphene sensor performance, either from its sensitivity or its baseline stability, was observed when operated at room temperature. Additional measurements performed at operating temperatures above the ambient indicate that the response decreases when graphene is heated. This decrease in responsiveness is probably due to the fact that heating promotes desorption of the weakly physisorbed VOC molecules from the graphene surface. Such results are presented in the Supplementary Materials file, Figure S1 and Table S1.

Sensing Response to Aromatic VOCs of Metal Oxide Nanoparticle-Decorated Graphene
It is well known that semiconductor metal oxide materials such as tin or tungsten oxides can be used as sensitive materials for the detection of vapors and gases [44][45][46]. Here we studied the effect of decorating graphene with nanoparticles of either tungsten or tin oxide for tuning gas sensing properties. First, an optimization of the sensor operating temperature was conducted by

Sensing Response to Aromatic VOCs of Metal Oxide Nanoparticle-Decorated Graphene
It is well known that semiconductor metal oxide materials such as tin or tungsten oxides can be used as sensitive materials for the detection of vapors and gases [44][45][46]. Here we studied the effect of decorating graphene with nanoparticles of either tungsten or tin oxide for tuning gas sensing properties. First, an optimization of the sensor operating temperature was conducted by characterizing sensor response at different temperatures ranging from room temperature to 300 • C. It was observed that the response of WO 3 /graphene and SnO 2 /graphene sensors towards the VOCs tested increased when the working temperature was raised up to 250 • C and then decreased again at 300 • C. The substantial increase in the optimal operating temperature of hybrid sensors in comparison to the one of bare graphene is because metal oxide nanoparticles need such high temperatures for activating the chemisorption and reaction of VOCs. However, a too high operating temperature hinders the adsorption of gas molecules and results in decreased responsiveness [47]. The results on the performance of metal oxide-decorated graphene sensors as a function of the operating temperature can be found in the Supplementary Materials file, Figures S2-S6 and Tables S2 and S3. The following sub-sections show the results obtained when sensors were operated at their optimal working temperature of 250 • C.

Response of Tungsten Oxide Nanoparticle-Decorated Graphene
The responses of the WO 3 /graphene-based sensor operated at 250 • C towards the vapors of benzene, toluene, and xylene are presented in Figure 5.
Sensors 2020, 20, x FOR PEER REVIEW 8 of 16 300 °C. The substantial increase in the optimal operating temperature of hybrid sensors in comparison to the one of bare graphene is because metal oxide nanoparticles need such high temperatures for activating the chemisorption and reaction of VOCs. However, a too high operating temperature hinders the adsorption of gas molecules and results in decreased responsiveness [47]. The results on the performance of metal oxide-decorated graphene sensors as a function of the operating temperature can be found in the Supplementary Materials file, Figures S2-S6 and Tables S2 and S3. The following sub-sections show the results obtained when sensors were operated at their optimal working temperature of 250 °C.

Response of Tungsten Oxide Nanoparticle-Decorated Graphene
The responses of the WO3/graphene-based sensor operated at 250 °C towards the vapors of benzene, toluene, and xylene are presented in Figure 5. The observed response of the WO3/graphene-based sensor shows an increase in the sensor resistance with the increase of the injected vapor concentration, for any of the aromatic VOCs tested. A good repeatability was observed (Figure 5c) at 250 degrees. The mechanisms behind the responses correspond to an overall p-type semiconductor behavior of the gas sensitive film. The chemisorption of aromatic VOCs onto the tungsten oxide nanoparticles transfers electrons from the adsorbed molecule towards the conduction band of tungsten oxide. These electrons are eventually injected to graphene, shifting the Fermi level of graphene towards the conduction band. As a result, the number The observed response of the WO 3 /graphene-based sensor shows an increase in the sensor resistance with the increase of the injected vapor concentration, for any of the aromatic VOCs tested. A good repeatability was observed (Figure 5c) at 250 degrees. The mechanisms behind the responses correspond to an overall p-type semiconductor behavior of the gas sensitive film. The chemisorption of aromatic VOCs onto the tungsten oxide nanoparticles transfers electrons from the adsorbed molecule towards the conduction band of tungsten oxide. These electrons are eventually injected to graphene, shifting the Fermi level of graphene towards the conduction band. As a result, the number of holes in graphene decreases and the overall resistance increases [48]. The calibration curves presented in Figure 5d reveal a better sensitivity for xylene compared to benzene and toluene. The error bars in the calibration curves shown were calculated as the mean +/− the standard deviation over replicate measurements performed with two replicate WO 3 /graphene sensors. Once again, the lowest concentration tested were 2 ppm, 1.5 ppm, and 0.2 ppm for benzene, toluene, and xylene, respectively, which, combined to the very low noise levels observed in these responses, indicates that the lower detection limit for the aromatic VOCs is about 100 ppb for benzene and toluene and about 20 ppb for xylene. The lower detection limit was estimated in relation to the observed response noise of 0.3% at 250 • C. The response signal at these low concentrations is higher (roughly two times higher) than the one recorded with bare graphene. This implies that the limit of detection for VOCs in tungsten oxide-decorated graphene sensors is better (i.e., lower) than in bare graphene sensors.

Response of Tin Oxide Nanoparticle-Decorated Graphene
The responses of the SnO 2 /graphene-based sensor operated at 250 • C towards the vapors of benzene, toluene, and xylene are presented in Figure 6.
Sensors 2020, 20, x FOR PEER REVIEW 9 of 16 of holes in graphene decreases and the overall resistance increases [48]. The calibration curves presented in Figure 5d reveal a better sensitivity for xylene compared to benzene and toluene. The error bars in the calibration curves shown were calculated as the mean +/− the standard deviation over replicate measurements performed with two replicate WO3/graphene sensors. Once again, the lowest concentration tested were 2 ppm, 1.5 ppm, and 0.2 ppm for benzene, toluene, and xylene, respectively, which, combined to the very low noise levels observed in these responses, indicates that the lower detection limit for the aromatic VOCs is about 100 ppb for benzene and toluene and about 20 ppb for xylene. The lower detection limit was estimated in relation to the observed response noise of 0.3% at 250 °C. The response signal at these low concentrations is higher (roughly two times higher) than the one recorded with bare graphene. This implies that the limit of detection for VOCs in tungsten oxide-decorated graphene sensors is better (i.e., lower) than in bare graphene sensors.

Response of Tin Oxide Nanoparticle-Decorated Graphene
The responses of the SnO2/graphene-based sensor operated at 250 °C towards the vapors of benzene, toluene, and xylene are presented in Figure 6. For benzene vapors, the SnO2/graphene-based sensor shows the same p-type behavior already observed in bare and tungsten oxide-decorated graphene. Sensor resistance increases for increasing concentrations of benzene. However, for toluene and xylene vapors, the sensor behaves as an n-type semiconductor, because its resistance clearly decreases for increasing concentrations of these species. For benzene vapors, the SnO 2 /graphene-based sensor shows the same p-type behavior already observed in bare and tungsten oxide-decorated graphene. Sensor resistance increases for increasing concentrations of benzene. However, for toluene and xylene vapors, the sensor behaves as an n-type semiconductor, because its resistance clearly decreases for increasing concentrations of these species. Such transitions from p-type to n-type materials have been reported before in graphene nanomaterials and attributed to adsorbate induced charge transfer and charge carrier scattering effects [49,50]. These response inversion effects appear at different operating temperatures for tin oxide and tungsten oxide-decorated graphene, and the dipole moment (DM) of the molecule detected plays an important role. Indeed, for SnO 2 /graphene-based sensors, while this effect is not seen for benzene (DM = 0), it is observed for toluene and o-xylene, which have increasing DM. For ethanol vapors, a molecule with even higher DM, this response inversion effect appears not only in SnO 2 /graphene but also in WO 3 /graphene operated at 300 • C (these results are summarized in the Supplementary Materials).
The calibration curves presented in Figure 6d reveal a better sensitivity for xylene compared to benzene and toluene. The error bars in the calibration curves shown were calculated as the mean +/− the standard deviation over replicate measurements performed with two replicate SnO 2 /graphene sensors. Once more, the very low noise levels observed in the responses indicate that the lower detection limit for the aromatic VOCs lies in the range of a few hundred of ppb (i.e., about 100 ppb for benzene and toluene and about 20 ppb for xylene). The lower detection limit was estimated in relation to the observed response noise of 0.3% at 250 • C. A good repeatability was observed (see Figure 6c) at 250 degrees. The mechanism behind the response signals at these low concentrations is even higher than the one recorded with WO 3 /graphene. This implies that the limit of detection for VOCs in SnO 2 /graphene sensors is better (i.e., lower) than in tungsten WO 3 /graphene and bare graphene sensors.

Sensitivity to Aromatic VOCs
To calculate the sensitivities of the fabricated sensors to all the tested vapors, the calibration curves were fitted using a linear regression model. The sensitivity (slopes of the calibration curves) for pristine graphene, WO 3 /graphene, and SnO 2 /graphene sensors are presented in Table 1. The best performance is observed for the SnO 2 /graphene sensor for benzene, toluene, and xylene vapors. These differences can be related to the different dipolar moments of the aromatic VOCs tested and the different activation energy of the gas-sensitive materials. The higher sensitivity for tin oxide-decorated graphene could also be due to a better matching between the work function of graphene (near 4.6 eV) and that of tin oxide (near 4.5 eV) as compared to tungsten oxide (near 5 eV) [51][52][53]. Results on the long-term stability of sensors are shown in the Supplementary Materials. A 20% drop in sensitivity was observed after one month of operation under humid conditions. The sensitivity towards non-aromatic VOCs was tested by measuring ethanol (see Figure S6, Tables S2 and S3, Supplementary Materials). Ethanol sensitivity was found to be comparable to the one for benzene but significantly lower than the one for toluene (3-fold lower) or xylene (10-fold lower). There are very few reports in the literature about the detection of aromatic VOCs using carbon nanomaterials. Table 2 summarizes a comparison between the performance in the detection of BTX vapors reported here and previously published results. The normalized response of our metal oxide-decorated graphene sensors show higher values for xylene and toluene vapors than those reported in previously published results [3,[41][42][43]54]. Mirzaei and co-workers report in their review [3] a ZnO nanowire sensor modified with small amounts of reduced graphene oxide (rGO) with a remarkably high response to benzene vapors. This differs from our approach in which graphene is modified with small amounts of metal oxides. However, the rGO-loaded metal oxide sensor reported operates at rather high temperatures and is loaded with Pd [3]. At such temperatures, Pd acts as a catalyst and high cross-sensitivity to hydrogen, carbon monoxide, alcohols, or hydrocarbons can be foreseen. Table 2. Comparison of the normalized responses (NR) expressed as response % × ppm −1 (where response % is defined as Rg/R 0 (%)) for benzene, toluene, and xylene for different carbon nanomaterial gas sensors. Other parameters such as the limit of detection (LOD) expressed in ppb, operating temperature ( • C), and dynamic range of concentration measured experimentally (in ppm) are given too. It is worth mentioning that the loading level of metal oxide nanoparticles on graphene could be optimized further in order to possibly achieve an even higher sensitivity to aromatic VOCs. This would involve checking different loading levels in a narrow range near the 5% level used here. Loading levels below 1% do not significantly alter the properties of the bare material and loading levels higher than 10% result in the agglomeration of metal oxide nanoparticles, inhomogeneous distribution of these on the graphene surface, and loss of sensitivity.

Response Time and Recovery Time
The response and recovery times of graphene, SnO 2 /graphene, and WO 3 /graphene sensors toward aromatic vapors are presented in Tables 3 and 4, respectively. The response time was estimated as the time required for the sensor to respond from 10% to 90% of its maximum response value. The recovery time was estimated as the time required for the sensor signal to fall towards its baseline, from 90% to 10% of its response value during a cleaning step. The results show faster response times for the SnO 2 /graphene sensor towards toluene and xylene, roughly half a minute. These response times compare favorably to the reported results for graphene sensors [3,48]. For benzene, the best response time is displayed by the graphene sensor. The recovery times of the developed sensors range between 2.5 min and nearly 7 min. A rather slow return to the baseline is achieved compared to metal oxide chemoresistors [3]. This is possibly due to the strong interaction between VOC molecules and the surface of metal oxide-loaded graphene sensors and the lower operating temperatures used.

Discussion
It is well known that the exposure to high concentrations of BTX vapors, commonly used as organic solvents in the industrial production of paints, lacquers, adhesives, and detergents may cause serious carcinogenic effects. The monitoring of the occupational exposure to these compounds is thus of great importance for the preservation of workers' health. Currently, the eight-hour TVA level for benzene as indicated by OSHA is 1 ppm (the TVA levels for toluene and xylene are higher) [55]. The metal oxide-loaded graphene sensors reported here are sensitive enough and show promising selectivity for detecting BTX vapors at the relevant concentrations for monitoring occupational exposure. The results reported here can be used to further the progress towards the development of a sensor array for the selective detection of BTX.
In a previous research [54], we showed that a sensor employing multiwall carbon nanotubes (MWCNTs) decorated with gold nanoparticles and functionalized with a self-assembled monolayer of long-chain thiols (MHDA) was selective to non-aromatic vapors. This material was insensitive to aromatic VOCs, which can be used as a negative test for confirming/ruling out the presence of aromatic vapors in a complex environment. The three materials studied here (i.e., graphene and the two metal oxide-decorated graphene) display different, overlapping sensitivities for aromatic VOCs. However, the fact that tin oxide-loaded graphene shows a characteristic response inversion for toluene and xylene vapors (and not for benzene) can be used as a confirmation test for the presence of benzene in the environment analyzed. To illustrate this approach, a lack of response from the MHDA-Au-MWCNT sensor combined with p-type responses from the other three sensors would be indicative of the presence of benzene. In contrast, a lack of response from the MHDA-Au-MWCNT sensor combined with a p-type response from the graphene and WO 3 /graphene sensors together with an n-type response from the SnO 2 /graphene sensor would be indicative of the presence of toluene or xylene. Finally, a response from the MHDA-Au-MWCNT sensor would indicate the presence of a non-aromatic VOC such as methanol, ethanol, or acetone.

Conclusions
In this paper, we studied the effect of tin oxide (SnO 2 ) and tungsten oxide (WO 3 ) nanoparticles decorating graphene on the performance of the resulting hybrid nanomaterials as chemiresistive sensors for detecting BTX vapors. The decoration of few layer graphene flakes resulted in high sensitivity towards benzene, toluene, and xylene vapors. The sensors could be suitable for the inexpensive monitoring occupational exposure to BTX. In a future research, the materials developed could be used for integrating a four-element sensor array on a single-chip MEMS configuration, and thus could provide a unique double signal confirmation response associated with the presence of highly toxic benzene vapors (as described in the discussion section). The fabrication of this single-chip sensor, the fine-tuning of the loading level of graphene with metal oxide nanoparticles to maximize the response to aromatic VOCs, and the study of the effect of ambient moisture will be the subject of our next research.