Self-Assembled Vanadium Oxide Nanoflakes for p-Type Ammonia Sensors at Room Temperature

VO2(B), VO2(M), and V2O5 are the most famous compounds in the vanadium oxide family. Here, their gas-sensing properties were investigated and compared. VO2(B) nanoflakes were first self-assembled via a hydrothermal method, and then VO2(M) and V2O5 nanoflakes were obtained after a heat-phase transformation in nitrogen and air, respectively. Their microstructures were evaluated using X-ray diffraction and scanning and transmission electron microscopies, respectively. Gas sensing measurements indicated that VO2(M) nanoflakes were gas-insensitive, while both VO2(B) and V2O5 nanoflakes were highly selective to ammonia at room temperature. As ammonia sensors, both VO2(B) and V2O5 nanoflakes showed abnormal p-type sensing characteristics, although vanadium oxides are generally considered as n-type semiconductors. Moreover, V2O5 nanoflakes exhibited superior ammonia sensing performance compared to VO2(B) nanoflakes, with one order of magnitude higher sensitivity, a shorter response time of 14–22 s, and a shorter recovery time of 14–20 s. These characteristics showed the excellent potential of V2O5 nanostructures as ammonia sensors.


Introduction
In many industries, hazardous gases have become increasingly important raw materials, and for this reason it has become very important to develop highly sensitive gas sensors to monitor them in the manufacturing process. To this end, finding suitable materials with the required surface/bulk properties is essential for the gas sensor development. So far, metal oxide semiconductors (MOS) are the most attractive gas-sensitive materials [1]. Because nanostructured materials have a large surface-to-volume ratio and abundant surface states, many gas sensors based on SnO 2 [2][3][4], TiO 2 [5,6], ZnO [2,[7][8][9], WO 3 [10,11], In 2 O 3 [12][13][14], and other metal oxide nanostructures have been successfully fabricated to detect and quantify gaseous species. MOS sensors are applicable in many areas of human activity. Different MOS gas sensors can be used to construct sensor matrixes named electronic noses to identify different odorous compounds and quantify their concentration [15][16][17][18][19]. Among the numerous metal oxides, vanadium oxides form an interesting material group because of their varying oxidation states between V 2+ and V 5+ [20]. At least 15 different vanadium oxides (V x O y ), such as VO, V 2 O 3 , VO 2 , V 2 O 5 , V n O 2n−1 , and V 2n O 5n−2 , have been reported [21,22]. Among them, VO 2 (B phase and M/R phase) and V 2 O 5 are the most famous, attracting great interest due to their special chemical/physical properties and their potential application in many fields. Table 1 shows the crystal structure, properties, Recently, vanadium oxides have been considered as new candidates for gas sensors. Their sensing properties for inorganic/organic gases such as nitrogen oxides [27,28], ethanol [1,[29][30][31], butyl-amine [32], and ammonia [33][34][35] have been reported. However, the gas-sensing properties of vanadium oxide nanostructures are strongly dependent on the actual synthesis environment and closely correlated with material morphologies, surface states, and microstructures. For example, several groups reported that V 2 O 5 nanostructures (e.g. nanorods, flow-like nanostructures) were more sensitive to ethanol than to ammonia at room temperature [29][30][31]; however, an opposite higher sensitivity to ammonia is also reported by Hakim et al. in V 2 O 5 nanoneedles [36]. Some groups even reported ultrahigh sensitivities of V 2 O 5 nanostructures (e.g. nanofibers, nanofilms) for ultralow level (<1 ppm) ammonia detection [33][34][35]. In addition, the conduction type n or type p usually determines the direction of resistance change when they are exposed to target gases. When n-type MOS gas sensors are utilized to detect reducing gases, reductive gas species react with the chemisorbed oxygen on the surface and electrons trapped by oxygen are released into the conduction band of MOS, leading to a decrease in resistance. With regard to p-type MOS gas sensors, the direction of resistance change is opposite due to the combination of holes with electrons released from the surface reaction. Vanadium oxides' nanostructures are generally regarded as n-type semiconductors, which exhibit n-type gas-sensing responses in most literature reports. However, interesting p-type sensing behaviors have also been reported in a few studies. Yu et al. reported a p-type sensing response to NO 2 in vanadium oxide nanotubes at 80 • C [37]. Also, p-type sensing responses to either ethanol or ammonia at room temperature were reported in V 2 O 5 and VO 2 (B) nanostructures [29,38]. Obviously, the synthesis environment and the microstructure played a crucial role in gas sensors of vanadium oxide nanostructures.
In this work, self-assembled VO 2 (B), VO 2 (M), and V 2 O 5 nanoflakes were synthesized and their gas-sensing properties were comparatively investigated. VO 2 (B) nanoflakes were first self-assembled via a hydrothermal method, and then V 2 O 5 and VO 2 (M) specimens were obtained by annealing VO 2 (B) nanoflakes in air and nitrogen, respectively. Their gas-sensing properties demonstrated that VO 2 (B) and V 2 O 5 specimens displayed superior selectivity to ammonia, while the VO 2 (M) specimen was gas-insensitive. Unlike the common n-type gas-sensing mechanism most reported, both VO 2 (B) and V 2 O 5 nanoflakes exhibited abnormal p-type sensing properties in ammonia. Compared to the VO 2 (B) Nanomaterials 2019, 9, 317 3 of 12 specimen, the ammonia sensor of V 2 O 5 nanoflakes showed a one order of magnitude higher sensitivity, faster response speed, and better reproducibility, exhibiting wide potential in ammonia sensors.

Synthesis of Self-Assembled Vanadium Oxide Nanoflakes
Self-assembled VO 2 (B) nanoflakes were prepared via a hydrothermal method, as described in previous works [39]. Briefly, 0.48 g commercial V 2 O 5 powder was added to 80 mL oxalic acid (0.1 mol/L) in an aqueous solution to form a yellow slurry. The slurry was stirred for 30 min and then transferred to a 100-mL autoclave with a Teflon liner. The autoclave was maintained at 180 • C for 36 h and then air-cooled to room temperature. The resulting dark blue precipitates (VO 2 (B)) were collected and washed with distilled water and ethanol several times and then dried at 60 • C under a vacuum for 10 h. Finally, the VO 2 (M) specimen was obtained by heating VO 2 (B) at 550 • C in an N 2 atmosphere for 1 h, and the V 2 O 5 specimen was obtained by heating VO 2 (B) at 550 • C in air for 1 h. In all reactions, all chemicals used were purchased without further purification and without adding any additives or surfactants.

Sensor Fabrication and Gas-Sensing Tests
The diagram of gas sensors is shown in Figure 1. Synthesized vanadium oxide samples were deposited on a SiO 2 tile for gas sensing tests. The SiO 2 tile was pre-patterned with interdigitated gold electrodes. For each sensor fabrication, the individual substrate was placed in a hot plate and heated to 50 • C. VO 2 (B), VO 2 (M), and V 2 O 5 samples were each separately dispersed in ethanol at a concentration of 200 mg mL −1 and then deposited on the hot sensor substrates. Finally, the sensor substrates with vanadium oxide samples across the interdigitated gold electrodes were dried in air for 40 min.

Synthesis of Self-Assembled Vanadium Oxide Nanoflakes
Self-assembled VO2(B) nanoflakes were prepared via a hydrothermal method, as described in previous works [39]. Briefly, 0.48 g commercial V2O5 powder was added to 80 mL oxalic acid (0.1 mol/L) in an aqueous solution to form a yellow slurry. The slurry was stirred for 30 min and then transferred to a 100-mL autoclave with a Teflon liner. The autoclave was maintained at 180 °C for 36 h and then air-cooled to room temperature. The resulting dark blue precipitates (VO2(B)) were collected and washed with distilled water and ethanol several times and then dried at 60 °C under a vacuum for 10 h. Finally, the VO2(M) specimen was obtained by heating VO2(B) at 550 °C in an N2 atmosphere for 1 h, and the V2O5 specimen was obtained by heating VO2(B) at 550 °C in air for 1 h. In all reactions, all chemicals used were purchased without further purification and without adding any additives or surfactants.

Sensor Fabrication and Gas-Sensing Tests
The diagram of gas sensors is shown in Figure 1. Synthesized vanadium oxide samples were deposited on a SiO2 tile for gas sensing tests. The SiO2 tile was pre-patterned with interdigitated gold electrodes. For each sensor fabrication, the individual substrate was placed in a hot plate and heated to 50 °C. VO2(B), VO2(M), and V2O5 samples were each separately dispersed in ethanol at a concentration of 200 mg mL −1 and then deposited on the hot sensor substrates. Finally, the sensor substrates with vanadium oxide samples across the interdigitated gold electrodes were dried in air for 40 min.
The gas-sensing properties of sensors were evaluated on a measuring system containing a static gas distribution chamber with a volume of 20 L. To simulate the actual measuring circumstances, we referenced the measuring method used by Qin et al. [29]. The ambient air was introduced into the chamber and target gases/liquids were injected and diluted/volatilized in the chamber. The relative chamber humidity was monitored using an internal humidity meter. During the measurement, the relative humidity is about 37% and the room temperature is about 24 °C.

Results
VO2(B) specimen was prepared via a hydrothermal procedure, and their representative SEM images are shown in Figure 2. Obviously, the VO2(B) specimen consists of large quantities of nanoflakes. It is noted that these nanoflakes are not distributed randomly in the products, but parallel arranged and self-assembled forming clumps as shown in Figure 2a. The medium-and highmagnification SEM images indicate that these nanoflakes are typically several micrometers in length, hundreds of nanometers in width (Figure 2b), and 20-50 nm in thickness ( Figure 2c). In addition, these nanoflakes are also assembled together in the TEM image ( Figure 2d). The corresponding high- The gas-sensing properties of sensors were evaluated on a measuring system containing a static gas distribution chamber with a volume of 20 L. To simulate the actual measuring circumstances, we referenced the measuring method used by Qin et al. [29]. The ambient air was introduced into the chamber and target gases/liquids were injected and diluted/volatilized in the chamber. The relative chamber humidity was monitored using an internal humidity meter. During the measurement, the relative humidity is about 37% and the room temperature is about 24 • C.

Results
VO 2 (B) specimen was prepared via a hydrothermal procedure, and their representative SEM images are shown in Figure 2. Obviously, the VO 2 (B) specimen consists of large quantities of nanoflakes.
It is noted that these nanoflakes are not distributed randomly in the products, but parallel arranged and self-assembled forming clumps as shown in Figure 2a. The medium-and high-magnification SEM images indicate that these nanoflakes are typically several micrometers in length, hundreds of nanometers in width (Figure 2b), and 20-50 nm in thickness ( Figure 2c). In addition, these nanoflakes are also assembled together in the TEM image ( Figure 2d). The corresponding high-resolution (HR) TEM image ( By oxidizing self-assembled VO2(B) nanoflakes in air at 550 °C, a V2O5 specimen was obtained. As shown in Figure 3a,b, the V2O5 specimen displays self-assembled flake-like nanostructures similar to the VO2(B) specimen. However, these V2O5 nanoflakes were slightly cracked after the calcination, showing a surface roughness increment in Figure 3b   By oxidizing self-assembled VO 2 (B) nanoflakes in air at 550 • C, a V 2 O 5 specimen was obtained. As shown in Figure 3a,b, the V 2 O 5 specimen displays self-assembled flake-like nanostructures similar to the VO 2 (B) specimen. However, these V 2 O 5 nanoflakes were slightly cracked after the calcination, showing a surface roughness increment in Figure 3b  annealing VO2(B) nanoflakes in nitrogen at 550 °C, a VO2(M) specimen was obtained. As shown in Figure 3d,e, VO2(B) nanoflakes transform to rod-and particle-like VO2(M) nanostructures and the self-assembled flake-like structure is destroyed to some extent. The corresponding HRTEM image (upper inset of Figure 3f) indicates clear lattice fringes. The spacing between two adjacent lattice planes is 0.32 and 0.23 nm, respectively, corresponding to the (011) and the (002) plane of M phase VO2. The corresponding FFT patterns (lower inset of Figure 3f) confirm the single-crystalline nature of these VO2(M) nanostructures.   In this work, their gas-sensing properties are investigated and compared. When sensors based on VO2(B), VO2(M), and V2O5 specimens are exposed in 100 ppm ammonia, ethanol, hydrogen, acetone, and isopropanol, they show different resistance response. As shown in Figure 4b, all vanadium oxide specimens exhibit higher response to ammonia, while no obvious response towards other target gases In this work, their gas-sensing properties are investigated and compared. When sensors based on VO 2 (B), VO 2 (M), and V 2 O 5 specimens are exposed in 100 ppm ammonia, ethanol, hydrogen, acetone, and isopropanol, they show different resistance response. As shown in Figure 4b, all vanadium oxide specimens exhibit higher response to ammonia, while no obvious response towards other target gases is observed. The resistance increment in 100 ppm ammonia are of 56.3%, 67.2%, and 12.4%, corresponding to VO 2 (B), V 2 O 5 , and VO 2 (M), respectively; thus, the three sensors show higher selectivity for ammonia than for other target gases. Figure 4c shows the resistance variation of VO 2 (B), VO 2 (M), and V 2 O 5 sensors when they are exposed in 50-600 ppm ammonia at room temperature. Figure 4d and the inset show their corresponding sensitivity. It is noted that an approximated power-law dependence between the sensitivity and the NH 3 concentration is observed for both V 2 O 5 and VO 2 (B) nanoflakes, matching the sensor response law developed by Gurlo et al. [40]. Here, the sensitivity is defined as S = R g − R 0 /R 0 , where R 0 and R g are the resistance of sensors in air and in ammonia, respectively. The resistance R 0 for sensors of VO 2 (B), VO 2 (M) and V 2 O 5 are (1.41 ± 0.063) × 10 4 Ω, (2.38 ± 0.078) × 10 6 Ω, and (1.73 ± 0.059) × 10 5 Ω, respectively. The difference in resistance can be interpreted by their different intrinsic electrical properties. V 2 O 5 is a semiconducting/insulating oxide with a band gap of 2.2 eV, and usually exhibits low conductivity due to the empty 3d orbital [41]. VO 2 (M) undergoes a near room-temperature MIT accompanied by a rapid change in resistivity, usually showing high resistivity at room temperature [41]. Although VO 2 (B) is generally regarded as an n-type semiconductor in gas sensing research [38], it is a theoretical semimetal/metal phase at room temperature, usually exhibiting high conductivity compared to VO 2 (M) and V 2 O 5 [42]. As the ammonia concentration increases from 0 to 600 ppm, both the VO 2 (B) and V 2 O 5 sensors show nonlinear increases of resistance. The sensor resistance of VO 2 (B) in 600 ppm ammonia is (9.58 ± 0.043) × 10 4 Ω, showing a~5.80-fold increment, while the value for the V 2 O 5 sensor is (9.58 ± 0.067) × 10 6 Ω, showing a~54.4-fold increment. Obviously, the sensor sensitivity of V 2 O 5 is superior, about one order of magnitude higher than the sensitivity of VO 2 (B). Unlike sensors of VO 2 (B) and V 2 O 5 , the resistance of the VO 2 (M) sensor first increases with the ammonia concentration, reaches a maximum value of (2.92 ± 0.033) × 10 6 Ω at 350 ppm, and then gradually decreases to (2.5 ± 0.028) × 10 6 Ω at 600 ppm. Therefore, the VO 2 (M) sensor whose resistance and sensitivity fluctuate in a narrow range ( Figure 4c and the inset of Figure 4d) is not suitable for the ammonia detection.
To investigate the influence of the specific surface area on the ammonia sensing performance, we used the BET method of adsorption and desorption of nitrogen gas to measure the specific surface area of VO 2 (B) and V 2 O 5 specimens. As shown in Figure 5, isotherms for both VO 2 (B) and V 2 O 5 nanoflakes followed a typical IV-type curve with a clear hysteresis loop at p/p 0 values of 0.55-0.99 for VO 2 (B) and 0.40-0.99 for V 2 O 5 . The BET specific surface area calculated from the nitrogen isotherms is 23.8 and 30.3 m 2 g −1 , corresponding to VO 2 (B) and V 2 O 5 nanoflakes, respectively. Although V 2 O 5 nanoflakes show a larger specific surface area, the two values are comparable, while the sensitivity of V 2 O 5 in this work is about 10 times that of VO 2 (B). Moreover, V 2 O 5 always showed higher sensing performance to ammonia than VO 2 (B) in the previous literature [38,[43][44][45]. Therefore, one reasonable explanation is that the intrinsic properties of V 2 O 5 determine its higher sensing performance. VO 2 (B) and V 2 O 5 are generally regarded as n-type semiconductors in gas-sensing investigations [38]; however, VO 2 (B) is a theoretical semimetal/metal phase at room temperature [42]. Therefore, V 2 O 5 should provide more absorbed oxygen ions on the surface, leading to high sensing performance.
The surface-depletion model is usually used to describe the sensing mechanism of the resistance-type metal-oxide semiconductor sensor. When the sensor is exposed to air, some oxygen molecules will be adsorbed on the surface, and then some oxygen ions, (O − (ads) ), will be formed at the surface. When the sensor is put in the reducing gas, ammonia in this work, the reducing gas molecules will react with the oxygen ions on the surface as described below: nanoflakes show a larger specific surface area, the two values are comparable, while the sensitivity of V2O5 in this work is about 10 times that of VO2(B). Moreover, V2O5 always showed higher sensing performance to ammonia than VO2(B) in the previous literature [38,[43][44][45]. Therefore, one reasonable explanation is that the intrinsic properties of V2O5 determine its higher sensing performance. VO2(B) and V2O5 are generally regarded as n-type semiconductors in gas-sensing investigations [38]; however, VO2(B) is a theoretical semimetal/metal phase at room temperature [42]. Therefore, V2O5 should provide more absorbed oxygen ions on the surface, leading to high sensing performance. Figure 5. N2 adsorption-desorption isotherms of self-assembled VO2(B) and V2O5 nanoflakes. Figure 5. N 2 adsorption-desorption isotherms of self-assembled VO 2 (B) and V 2 O 5 nanoflakes.
Thus the carrier concentration changes with the ammonia level and, consequently, causes the sensor resistance to change. It is well known that semiconductor metal oxides are usually classified as either n-type or p-type. Conductivity type plays an important role in sensor responses. For n-type semiconductor sensors, the reductive gas species reacts with the adsorbed oxygen ions and the electrons trapped by oxygen are released into the conduction band, leading to a decrease in resistance. With regard to p-type semiconductor sensors, the opposite change in the resistance is observed due to the combination of holes with electrons released from the surface reaction. Both VO 2 (B) and V 2 O 5 are generally considered to be n-type semiconductors in gas-sensing investigations, and there are many studies reporting their n-type gas-sensing responses. In studies by Raj et al. [31] and Modafferi et al. [34], sensors based on V 2 O 5 displayed a resistance decrease when they were exposed to an ascending ammonia level and the material was treated as n-type. However, our measured results indicate that the sensor resistance of VO 2 (B) and V 2 O 5 displays a unidirectional increase as the ammonia level increases from 0 to 600 ppm, showing typical p-type resistance responses. Similar intriguing p-type gas-sensing behaviors have been reported previously. Qin et al. [29] and Evans et al. [38] reported p-type resistance responses in V 2 O 5 hierarchical networks and VO 2 (B) nanoparticles, and attributed them to the inversion layer formed at the surface due to a larger quantity of surface oxygen vacancy [29]. These contrasting results suggest that the sensing response type of vanadium oxides at room temperature is highly dependent on the synthesis environment and the resultant surface species.
I-V curves for sensors of VO 2 (B) and V 2 O 5 in 0-600 ppm ammonia are measured at room temperature and shown in Figure 6a,b. All of the curves appear to be linear, indicating that good ohmic contacts formed between the metal electrodes and nanoflakes. From the I-V curves, it is seen that the conduction of two devices progressively decays with the ascending ammonia level, showing p-type sensing responses and agreeing with the resistance measurement (Figure 4c). Figure 6c,d present the dynamic responses of VO 2 (B) and V 2 O 5 sensors in the ascending ammonia levels. When the ammonia was continuously injected into the testing chamber with a step of 50 ppm, both sensors showed nonlinear increases in their resistances. From Figure 6c,d, it is obvious that in each ammonia level of both sensors the resistance ladder is very clear, revealing good dynamic response characteristics. Moreover, the V 2 O 5 sensor shows a one order of magnitude higher sensitivity than the VO 2 (B) sensor. In both sensors, NH 3 reacts with the surface adsorbed oxygen ions, releasing electrons and inducing an electrical response to the ascending ammonia level. However, VO 2 and V 2 O 5 are normally considered n-type semiconductors at room temperature, so the formation of a surface inversion layer is a reliable interpretation for the p-type behaviors [29,38], where the released electrons would reduce the number of holes (the majority charge carriers in the inversion layer) and result in a p-type increase of the sensor resistivity. Further dynamic testing procedures were carried out, which provided more information on the most important parameters for a sensing device: sensitivity, response and recovery time, and reproducibility. Figure 7a,b show the dynamic responses of VO2(B) and V2O5 sensors for 10 cycles where the dynamic mesurement was performed between air and 550 ppm ammonia at room temperature. For clear observation, only 10 cycles are given. Similarly, both VO2(B) and V2O5 sensors indicate good reproducibility, as revealed in the repeated measurements (Figure 7a,b). When sensors of VO2(B) and V2O5 were exposed to 550 ppm ammonia, the ammonia adsorption was triggered, and simultaneously the sensor resistance increased abruptly and then reached a relative stable value. When the sensors were switched to air again, ammonia desorption happened, and the sensor resistance decreased abruptly and then reached a relative stable value. The response time and recovery time for the VO2(B) sensor (defined as 95% of the time between the peak value and the valley value) were 63-85 s and 11-16 s (Figure 7c), respectively. Comparatively, the V2O5 sensor indicated a faster response: the response time was 14-22 s and the recovery time was 14-20 s (Figure 7d). Compared with other ammonia sensors of VO2(B) and V2O5 reported previously, the sensors in this work showed a fast response/recovery rate ( Table 2). As shown in Figure 7c, it is clear that the resistance of the VO2(B) sensor fluctuated significantly during the dynamic test and the response time fluctuated in a wide range of 63-85 s, while the V2O5 sensor showed a slight fluctuation in the resistance and a narrow range in the response time (14-22 s). Clearly, the V2O5 sensor showed superior reproducibility compared to the VO2(B) sensor. In addition, a mass of repeated measurements revealed the response sensitivity of both sensors decreasing after long-term work, and if the sensor was kept in air for about two hours, the response sensitivity would still recover to the original value. This phenomenon can be perfectly interpreted by interactions of surface hydroxyl Further dynamic testing procedures were carried out, which provided more information on the most important parameters for a sensing device: sensitivity, response and recovery time, and reproducibility. Figure 7a,b show the dynamic responses of VO 2 (B) and V 2 O 5 sensors for 10 cycles where the dynamic mesurement was performed between air and 550 ppm ammonia at room temperature. For clear observation, only 10 cycles are given. Similarly, both VO 2 (B) and V 2 O 5 sensors indicate good reproducibility, as revealed in the repeated measurements (Figure 7a,b). When sensors of VO 2 (B) and V 2 O 5 were exposed to 550 ppm ammonia, the ammonia adsorption was triggered, and simultaneously the sensor resistance increased abruptly and then reached a relative stable value. When the sensors were switched to air again, ammonia desorption happened, and the sensor resistance decreased abruptly and then reached a relative stable value. The response time and recovery time for the VO 2 (B) sensor (defined as 95% of the time between the peak value and the valley value) were 63-85 s and 11-16 s (Figure 7c), respectively. Comparatively, the V 2 O 5 sensor indicated a faster response: the response time was 14-22 s and the recovery time was 14-20 s (Figure 7d). Compared with other ammonia sensors of VO 2 (B) and V 2 O 5 reported previously, the sensors in this work showed a fast response/recovery rate ( Table 2). As shown in Figure 7c, it is clear that the resistance of the VO 2 (B) sensor fluctuated significantly during the dynamic test and the response time fluctuated in a wide range of 63-85 s, while the V 2 O 5 sensor showed a slight fluctuation in the resistance and a narrow range in the response time (14-22 s). Clearly, the V 2 O 5 sensor showed superior reproducibility compared to the VO 2 (B) sensor. In addition, a mass of repeated measurements revealed the response sensitivity of both sensors decreasing after long-term work, and if the sensor was kept in air for about two hours, the response sensitivity would still recover to the original value. This phenomenon can be perfectly interpreted by interactions of surface hydroxyl groups with ammonia. As revealed in previous studies, the humidity in air can improve the ammonia adsorption and resultantly increase the sensor response [46,47]. A long-term work could decrease the site activities of adsorbed hydroxyl groups on the surface, leading to the sensitivity decreasing. After a few hours in air, the ammonia retained on hydroxyl groups should be removed during air purging, thus causing sensitivity recovery. Nanomaterials 2019, 9, x FOR PEER REVIEW 9 of 12 groups with ammonia. As revealed in previous studies, the humidity in air can improve the ammonia adsorption and resultantly increase the sensor response [46,47]. A long-term work could decrease the site activities of adsorbed hydroxyl groups on the surface, leading to the sensitivity decreasing. After a few hours in air, the ammonia retained on hydroxyl groups should be removed during air purging, thus causing sensitivity recovery.

Conclusions
In summary, self-assembled VO2(B) nanoflakes were synthesized via a simple hydrothermal method, and VO2(M) and V2O5 nanoflakes were obtained through a high-temperature phase transition in nitrogen and air, respectively. Sensors based on the three famous vanadium oxide

Conclusions
In summary, self-assembled VO 2 (B) nanoflakes were synthesized via a simple hydrothermal method, and VO 2 (M) and V 2 O 5 nanoflakes were obtained through a high-temperature phase transition in nitrogen and air, respectively. Sensors based on the three famous vanadium oxide compounds were fabricated and their gas-sensing characteristics were comparatively investigated at room temperature. It was found that VO 2 (M) nanoflakes were gas-insensitive, while both VO 2 (B) and V 2 O 5 nanoflakes were highly selective to ammonia. As an ammonia sensor, V 2 O 5 nanoflakes showed higher sensitivity, faster response, and better reproducibility to ammonia than VO 2 (B) nanoflakes. The sensitivity was about one order of magnitude higher than that of VO 2 (B) nanoflakes, and the response time and recovery time were 14-22 s and 14-20 s, respectively. Interestingly, although vanadium oxides are generally regarded as n-type semiconductors, both VO 2 (B) and V 2 O 5 nanoflakes showed p-type sensing responses to ammonia, which can be attributed to the surface inversion layer formation.