The experimental setup for the synthesis of pure SnO₂, 0.2–3 wt%Ru/SnO₂ nanopowders by FSP is shown in Figure 1. The flame-spray-made (5/5) pure SnO₂ was designated as P0 while the SnO₂ nanopowders doped with 0.2, 0.6, 1, 2, and 3 wt%Ru were designated as P0.2, P0.6, P1, P2, and P3, respectively. Precursor solutions (0.5 M) were prepared by dissolving appropriate amounts of tin (II) 2-ethylhexanoate (Aldrich, 95%) and ruthenium (III) acetylacetonate (Aldrich, 97%) used as Sn and Ru precursors in xylene (Carlo Erba, 98.5%), respectively. In a typical run, the precursor mixture was fed into a nozzle at a constant feed rate of 5 mL/min using a syringe pump. At the end of the nozzle the precursor solution was dispersed by 4.30 L/min oxygen forming a spray with a pressure drop at the capillary tip kept constant at 1.5 bars by adjusting the orifice gap area. A sheath gas flow of 3.92 L/min of O₂ was issued concentrically around the nozzle to stabilize and contain the spray flame. The spray was ignited by supporting flamelets fed with oxygen (2.46 L/min) and methane (1.19 L/min) which are positioned in a ring around the nozzle outlet. The observed flame height was approximately 10-12 cm, and it increased slightly with increasing combustion enthalpy. The combustion enthalpies are directly dependent on the particular solvent, starting materials, and dopants used. Pure SnO₂ samples show an light orange and Ru doped samples show light pink color in the base and middle of the flame, and also light orange on the top of the flame, as shown in Figure 2. After evaporation and combustion of precursor droplets, particles are formed by nucleation, condensation, coagulation, coalescence, and Ru deposit on the SnO₂ support. Finally, the nanoparticles were collected on a glass microfibre filters (Whatmann GF/A, 25.7 cm in diameter) with the aid of a vacuum pump (Busch, Seco SV 1040C).

The powder phases were analyzed by X-ray diffraction (XRD) [Phillips X-′pert] using CuKα radiation (20 kV, 20 mA) with a scanning speed of 5°/minute. The specific surface areas of the nanopowders were obtained from BET measurements [Autosorb 1 MP, Quantachrome]. All samples were degassed at 120 °C for 2 h prior to analysis. The diameter of particles were calculated from d_BET = 6/SSA_BET × ρ_sample, where SSA_BET is the specific surface area (m²/g), ρ_samples are the average density of SnO₂ (ρ_SnO2 = 6.85 g/cm³ [1]) and the density of ruthenium (ρ_Ru = 10.65 g/cm³ taken into account for their weight content of different doping [26]). The accurate morphologies of the nanoparticles and cross-section structures of sensor were analyzed by TEM [JSM-2010, JEOL], SEM [JSM-6335F, JEOL], and EDS analyses.

An appropriate quantity of 0.28 mL homogeneous mixed solution was prepared by stirring and heating at 80 °C for 12 hr with ethyl cellulose (Fluka, 30–70 mPa.s) as the temporary binder and terpineol (Aldrich, 90%) as a solvent. The liquid mixture was combined with 60 mg samples of the P0, P0.2, P1, and P3 nanopowders and mixed for 30 min to form a paste prior to spin-coating. The resulting paste was firstly spin-coated at 700 rpm for 10 s, and then subsequently at 3,000 rpm for 30 s on the Al₂O₃ substrates interdigitated with Au electrodes (0.5 × 0.5 cm) to deposit sensing films. The resulting substrates were annealed in an oven at 150 °C for 1 h with an annealing rate of 1 °C/min and at 400 °C for 1h with an annealing rate of 1 °C/min for binder removal prior to the sensing test [28].

The sensor characteristics of sensing films were characterized toward the high concentration of H₂ gas (500–10,000 ppm). The flow through technique was used to test the gas-sensing properties of sensing films. A constant flux of synthetic air of 2 L/min as gas carrier was flown to mix with the desired concentration of pollutants dispersed in synthetic air. All measurements were conducted in a temperature-stabilized sealed chamber at 20 °C under controlled humidity. The gas flow rates were precisely manipulated using a computer controlled multi-channel mass flow controller. The external NiCr heater was heated by a regulated DC power supply to different operating temperatures. The operating temperature was varied from 200 °C to 350 °C. The resistances of various sensors were continuously monitored with a computer-controlled system by voltage-amperometric technique with 10 V DC bias and current measurement through a picoammeter. The sensor was exposed to a gas sample for ∼5 minutes for each gas concentration testing and then the air flux was restored for 15 minutes. The sensitivity (S) is defined in the following as the resistance ratio R_a/R_g [11,27-30], where R_a is the resistance in dry air, and R_g is the resistance in the test gas. The response time (T_res) is defined as the time required until 90% of the response signal is reached. The recovery time (T_rec) denotes the time needed until 90% of the original baseline signal is recovered. After the sensors fabricated using samples P0, P0.2, P1, and P3 had been tested with varied the operating temperatures, they were designated as S0, S0.2, S1, and S3, respectively. Finally, the morphologies, film thickness of sensing layers and elemental compositions were further analyzed by SEM and EDS line-scan mode analyses.

Figure 3(a) shows the XRD patterns of flame-spray-made pure SnO₂ and 0.2–3 wt%Pd/SnO₂ nanopowders. All samples were highly crystalline, and all peaks can be confirmed to be the cassiterite-tetragonal phase of SnO₂, which matched well with the JCPDS file No. 77-447. Ru peaks were not found in these patterns (JCPDS file No. 6-663). It can be assumed that the amount of Ru concentration was very low, which affected the appearance of the Ru peaks.

The diffraction peak for 0.2 wt% Ru/SnO₂ nanopowder was the broadest compared to other doping levels, suggesting relatively well-dispersed smaller Ru particles. As the Ru concentration increased, all peaks were slightly sharpened and increased in intensity, indicating that the poor-dispersion of larger Ru particles leads to rough agglomeration at higher Ru doping levels. These results were consistent with the BET data, as shown in Figure 3(b). The specific surface area (SSA_BET) drastically increased from 141.6 m²/g (bare SnO₂) to 183.8 m²/g (0.2 wt%Ru/SnO₂). When the Ru concentration increased (0.2 to 3 wt%Ru), the SSA_BET were found to linearly decrease (183.8 to 113.5 m²/g), with an increase in the average BET-equivalent particle diameter (d_BET) (bare SnO₂: 6.2 nm, 0.2–3 wt%Ru/SnO₂: 4.7 to 7.6 nm). This trend was consistent with Niranjan et al. [11] who studied the effect of Ru concentration on crystalline SnO₂ nanoparticles. To explain this result, it can be speculated as follows: during the processes of Ru particle formation and deposition on the particle support (SnO₂) in the flame, the Ru created a new nucleation center, which in turn changed the nucleation type from homogeneous to heterogeneous, and deteriorated the deposition formation leading to the agglomeration of the tiny Ru particles at high doping levels. This can be confirmed from the accurate morphology by TEM bright-field images. The FSP afforded small Ru particles attached to the surface of the supporting SnO₂ nanoparticles indicating a high SSA_BET. The well-dispersed flame-made 0.2 wt%Ru/SnO₂ nanoparticles were confined to the SnO₂ surface. The larger crystallite diameters indicate clumping and clusters of Ru, translating into a poor dispersion of the Ru nanoparticles on SnO₂ support which affected to the decrease of the SSA_BET. The SEM micrograph [Figure 4(a)] and the elemental compositions of the agglomerated nanoparticles formed with the sample with the highest Ru concentration (P3) are shown by the EDS spectra in Figure 4(b). Interestingly, the analyzed square regions [Figure 4(b)] were composed of the agglomerated nanoparticles, the copper grid, and gold sputtering prior to an analysis. The EDS spectra showed elemental compositions rich in copper (Cu), caused by the contamination of copper foil, poor gold (Au) caused by the contamination of gold sputtering which used to prepared the samples prior to an analyzing, tin (Sn), oxygen (O), and poor ruthenium (Ru) elements.

Figure 5 (a–f) show TEM bright-field images of P0-P3. The corresponding diffraction patterns are shown in the insets. The diffraction patterns illustrated spot patterns corresponding to the tetragonal-cassiterite structure of SnO₂, indicating the SnO₂ nanoparticles were highly crystalline, in good agreement with the XRD data. The TEM bright-field images of the FSP (5/5)-made nanoparticles, indicate polyhedral aggregates of primary particles. The morphologies of flame made (5/5) SnO₂ and 0.2–3 wt% Ru/SnO₂ nanoparticles contained mainly spherical particles, with diameters ranging from 3–10 nm, with occasional rectangular, hexagonal (3–10 nm) and rod-like (3–5 nm in width, and 5–20 nm in length) particles. Ru nanoparticles were not found in these micrographs. This is because Ru is very small when compared with the size of SnO₂ nano-support. The primary particle diameters observed by TEM were consistent with the d_BET. From these data, it can be clearly seen that the amount of Ru concentrations would not affect to change the size of SnO₂ nanoparticles. We could assume this doping formation from the Hume-Rothery rules [31-33], which commonly used to explain the solid mixtures called solid solutions.

In the doping of materials, atoms of the solvent (host material; Sn) are successfully replaced by the solute-atoms (the doping atom; Ru) from their lattice positions (interstitial solid solutions are not discussed here). In the other words, one material gets dissolved in the other, without disturbing the crystal structure, except for lattice distortions (expansions or compressions). For the formation of solid solutions, according to the Hume-Rothery rules, some criteria have to be fulfilled: (1) the atomic radii of the solute (Ru = 178 pm) [34] and solvent (Sn = 145 pm) [34] atoms must differ by no more than 15% (∼22.75%). If not, it is likely to have a low solubility. This is the first rule which must be considered. The atomic size factor was said to be unfavorable; (2) the solute and solvent should have similar electronegativity (Ru = 2.2, Sn = 1.8) [33], compared to the host. If the electronegativity difference is too great, the metals will tend to form intermetallic compounds instead of solid solutions. Its solubility in the host would therefore be limited, because of the so-called electronegative valency effect; (3) a metal with lower valency is more likely to dissolve one which has a higher valency, than vice versa (relative valency effect). The valence electrons are the electrons in the last shell or energy level of an atom. Maximum solubility occurs when the solvent (Sn) [35] and solute (Ru) [36,37] have the same valency. Moreover, the thermodynamic instability of the lower oxidation states of Ru was discussed by Wiley et al. [36-38] to explain their inability to synthesize oxygen deficient Ru-bearing perovskites for catalysis. Although spectroscopic data indicated that other oxidation states (Ru²⁺, Ru³⁺, Ru⁵⁺) could exist in oxides, species other than Ru⁴⁺ generally occurred in mixed-valence phases dominated by Ru⁴⁺. Exceptions exist, however, and Ru⁵⁺ and even Ru⁷⁺ occurred in oxide compounds where there were essential structural constituents and the only Ru species. For this reason, a more precise generalization that Ru⁴⁺ was the lowest valence in oxides which was not induced by the special defect equilibrium. Metals with lower valency will tend to dissolve metals with higher valency; and (4) the crystal structures of solute (Ru = hexagonal) and solvent (Sn = tetragonal) must match. Thus the size of particles in the doped sample were not affected by Ru due to the fact that Ru could not get in solid solution into the unit cell of SnO₂ crystal structure.

Figures 6(a-c) show the plot of sensitivity (S) and response times (T_res) versus hydrogen concentrations ranging from 500–10,000 ppm for the sensors S0, S0.2, S1, and S3 during a forward cycle at operating temperature ranging from 200–350 °C. It was found that the sensitivity increased with operating temperature to the maximum at 350 °C [Figure 6(c)]. Interestingly, the temperature of maximum sensitivity was found to shift towards lower Ru concentrations, which can be attributed to the effect of particles size and the specific surface area, as a result of a well-dispersed Ru incorporation into the SnO₂ matrix. At the operating temperature of 200 °C, the sensitivity of all Ru doping materials was seen to be higher than that of pure SnO₂. The sensitivity (filled symbols, left axis) increased and the response time (open symbols, right axis) decreased with increasing H₂ concentrations. Moreover, it was found that the highest Ru concentration (3 wt%) showed the best sensing performance in terms of sensitivity (S = 8.6) and response time. The response time of a 3 wt% Ru/SnO₂ sensor for 10,000 ppm at 200 °C was 16 s (open circles, right axis), which was better than pure SnO₂ (178 s) (open diamonds, right axis) and the other doping levels (0.2 wt% Ru/SnO₂ = 70 s (open rectangles, right axis), and 1 wt% Ru/SnO₂ = 22 s (open triangles, right axis)). On the other hand, both the operating temperatures of 300 °C and 350 °C had better sensing performance than 200 °C in terms of sensitivity (filled symbols, left axis) and faster response time (open symbols, right axis). Also, in the case of Ru doping the best performance was achieved at a sensor operating temperature of 350 °C. However, the situation was completely different when more Ru was added. Here, 1 wt% (300 °C) and 3 wt% Ru/SnO₂ (350 °C) also displayed evidently reduced sensing performance in terms of sensitivity. Note that these tests were performed with a set of four sensors placed in the chamber. The sensitivities of all sensors were found to increase rather linearly with increasing H₂ concentrations. As the Ru concentration increased from 0.2 to 3 wt%Ru, the lowest Ru concentration (S0.2) the sensor behaviors improved in terms of the best sensitivity (to 10,000 ppm, S = 27) (filled rectangles, left axis) and very fast response times (T_res = 6 s) (open rectangles, right axis) at 350 °C, which evidently were better than S1, and S3. The sensor S0.2 showed very fast response to H₂ gas, whereas the response of the pure SnO₂ sensor (S0) was somewhat sluggish. Figures 6(b) and 6(c) indicate the dependence of the sensitivity on the Ru concentration at an operating temperature of 300 °C and 350 °C, respectively. The amount and distribution of Ru species in the SnO₂ support were important parameters governing the sensitivity, being maximum (S = 27) at 350 °C for a SnO₂ containing 0.2 wt% Ru. The sensitivity consistently increased with increasing H₂ concentration. The role of the Ru in enhancing the sensitivity and response rate of the sensor could be due to the electronic interaction between the sensitizer and the semi-conducting material. Ru acts as a catalyst and enhances the reaction rate, especially because χ_o − χ_Ru < χ_o − χ_Sn, where χ_o represents the electronegativity value (χ_o, χ_Sn, χ_Ru = 3.5, 1.8, 2.2, respectively) [11]. Thus, when oxygen is adsorbed on the Ru zones of strong localization at elevated temperatures, the potential between the SnO₂ grains may be raised and as a result, the total resistance increases as compared with the sample without Ru. The decrease of the amount of Ru concentration leads to well-dispersed Ru on the SnO₂ surface arising from the chemisorbed oxygen species. Moreover, Figure 7(a) shows the response to high concentrations of H₂ (500–10,000 ppm) of sensors which were functionalized in situ with 0.2 wt% Ru. Doping the SnO₂ with 0.2 wt% Ru results in a much steeper calibration curve and the highest sensor signal compared to pure SnO₂ [see Figure 7(a)]. The higher sensor signal, and especially the higher sensitivity (i.e., the steeper response curve), demonstrate an enhanced sensor performance.

Figure 6 shows the selectivity histogram for 0.2 vol% of different gases at an operating temperature of 350 °C. The sensors S0 and S0.2 exhibited similar selectivity towards the flammable H₂ and C₂H₂ gases and toxic CO gas. This can be attributed to the identical reducing behavior of both gas types. The S0.2 sensor has a good gas selectivity for 0.2 vol% H₂ of 7 at 350 °C. The sensitivity of S0.2 sensor of C₂H₂ and CO gases were 2.3 and 1.8 at 0.2 vol% H₂ at 350 °C. Thus, the gas sensitivity of S0.2 sensor was higher than that of C₂H₂ and CO gases. The H₂ selectivity of S0.2 sensor was substantially higher compared to pure SnO₂ gas sensor (S0). On the other hand, C₂H₂ and CO gases sensitivity/selectivity of S0.2 sensor was also evidently deteriorated compared to that of pure SnO₂ gas sensor (S0). Ru cannot improve the sensing performance and is unsuitable for use as dopant in SnO₂ sensor for both C₂H₂ and CO gases. This is because the absorption configurations of the gas molecules and the surface fragmentation reactions on the Ru sites are responsible for the similar sensitivity values towards all gases.

The microstructures of high density Al₂O₃ (dark view) substrate interdigitated with Au electrodes (bright view) was evidently seen as the phase boundaries in Figure 8(a). The cross-section, film thickness, and surface morphology of the sensing film layer (S0.2) after a sensing test at 200–350 °C were observed using SEM analysis as shown in Figure 8(b). The film thickness of sensing film was about 10 μm, which was of tremendous benefit to the H₂ gas sensing properties. The microstructure of high density Al₂O₃ substrate was visible. The square emphasized the investigation selected area at high magnification to an aggregated of primary particles after sensing test. The particle sizes of nanoparticles slightly changed after annealing and sensing test were also shown in the inset. In addition, the trends in the elemental composition of the agglomerated nanoparticles formed of sample P0.2 was shown by the EDS line scan mode in Figure 8(c). Interestingly, the analyzed regions composed of the nanoparticles, the copper grid, and gold sputtering prior to an analysis. The line scan across the agglomerate for sensor P0.2 is indicated in Figure 8(c). The elemental-line histograms are shown as a series of solid lines corresponding to a rich in copper (Cu) caused by the contamination of copper grid, poor gold (Au), tin (Sn), oxygen (O), and ruthenium (Ru) elements. After annealing process, a denser film layer was formed. Regularities and preciseness in the film thickness stem from the spin coating technique.