Crystal growth during operation at the elevated working-temperatures (250–500 °C) of metal-oxide gas sensors is considered to contribute to the drift of their baseline (film resistance without the analyte) and poor long-term stability of their response [2,3]. The BET and XRD size were within 1 nm suggesting formation of mainly monocrystalline particles. Figure 1 shows the average crystal size of large (triangles) and small (squares) SnO₂ nanoparticles as a function of the sintering time at 400 °C. The crystal size of the better performing, small SnO₂ nanoparticles (Figure 1, squares) increased from 12 to 22 nm with increasing sintering time from 0 to 24 h. A crystal size of 21.6 nm was obtained already after 12 h sintering (Figure 1, squares). In contrast, the crystal size of the large SnO₂ nanoparticles (Figure 1, triangles) increased only from 21.9 to 24.2 with increasing sintering time from 0 to 24 h. This indicates that, for sensor operation at 400 °C, flame-made nanoparticles constituted by pure SnO₂ crystals have a thermodynamically stable size of nearly 24 nm in agreement with the poor thermal stability of small SnO₂ nanoparticles and with the grain size stability conditions reported for several other synthesis methods [3]. This is in line with the rapid crystal and grain growth of flame-made SnO₂ nanoparticles observed already at low sintering temperatures [7] suggesting that obtaining stable sensor responses requires testing of the sensors for several consecutive days. In particular, the asymptotic-like growth of the small SnO₂ nanoparticles (Figure 1, squares) toward 22 nm suggests that small drift of sensor response and baseline may continue for a very long time span (>>24 h). In fact, the thermodynamically stable crystal size (at 400 °C) of 24 nm was still not obtained upon 24 h sintering.

Cosynthesis of SiO₂ led already at very low content (1.4 wt%) to remarkable long-term thermal-stabilization of the small SnO₂ nanoparticles. The average crystal size of these SnO₂-SiO₂ nanocomposites (Figure 1, circles) increased only from 9.5 to 10.5 nm with increasing sintering time from 0 to 24 h. The particles were mainly polyhedrical (Figure 2(c,d)) consisting of a crystalline SnO₂ core and some dispersed SiO₂ phase. At high SiO₂ content, the SnO₂ and SiO₂ phase were segregated in crystalline and amorphous domains, respectively. This is in line with the reported thermal stabilization and performance maximization of SnO₂- [7] and WO₃-based [19] gas sensors by Si-doping. Here, it is shown that this noble metal-free approach to improve the performance of metal-oxide chemoresistive gas sensors offers also superior long-term structural stability.

The thermal stabilization mechanism of the small SnO₂ crystals by SiO₂ cosynthesis was further investigated by electron diffraction analysis, XRD and TEM analysis of the nanoparticles upon sintering for 4 h at 400 °C. In line with previous results [7], increasing the SiO₂ content increased the homogeneity of the visible SnO₂ TEM size both for the as-prepared and sintered samples. The electron diffraction (ED) patterns of the small SnO₂ nanoparticles (Figure 2(a)) showed the presence of crystalline structures in line with the XRD analysis [7] that corresponded to 100 wt% cassiterite phase [16] while the numerous bright spots indicate the formation of large crystals already upon short (4 h) sintering at 400 °C. This is in line with the rapid increase in average crystal size of the small SnO₂ nanoparticles (Figure 1, squares) with increasing sintering time and suggests a polydisperse crystal size distribution. In contrast, the ED patterns of the 1.4 wt% SnO₂-SiO₂ nanocomposites (Figure 2(b)) did not shown nearly any bright spot. This indicates that the formation of large crystals is homogeneously inhibited by cosynthesis of SiO₂. In this respect, a possible thermal stabilization mechanism is the pinning of the SnO₂ crystal boundaries by SiO₂ [13,14]. Condensation of the SiO₂ molecules on the surface of the nucleated SnO₂ clusters during flame-synthesis may explain the lower sintering rates of these SnO₂-SiO₂ nanoparticles already at very low SiO₂ content. In fact, it is expected that SiO₂ segregates from the SnO₂ already at low content (ca. 2 wt%) [7].

This homogeneous inhibition of SnO₂ crystal growth is necessary to achieve long-term stability of the sensing properties. In fact, irregular growth of some un-stabilized SnO₂ nanoparticles would also lead to change in the structural and electronic properties of the sensing film. However, the crystal size of the SnO₂-SiO₂ nanocomposites (Figure 1, circles) was not completely stable and approached slowly 10.5 nm. Although, this crystal growth is very small with respect to that of pure SnO₂ (Figure 1, squares), it still indicates a restructuring of the nanoparticle interface. In particular, growth of sintering necks may drastically change the performance of the sensing film while showing very small variations in the measured crystal size [7].

Reduction of the long-term drift of the SnO₂ and SnO₂-SiO₂ sensors may be obtained by pre-sintering of the films at temperatures above the operational ones (e.g., at 600 °C) [7] leading to the achievement of a thermodynamically stable grain size prior to sensor testing (e.g., at 400 °C). Figure 3 shows the average SnO₂ crystal size of several SnO₂-SiO₂ nanocomposites as a function of such a pre-sintering step at 600 °C. The 1 wt% SnO₂-SiO₂ crystal size (Figure 3, triangles up) increased from 9.7 to 15.4 with increasing sintering time at 600 °C from 0 to 24 h. This shows that even the smallest addition of SiO₂ leads to stabilization of the SnO₂ crystal size far below that of pure SnO₂ at 400 °C (Figure 1, solid triangles). In particular, the 1.4 wt% SnO₂-SiO₂ reached a size of 11.4 nm already after 4 h sintering (Figure 3, circles). This is more than the thermodynamically stable size (≈10.5 nm) at 400 °C and thus pre-sintering of the sensing films prior to sensor utilization may be utilized to considerably shorten the time required for achievement of a stable sensor response. Furthermore, up to 4 wt% SiO₂, the as-prepared SnO₂ crystal size (Figure 3) of these nanocomposites was very close (ca. 10 ± 1.5 nm) suggesting further that SiO₂ may condense directly on the formed SnO₂ nanoparticles inhibiting further crystal and grain growth during flame-synthesis. In comparison, the initial crystal size of the pure SnO₂ nanoparticles was 12 nm (Figure 1, solid squares) which is attributed to particle coagulation during the residence time in the flame. In line, the as-prepared powder SSA increased from 100 to 211 m²/g with increasing SiO₂ content from 0 to 15 wt%. The 15 wt% SiO₂-SnO₂ demonstrated the highest long-term stability growing only from 4.5 to 5 nm (Figure 3, diamonds) with increasing sintering time from 0 to 24 h. This is in agreement with the grain growth inhibition demonstrated by SiO₂ cosynthesis [7]. However, utilization of such high SiO₂ contents results in the formation of insulating domains and a drastic drop of the sensing performance [7] and thus, here, the dynamics of the sensor response stabilization has been investigated at low SiO₂ content (1–4 wt%).

The sensing properties of these SnO₂-SiO₂ nanoparticles were tested with EtOH, a standard volatile organic compound that is particularly important for detection of drunken drivers and is increasingly investigated also for non-invasive breath analysis. Figure 4 shows the response to 10 ppm ethanol of a pure SnO₂ (d_XRD = 12 nm) gas sensor, that was not stabilized by a pre-sintering step, for several operation temperatures. The response of this sensor (Figure 4) decreased considerably with increasing operation temperature from 220 to 320 °C. This is surprising as pure SnO₂ has maximal response to EtOH at around 300–350 °C [20]. The drop in the sensor response was attributed to the sintering of the SnO₂ nanoparticles already during operation at such moderate temperatures. This is in line with the measured crystal growth of the small SnO₂ nanoparticles (Figure 1, solid squares) that is expected to drastically reduce their sensitivity [3,5,7]. Additionally, operation of the SnO₂ sensor at 220 °C was characterized (Figure 4, dotted line) by an unstable response and it was not possible to fully recover the initial baseline. This indicates that without prior stabilization the sensing behavior of pure SnO₂ nanoparticles is characterized by very poor long-term stability. After two days at 320 °C (Figure 4, solid line), the sensor properties were considerably more stable demonstrating a well-defined response to 10 ppm EtOH and full recovery of the initial baseline. Nevertheless, increasing the EtOH concentration to 30 and 50 ppm (Figure 5) resulted in very long response times.

The step-wise increase of the EtOH concentration (Figure 5) from 10 to 30 ppm at 320 °C demonstrated sufficient sensor sensitivity for discerning among small (<10 ppm) EtOH variations. Furthermore, a very good recovery of the initial baseline was observed (Figure 5) in line with the single EtOH step at 320 °C (Figure 4, solid line). However, the sensitivity of this SnO₂ sensor to EtOH was pretty low with respect to noble-metals [6] or metal-oxide [7] doped nanoparticles. This was also attributed to the growth of the SnO₂ crystals during operation and that was found to undermine the reproducibility of the sensor performance. To have more stable sensing properties, novel sensors made of pure SnO₂ nanoparticles (d_XRD = 12 nm) were also tested for EtOH sensing after a sintering step at 600 °C (12 h). Their performance was compared to the sensors (Figure 4 and Figure 5) with the as prepared nanoparticle films.

Figure 6 shows the response to increasing EtOH concentrations of a SnO₂ sensor made of small nanoparticles that was sintered at 600 °C for 12 h prior to gas sensing. The sensor shows (Figure 6) similar response and recovery time to that having an as prepared SnO₂ nanoparticle film (Figure 5). This is in agreement with the identical surface composition and film thickness of the as prepared and sintered SnO₂ films. However, the magnitude of the sensor response was drastically increased with respect to the latter. In particular, the sintered sensor (Figure 6) had a response of nearly 17 to 10 ppm of EtOH while the as prepared barely reached 8. This is surprising as high temperature sintering is expected to increase crystal size leading to lower sensitivity [3,5]. Here, it is suggested that the enhancement of the sensor response arise from the formation of sintering necks with size below that of the main grains between the SnO₂ particles. Formation of partially or fully depleted sintering necks can increase the sensitivity of metal-oxide gas sensors and can hardly be measured by XRD or nitrogen adsorption [7]. As a result, two instability mechanisms are suggested for the pure SnO₂ nanoparticles. A first, where the average crystal (and grain) size is increased (Figure 4) resulting in a drop of the sensor response, and a second, where partially depleted sintering necks are formed increasing the sensitivity. Similar effects were observed for size selected SnO₂ agglomerates [21]. There, very small changes in the sintering properties of the agglomerates that could hardly be tracked by XRD analysis led to drastic variations in their sensing response to EtOH. Both dynamics can be accelerated by a pre-sintering step leading to (Figure 6) higher response and more stable sensing properties.

The sensing dynamics of the SnO₂-SiO₂ composites was different from that of the pure SnO₂ nanoparticles (Figure 4, Figure 5 and Figure 6). To better investigate the effect of SiO₂ cosynthesis on the sensing properties, all synthesis parameters were kept constant and only the Si-content was varied. Figure 7 shows the response to step-wise increases in EtOH concentration of a sensor made of as prepared 1.4 wt% SnO₂-SiO₂. The response and recovery times of these nanocomposites (Figure 7) were comparable to that of the pure SnO₂ (Figure 5). Nevertheless, the magnitude of the sensor response was initially lower than that of the SnO₂ nanoparticles (Figure 5) reaching about 37 at 50 ppm EtOH (Figure 7). This is in contrast to the smaller crystal size of these nanocomposites (Figure 3, triangles up) that should lead to higher sensitivity [5]. This is attributed to the sintering inhibition effect of the SiO₂ that may have limited the growth of sintering neck between the main SnO₂ grains.

After sintering these films at 600 °C for 12 h, their response was greatly increased (Figure 8). More in details, upon this stabilization step, the response of the 1.4 wt% SnO₂-SiO₂ to 50 ppm EtOH increased from 37 (Figure 7) to 153 (Figure 8). This 4 fold increase in sensitivity is in line with the reported optimal Si-doping of SnO₂ nanoparticles [7]. Furthermore, it confirms the long-term instability mechanisms observed for pure SnO₂. As the presence of SiO₂ drastically inhibit the crystal growth, the sintering necks formed at 600 °C are smaller and thus more depleted than for pure SnO₂(Figure 5 and Figure 6) resulting in a more drastic enhancement of their sensing performance (Figure 7 and Figure 8). A more detailed analysis of the neck morphologies and growth dynamics is required to quantitatively describe the sensing response enhancement of these nanocomposites [4]. Higher SiO₂ contents (Figure 3), up to 4 wt%, resulted in a similar enhancement of the sensing properties. Overall, cosynthesis of SiO₂ increases the variation of sensor resistance during injection of EtOH concentration as it was previously investigated in details [7].

The stability of the sensor response toward variations in relative humidity is of major importance for several applications [2,22]. Doping of SnO₂ nanoparticles with Ti has been reported to drastically decrease their cross-sensitivity to humidity [9]. Recently, this has been attributed to thermodynamically dictated enrichment of the SnO₂ surface with Ti atoms [23] that lower the adsorption energy of dissociatively adsorbed H₂O [24]. The effect of SiO₂ cosynthesis on the cross-sensitivity (CS) to humidity of SnO₂ nanoparticles, however, has not been investigated yet. This is particularly important as flame-made SiO₂ has high surface concentration of hydroxyl groups that facilitate the binding of H₂O molecules [10] and thus may result in a strong enhancement of the CS to humidity. Figure 9 shows the CS to humidity during EtOH detection, defined as change of the sensor response in dry air (Equation (2)), of the pure SnO₂ (triangles solid), 1 wt% (empty triangles) and 2.5 wt% (empty squares) SnO₂-SiO₂ nanocomposites as a function of the relative humidity. The CS to humidity of the pure SnO₂ sensors (Figure 9, solid triangles) increased from 51 to 74% with increasing r.h. from 20 to 60%. This is in agreement with the drastic change in sensor response reported for SnO₂ nanoparticles with increasing r.h. content [9]. The continuous increase in CS above 20% r.h. (Figure 9, solid triangles) indicates that the SnO₂ surface has not yet been saturated with adsorbed H₂O species. More important, the CS of both SnO₂-SiO₂ sensors (Figure 9, empty squares and triangles) was comparable to that of the pure SnO₂ (Figure 9, solid triangles). This is different than the effect of Ti-doping [7] and in contrast to the super-hydrophilic properties of flame-made SiO₂ [10]. However, as SiO₂ is an isolator, localized SiO₂ molecules/clusters on the SnO₂ surface may act as active sites for H₂O binding but still have minimal impact on the sensing properties of the SnO₂ nanocrystals due to its inefficient electron conduction properties. As a result, SiO₂ cosynthesis leads to the same CS than pure SnO₂ nanoparticles. This is in contrast to the modification of SnO₂ crystals with hydrophilic zeolites [25,26] where notable variations from the sensing response of the pure SnO₂ were observed. Minimization of the CS while improving the long-term stability and sensitivity of SnO₂-based gas sensors may be achieved by synthesis of Sn_1−xTi_xO₂-SiO₂ nanocomposites [9].

It should be pointed out that the sensing chamber geometry has considerable influence on the resulting sensor response dynamics as previously discussed in details [27]. The setup utilized here was previously tested for several other materials [7,9,15,17] and, for the utilized total gas flow (1 L/min), results in reaction limited sensor responses [27].