Thin Solid Films 500 (2006) 302 308 www.elsevier.

com/locate/tsf

On-line monitoring of CO2 quality using doped WO3 thin film sensors
M. Stankova a, X. Vilanova a, E. Llobet a,*, J. Calderer b, M. Vinaixa a, I. Gra ` cia c, c a C. Cane , X. Correig
a

Departament dEnginyeria Electro ` nica, Ele ` ctrica i Automa ` tica, Universitat Rovira i Virgili, Avda. Pa sos Catalans, 26, 43007 Tarragona, Spain b Departament dEnginyeria Electro ` nica, Universitat Polite ` cnica de Catalunya, Barcelona, Spain c Departament de Microsistemes i Tecnologies del Silici, Centro Nacional de Microelectro nica, Bellaterra, Spain Received 18 February 2005; received in revised form 12 October 2005; accepted 9 November 2005

Abstract Thin films of either pure or doped tungsten oxide were grown by radiofrequency (rf) sputtering onto silicon micromachined substrates. Up to 7 different dopant materials (noble metals or metal oxides) were deposited by rf sputtering or by evaporation onto the tungsten oxide films. The responsiveness of the resulting micromachined sensors towards sulfur dioxide and hydrogen sulfide was studied. Other pollutants in CO2 such as ethylene and methane were also tested. It was found that Au-doped tungsten oxide sensors were highly sensitive to H2S, poorly sensitive to SO2 and almost insensitive to hydrocarbons. On the other hand, Pt-doped tungsten oxide was highly sensitive to SO2, poorly responsive to H2S and nearly insensitive to hydrocarbons. By applying a principal component analysis (PCA), we show that it would be possible to selectively detect traces of H2S and SO2 in a CO2 stream using doped WO3 microsensors. These sensors could be used in a lowcost analyzer of beverage-grade CO2. D 2005 Elsevier B.V. All rights reserved.
Keywords: Tungsten oxide; Gas sensor; Micromachined sensor arrays; Doped oxides

1. Introduction In recent years, there has been a continuous improvement towards quality assurance in the beverage industry. The International Society of Beverage Technologists (ISBT) has established quality guidelines for carbon dioxide used in beers, mineral waters and soft drinks [1]. Carbon dioxide can be produced from a wide variety of processes. Each process has the potential of leaving residues that reduce the purity of CO2. Furthermore, contaminants may be introduced from storage or transport vessels. In the quality control of CO2, the methods generally used are analytical techniques such as gas chromatography or gas chromatography coupled to mass spectrometry (GC/MS) [2]. However, these techniques are very expensive and bulky. That is why, a low cost, small size analyzer to monitor on-line the quality of CO2 would be of great interest, especially for breweries or soft drink companies, where acquiring a GC/MS system is out of question. On this basis,
* Corresponding author. E-mail addresses: mariana.stefanova@urv.net (M. Stankova), eduard.llobet@urv.net (E. Llobet). 0040-6090/$ - see front matter D 2005 Elsevier B.V. All rights reserved. doi:10.1016/j.tsf.2005.11.021

equipment based on an array of metal oxide gas sensors would be a good solution that could be installed in any beverage production plant. It is well known that atmospheric oxygen plays an important role in gas detection by metal oxide gas sensors [3,4]. Kohl [5] reported some surface processes for the detection of reducing gases, which did not require the presence of atmospheric oxygen. In the last years, several groups have investigated the sensing mechanism of metal oxides operated under low oxygen concentrations [6] and in streams of inert gases such as argon [7]. In 2003, Vilanova et al. [8,9] developed a multisensor system (tin oxide sensors doped with noble metals) to detect methane, ethylene and sulfur dioxide in a CO2 stream. However, these sensors (based on SnO 2) showed low sensitivity to sulfur compounds, especially to H2S. That is why, a system based on WO3 sensors (pure and doped with platinum) was developed for the detection of sulfur species in carbon dioxide [10]. In this paper, we further study the detection of sulfur compounds in CO2 by analyzing the effect of 7 different dopants on tungsten oxide based sensors. The objective is the selective detection of these compounds at a low operating temperature (below 300 -C).

M. Stankova et al. / Thin Solid Films 500 (2006) 302 308

303

2. Experimental details 2.1. Sensor fabrication Four element microhotplate sensor arrays were fabricated. Details on the design and fabrication can be found in a previous article [10]. The active layer of WO3 was radiofrequency (rf) sputtered and patterned by lift-off. The deposition was made using a target of W (99.95% of purity) that was fixed at 70 mm from the substrate. The substrate was kept at room temperature during the process. The sputtering atmosphere consisted of 1:1 Ar/O2. The forward input power was maintained at 200 W with zero reflected power. The pressure in the chamber during deposition was 0.5 Pa. After the deposition of tungsten trioxide, the active layer was doped. As doping materials, the following compounds were used: Pt, Au, Ag, Ti, SnO2, ZnO and ITO (indium tin oxide). Most of them (Pt, Au, SnO2, ZnO and ITO) were deposited by rf magnetron sputtering. The conditions for these processes are summarized in Table 1. For the deposition of Ag and Ti, an e-gun evaporation was employed. In this last case, the target was placed 17 cm away from the sample. The thickness of each dopant was below 10 nm (layer thickness was estimated by a 4-probe resistance measurement system). As a final step, the active layer was annealed in dry air for 2 h at 400 -C. 2.2. Structural analysis

Table 2 Typical contaminants with their maximum concentrations allowed in 99.95% purity CO2 (after ISTB [1]) Pollutant Sulfur dioxide Hydrogen sulfide Other sulfur compounds Nitrogen dioxide Nitric oxide Ammonia Nitrogen Carbon monoxide Benzene Methane Heavy hydrocarbons Volatile hydrocarbons Total aldehydes Oxygen Water Maximum allowed concentration 2 ppm 500 ppb 500 ppb 2.5 ppm 2.5 ppm 2 ppm 40 ppm 2 ppm 20 ppb 30 ppm 1 ppm 20 ppm 200 ppb 9 ppm 8 ppm

geometry and vertical h h goniometer) fitted with a grazing incidence (x = 0.52-) attachment for thin film analysis and scintillation counter as a detector. The angular 2h diffraction range was between 21.0- and 70.0-. The data were collected with an angular step of 0.05- at 3 s per step and sample rotation. CuKa radiation was obtained from a copper X-ray tube operated at 40 kV and 30 mA. 2.3. Gas sensitivity measurements

The morphology of platinum-doped WO3 active layers was studied by transmission electron microscopy (TEM). The aim of this analysis was to study the surface distribution of dopants and to estimate the thickness of the layer. For this purpose, a film of Ptdoped WO3 was deposited directly on a Si wafer. A Si wafer was used instead of a micromachined sensor substrate because the latter would have broken during the measurement process. The film was annealed at 400 -C in air for 2 h. Two samples were prepared for TEM analysis: one of them was cross-sectional and the other was planar. Using the first one, we could estimate the thickness of the layer, while in the second, the surface distribution of the doping material could be observed. Micrographs were taken using a Hitachi H-800-MT microscope equipped with a Gatan Multiscan camera, working at 200 kV. The same equipment was applied to make Selected Area Electron Diffraction (SAED). The WO3 phase was determined also by using X-ray diffraction (XRD). XRD measurements were made using a Siemens D5000 diffractometer (Bragg Brentano parafocusing
Table 1 Deposition conditions used to sputter some of the dopant materials Dopant Pt Au SnO2 ZnO ITO Target Pt (99.99%) Au (99.99%) Sn (99.95%) Zn (99.99%) ITO (99.99%) Ar/O2 (sccm) 14.6/0 14.6/0 7/7 7/7 14/0 Power (W) 50 100 100 100 50 Time 10 s 30 s 2 min 30 s 10 min 4 min

The sensors were introduced in a temperature controlled, 16-ml chamber. First, pure CO2 (90 ml/min) flowed thought the chamber and the sensor baseline was established. Then the stream was switched (by the means of electrovalves) to CO2 with a given contaminant (from calibrated bottles). During all the measurements, the sensor resistance was acquired and stored for further processing. Four contaminants were tested, namely, 1 ppm SO2, 100 ppb H2S, 30 ppm CH4 and 20 ppm C2H4. These concentrations were chosen based on the maximum allowed in beverage-grade CO2 according to ISBT (see Table 2). The sensors were operated at 110, 200 and 260 -C. Each pollutant was measured at the different operating temperatures. Every measurement was replicated ten times. 3. Results and discussion 3.1. Active layer characterization First, the cross-sectional sample was studied (Fig. 1(a)). From this image, the thickness of the Pt film was estimated to be 3 4 nm. This thickness was less than the expected value (about 10 nm). The expected value was extrapolated based on previous depositions on Si wafers and four-point measurements of layer resistance. It is well known that growth rate is not a linear function of time. Our deposition process took 10 s and, in such a short interval, it is very difficult to control the growth process. We assume that this is the reason to obtain a layer thinner than expected. However, we were interested in obtaining a very thin

The sputtering pressure was 5 Pa for all the processes. sccm standard cubic centimetre per minute.

304

M. Stankova et al. / Thin Solid Films 500 (2006) 302 308

Fig. 2. Mean diameter for the Pt grains from Fig. 1(a).

Fig. 1. (a) Cross-section and (b) planar TEM images of the Pt/WO3/Si sample.

layer, because the deposition of a superficial and porous Pt film onto tungsten trioxide was sought (to allow gases to react with the tungsten oxide underneath). From Fig. 1(a), it is obvious that the desired catalytic layer was obtained. Moreover, if the resulting Pt film were not so thin, then adhesion problems would have appeared, since platinum is prone to suffer from bad adhesion. A very thin Pt layer is appropriate for catalyzing reactions at the surface of the metal oxide film. In the second step, the planar sample was analyzed (Fig. 1(b)). The aim, in this case, was to study the distribution of platinum on the active layer surface. In this figure, the Pt grains can be observed. A statistical study on grain size distribution was conducted. To do so, the manual measurement of 100 grains within each TEM micrograph was performed using an appropriate program tool from Digital Micrograph (Gatan Inc.). An automatic particle analysis using standard image analysis software was not considered here because the edges of some particles were not sharply defined, which resulted in overestimating grain size. The grain size measurements performed, i.e. 100 measurements in total, underwent a standard statistical analysis. Considering that d is the mean diameter of the grains, S.D. is the standard deviation and S.E.M. is the standard error of the mean, the values obtained from the statistical analysis were as follows: mean diameter d = 2.63625 nm; S.D. = 0.38718 and S.E.M. = 0.03872. Its can be seen that the mean diameter of grains is very close to the layer thickness. The grain size distribution is shown in Fig. 2. SAED gave us a diffraction pattern formed by the elastic scattering of an electron beam by the atoms in the specimen. We applied this technique to obtain information about the

crystalline phase of the films. From the diffraction patterns and by identifying the spots and rings, two different states can be distinguished (see Fig. 3). The first one corresponds to monoclinic WO3 and the second-to cubic Pt. In Table 3, the spacing distances obtained from the diffraction patterns (the spots marked in Fig. 3) and the ones found in the databases (Joint Committee on Power Diffraction Standards: XRD tables of the International Center for Diffraction Data) [11] are compared. We used the database for triclinic and monoclinic WO3, platinum and platinum tungsten (Pt2W). The last one was used because the possibility of this compound being present was not discarded. A good accordance between the data for monoclinic (Pc) tungsten oxide and cubic platinum was obtained. It is well known that tungsten trioxide experiments several phase transitions with temperature [12,13]. It has been reported that WO3 is tetragonal above 720 -C, orthorhombic from 320 to 720 -C, monoclinic from 17 to 320 -C, and triclinic below 17 -C. A coexistence of triclinic and monoclinic phase is also possible. For example, the spot labeled 4 could belong to the triclinic or monoclinic phase. The experimental d -spacing value (see Table 3) is close to the value for the triclinic or the monoclinic phase (1.8222 and 1.8195, respectively). Fig. 4 shows half rings at the positions where the diffraction rings were identified by the software ProcessDif-

Fig. 3. Diffraction pattern for WO3 doped with Pt.

M. Stankova et al. / Thin Solid Films 500 (2006) 302 308 Table 3 Comparison of experimentally obtained d -spacing with those collected by International Centre for Diffraction Data (ICDD) (h k l are the Miller indexes) Spot no. 1 2 3 4 5 6 7 d -spacing ) experimental (A 3.665 3.052 2.304 1.823 1.512 1.398 1.205 d -spacing ) ICDD (A 3.6892 3.0595 2.27 1.8195 1.5116 1.5116 1.39 1.18 hkl 110 102 111 104 312 2 2 3 220 311 Phase WO3 Monoclinic WO3 Monoclinic Pt Cubic WO3 Monoclinic WO3 Monoclinic Pt Cubic Pt Cubic

305

The spots listed correspond to the ones marked on Fig. 3.

fraction [14]. Only diffraction rings with sharp definition are identified and analyzed. The presence of well defined spots and spots which forms arcs is due to the fact that there were two different compounds (monoclinic WO3 and cubic Pt). For some of the spots, two different planes were possible (see in Table 3, for example, the data for spot number 5). From the SAED analysis it can be concluded that the active layer consists of WO3 and a superficial doping layer, which confirms the expected result. Because of the possible presence of two phases for WO3, an XRD was performed. The result of this analysis is shown in Fig. 5. The peaks labeled on this figure correspond to monoclinic tungsten oxide. However, some of the picks characteristic of the monoclinic phase are absent and other peaks, which are characteristic of triclinic WO3, are present. This allows us to conclude that the main phase of the tungsten oxide is the monoclinic one, but some triclinic tungsten oxide coexists. 3.2. Sensor response The possibility to detect pollution by sulfur compounds (SO2 and H2S) below the maximum permissible concentrations for beverage-grade CO2 using the different sensors fabricated was investigated. The possible interference caused by other pollutants such as hydrocarbons, which are permitted at higher concentrations in CO2, was also studied. The results obtained for carbon dioxide polluted by hydrocarbons showed that most of the tungsten oxide based sensors were almost insensitive to

Fig. 5. XRD spectrum for WO3. The diffractograms for monoclinic (Pc) and triclinic (P-1) are shown for comparison.

the presence of hydrocarbons (see Table 4). On the other hand, the sensors were very sensitive to sulfur compounds. Fig. 6 presents some typical responses to the gases studied.
Table 4 Sensor response of the different sensors to the gases studied Active layer WO3 T (-C) 110 200 260 110 200 260 110 200 260 110 200 260 110 200 260 110 200 260 110 200 260 110 200 260 100 ppb H2S 3.06 3.08 3.11 1.20 1.79 1.89 4.10 11.14 13.11 3.75 5.82 6.92 1.38 1.48 1.63 1.10 1.13 1.21 1.25 1.05 1.02 1 1.05 1.07 1 ppm SO2 1.29 1.38 1.37 1.64 7.71 5.23 1.33 1.50 1.28 1.05 1.07 1.10 1.77 1.73 1.64 1.29 1.26 1.25 1 1 1.60 1 1 1.19 30 ppm CH4 1.11 1.11 1.18 1 1 1 1 1 1 1 1 1 1.26 1 1 1 1 1 1 1 1 1 1 1 20 ppm C2H4 1 1.19 1.16 1.12 1.13 1.19 1 1 1 1 1 1 1.10 1.26 1.33 1 1 1 2.27 1.03 1.02 1.15 1.17 1.05

WO3/Pt

WO3/Au

WO3/Ag

WO3/Ti

WO3/SnO2

WO3/ZnO

WO3/ITO

Fig. 4. Measured SEAD with detected circles.

For reducing gases (H2S, CH4 and C2H4), the sensor response was defined as R 0/R g and for oxidizing (SO2) as R g/R 0. In this way, the sensor response is always higher or equal to unity. For each type of sensor, the higher responses to gases appear in bold.

306

M. Stankova et al. / Thin Solid Films 500 (2006) 302 308

Fig. 6. Some typical responses to the gases studied. (A) Response of an Audoped WO3 sensor operated at 200 -C to 100 ppb of H2S; (B) response of a ZnO-doped sensor operated at 110 -C to 20 ppm of C2H4.

Table 4 shows the values of sensor response, which was defined as R 0/R g (for H2S, CH4 and C2H4) and R g/R 0 (for SO2), where R 0 is the baseline resistance in pure CO2 and R g is the sensor resistance in the presence of the polluted carbon dioxide. It has been shown [15] that when WO3 is exposed to SO2, an oxidizing gas for this temperature range, the electrical resistance of the sensor increases. This is why the sensor response was defined differently for this gas. The ratio between the standard deviation of the response and its mean (over 10 replicate measurements) was below 1.59 10 2 for each gas and operating temperature. This shows the good repeatability that was observed. These results demonstrate that the sensors studied make possible the detection of sulfur compounds in a CO2 stream and identify them even in the presence of hydrocarbons at higher concentrations. The sensors were insensitive to methane. Tin oxide-based sensors should be used to detect this gas [9]. In the detection of ethylene, sensor responses were very low, exception made of the ZnO-doped layers operated at 110 -C (R 0/ R g = 2.27). These sensors could be used for sensing C2H4, although, again, it is advisable to use tin oxide sensors. All the sensors were sensitive to H2S and SO2. However, important differences in responsiveness were found among them. In the case of hydrogen sulfide, the responsiveness was the highest for sensors doped with Au. Sensor resistance changed up to 13 times when these devices were operated at 260 -C. For Au-doped tungsten oxide, sensor response increased noticeably when the operating temperature was increased. This highest increase in responsiveness (3 times) is obtained when the temperature is increased from 110 -C to 200 -C. Responsiveness was also very good for sensors doped with Ag (see Table 4). In this case, the maximum value was about 7, and this is roughly two times less than for WO3/Au films. Again, sensor response increased with working temperature. Pure tungsten oxide layers also presented a very good response. It was three times less than that for WO3/Au and two times less than for WO3/Ag, which shows the positive role of doping materials. Just as before, pure tungsten oxide sensors increased their response when their operating temperature was increased. The responsiveness of sensors doped with Pt, Ti and SnO2 can

be considered as moderate. Only ZnO-doped sensors showed a different behavior because their responsiveness decreased when their working temperature was increased. The response of ZnO-doped sensors was moderate at 110 -C and very low at the other temperatures studied. Finally, the sensors doped with ITO were almost insensitive to H2S. At the lowest temperature, they did not respond at all and at higher temperatures their response was very low (1.05 1.07). For the detection of sulfur dioxide, the best results were obtained for sensors doped with Pt (see Table 4). The maximum responsiveness (7.71) was at the intermediate temperature (200 -C). Only these sensors showed a high response to this pollutant. The response of the others can be considered as moderate. If sensors are sorted according to their responsiveness, the result is as follows: (1) Titanium-doped films, which can be used at low temperature, because their response at 110 -C was 1.77 (a little higher than the one for WO3/Pt at the same operating temperature). The response decreased slightly with operating temperature. (2) ZnO-doped layers, whose responsiveness was 1.60 at 260 -C, but showed no response at lower temperatures. (3) Au-doped sensors showed a maximal responsiveness of 1.50 at T = 200 -C. (4) Pure tungsten oxide sensors showed the maximal responsiveness at 200 -C. However, the response at lower and higher operating temperatures was close to the maximum values. (5) The WO3/SnO2 devices show a responsiveness that decreased with operating temperature. The maximum was 1.29 at 110 -C. (6) ITO-doped films showed responsiveness when operated at 260 -C only. (7) Ag-doped sensors showed very low response to SO2. Shimizu et al. [15] had reported that silver improved the sensitivity to SO2. A possible explanation for this different result is as follows: 1. The concentration of the doping material is different. 2. While the paper by Shimidzu reported the detection of SO2 in air, here the balance gas was CO2. 3. Finally, the sensor operating temperatures were different (from 110 to 260 -C here, and from 300 to 600 -C in Shimizu et al. [15]). 3.3. Selectivity As it has been shown, the sensors can detect sulfur compounds in the presence of hydrocarbons. They can also selectively detect H2S and SO2 (a reducing and an oxidizing gas, respectively). The dependence of sensor response on the operating temperature is shown in Fig. 7. By comparing Figs. 7(A) and (B), it can be derived that the combination of Pt doped sensors with some of the WO3/Au, WO3/Ag or pure WO3 leads to the selective detection of the two sulfur compounds considered. For example, if an Au-doped film was chosen as active layer to detect H2S, there will be no cross sensitivity to SO2, because the response to the latter is very low. The situation is similar for Ag-doped films, which can also be applied to H2S sensing. Pt-doped layers are the best for detecting SO2. What is more, their response to H2S is low, so the selectivity is quite high. The best working temperature for this case is 200 -C (see Fig. 7(B)), although at 260 -C the response was also high. All this leads to the conclusion that if an array of sensors with different doping materials were

M. Stankova et al. / Thin Solid Films 500 (2006) 302 308

307

some degree of collinearity. This collinearity means that R will have some dominating types of variability that carry most of the available information. The objective of PCA is to express the information in the variables of R = {r k , k = 1, 2, . . ., K } K is the number of columns in the response matrix by a lower number of variables P = { p 1, p 2, . . ., p n } (n < K ) called principal components. The PCs are chosen to contain the maximum data variance and to be orthogonal. The response matrix is decomposed into a product of two matrices (scores and loadings). While the loadings matrix, L , contains the contribution of the original response vectors to the new response vectors or PCs, the scores matrix contains the response vectors projected onto the space defined by the PCs. Further details on this method can be found elsewhere [16]. Prior to perform the principal component analysis, the response matrix was autoscaled. The PCA was performed using the responses of all sensors measured. The response matrix was formed from the conductance change and had 24 columns (8 sensors with different active layers 3 operating temperatures) and 80 rows, which corresponded to the number of measurements (4 gases 10 replicates per measurement 2 sensors of each type). Scores and loadings biplots provide a useful tool of data analysis and allow the visual appraisal of the structure of large data matrices. It is specially revealing in PCA, where the biplot can show inter-unit distances and indicate clustering of units as well as display variances and correlation of the variables [17]. The relationship between the objects (i.e. measurements) and the variables (i.e. sensors) are often best displayed on a biplot. This is a plot with axes scaled as to include the scores-scores coordinates as well as the loading values. A vector drawn from the origin to a set of loadings coordinates represents the size of the loadings on the respective PCs, and, in addition, indicates the objects with which it is particularly associated [18]. If we draw the graph of sensor loadings and the biplot between this

Fig. 7. Dependence of the sensor response on the operating temperature for the doping materials studied: (A) for 100 ppb H2S; (B) for 1 ppm SO2.

fabricated, it would be possible to selectively detect sulfur compounds in a CO2 stream. What is more, if a combination of doped WO3 and SnO2 sensors were considered, then it would be possible to selectively detect not only H2S and SO2 but also traces of hydrocarbons such as ethylene and methane in CO2. 3.4. Gas analysis by principal component analysis (PCA) The gases studied were qualitatively analyzed using PCA. PCA is a linear, unsupervised pattern recognition method. PCA involves a mathematical procedure that transforms a number of correlated variables into a (smaller) number of uncorrelated variables called principal components. The first two principal components (PCs) are often employed, since they capture the most data variance, and therefore theoretically best describe the data. The most frequent application of PCA is in cases where the sensor response matrix, R , is expected to contain variables with

Fig. 8. Results from the PCA analysis performed on sensor response (biplot of the scores and loadings). The scores are the projection of measurements in the orthogonal base of PCs and the loadings are the contribution of each sensor to the PCs.

308

M. Stankova et al. / Thin Solid Films 500 (2006) 302 308

graph and the scores plot (Fig. 8), a combination of sensors with different active layers can be proposed to detect sulfur compounds. The different gases cluster together in the scores plot and there is a very good separation between them. Since the sensors were almost insensitive to methane (see Fig. 8), this component appears nearly at the (0,0) point of the plot. For ethylene, the results show a higher dispersion because only some of the sensors responded to this species and their responses were poor. The loadings present the distribution of the different sensors in relation to the different operating temperature. That is why some dispersion for sensors of the same type can be observed. To detect H2S the best solution is to use sensors doped with Ag and Au. The devices doped with SnO2 are selective to SO2. Sensors doped with ITO and the pure tungsten oxide ones are selective to ethylene. 4. Conclusions Microhotplate sensor arrays based on pure and doped tungsten oxide thin films were fabricated. A morphological study of the Pt doped layer was made using TEM. The thickness of the doping layer was found to be about 3 4 nm. The dopant material was distributed on the surface and no diffusion into the WO3 was detected. By SAED and XRD, the phase of the compounds was determined: WO3 was present in monoclinic phase and the Pt in cubic phase. The ability of the sensors based on pure and doped WO3 to detect the presence of sulfur compounds as pollutants in CO2 has been investigated. The sensors showed high and reversible responses to the presence of H2S and SO2 diluted in CO2, in absence of oxygen (O2 concentration below 15 ppm). The sensors doped with Au and Ag responded very well to H2S (100 ppb in CO2) and poorly to SO2 (1 ppm in CO2), while the sensors doped with Pt showed an opposite behavior. Besides, H2S behaves as a reducing gas (decreasing sensor resistance), while SO2 behaves as an oxidizing species (increasing sensor resistance). The responsiveness for H2S and SO2 were highest when the sensors were operated at 260 -C and 200 -C, respectively. Taking into account these results, an array combining active layers of WO3/Au and WO3/Pt is a promising system for detecting the pollution of CO2 by sulfur compounds. As it was confirmed by PCA, the detection of sulfur compounds was not influenced by the presence of hydro-

carbons such as methane and ethylene. However, the interference from other gases such as ammonia or nitrogen oxides (with maximum allowed concentration in CO2 of 2 and 2.5 ppm, respectively [1]), has to be analyzed prior to ensure that these materials are the best for this application. Acknowledgements This work has been funded by the Spanish Commission for Science and Technology (CICYT) under grant no. TIC200306301. References
[1] ISBT carbon dioxide quality guidelines and analytical procedure bibliography, International Society of Beverage Technologists, USA, March 2001. [2] Carbon dioxide source certification, quality standards and verification, IGC Doc 70/997E, European Industrial Gases Association, Brussels, 1999 (http://www.eiga.org/pdf/Doc%2070%2099%20E.pdf). [3] S.R. Morrison, in: S.M. Sze (Ed.), Semiconductor Sensors, John Wiley and Sons, Inc., USA, 1994, p. 383. [4] N. Ba rsan, U. Weimar, J. Phys., Condens. Matter 15 (2003) R813. [5] D. Kohl, Sens. Actuators, B 18 (1989) 71. [6] W. Schmid, N. Ba rsan, U. Weimar, Sens. Actuators, B 103 (2004) 362. [7] L.F. Reyes, S. Saukko, A. Hoel, V. Lantto, C.G. Granqvist, J. Eur. Ceram. Soc. 24 (2004) 1415. [8] X. Vilanova, X. Correig, E. Llobet, J. Brezmes, R. Calavia, X. Sa nchez, Spanish Patent No. ES2212739, 2003. [9] X. Vilanova, E. Llobet, J. Brezmes, R. Calavia, X. Sa nchez, X. Correig, in: Proceedings of IEEE Transducers 2003, Boston, U.S.A., June 8 12, vol. 2, 2003, p. 1347. [10] M. Stankova, X. Vilanova, J. Calderer, E. Llobet, P. Ivanov, I. Gra ` cia, C. Cane , X. Correig., Sens. Actuators, B 102 (2004) 219. [11] International Centre for Diffraction Data website (http://www.icdd.com/). [12] E. Salje, K. Viswanathan, Acta Crystallogr., A 31 (1975) 356. [13] P.M. Woodward, A.W. Sleight, T. Vogt, J. Phys. Chem. Solids 56 (1995) 1305. [14] J.L. La ba r, Proceedings of EUREM 12, Brno, Czech Republic, July 9 14, 2000, p. I379. [15] Y. Shimizu, N. Matsunaga, T. Hyoto, M. Egashira, Sens. Actuators, B 77 (2001) 35. [16] I.T. Joliffe, Principal Component Analysis, Springer Verlag, New York, 1986. [17] K.R. Gabriel, Biometrica 58 (1971) 453. [18] S. Kokot, M. Grigg, H. Panayiotou, T.D. Phuong, Electroanalysis 10 (1998) 1081.