Detection of chemical warfare agents using ... - Semantic Scholar

Sensors and Actuators B 108 (2005) 41–55

Detection of chemical warfare agents using nanostructured metal oxide sensors Alexey A. Tomchenko∗ , Gregory P. Harmer, Brent T. Marquis Sensor Research and Development Corp., Orono, ME 04473, USA Received 12 July 2004; received in revised form 12 November 2004; accepted 18 November 2004

Abstract The feasibility of thick-film chemical sensors based on various semiconductor metal oxides to reliably detect chemical warfare agents has been studied. Nanocrystalline semiconductor metal oxide (SMO) powders were used as initial materials for the sensors’ fabrication. The thick films were prepared using a simple drop-coating technique accompanied with in situ annealing of the deposited films by a heater that is integrated into the sensor’s platform. The sensors were exposed to mixtures of hexane, diesel oil vapor, methanol, 1,5-dichloropentane (DCP), or dimethyl methylphosphonate (DMMP) with air. DCP and DMMP were considered as simulants of mustard gas and nerve agents respectively. The performance of the sensors was investigated over a wide range of operating temperatures. They were additionally tested with mustard gas, sarin and soman at a certified live agent facility. The data obtained from the simulant and live agent testing are presented and discussed. In particular, attention is focused on the ability of an array of sensors to detect and identify agents in mixtures of interferents. © 2004 Elsevier B.V. All rights reserved. Keywords: Chemical warfare; Metal oxide gas sensor; Thick film; Sensor array; Principal component analysis; Linear discriminant analysis

1. Introduction The use of chemical weapons against civilians by terrorist groups or fanatic individuals is not just horror fiction anymore, but an absolute real threat. Two sarin gas attacks in Matsumoto and Tokyo, Japan in 1994–1995 confirmed this horrible reality. This demonstrates that there is a critical need for detectors and sensors that are able to warn about imminent CWA danger, to enable people to safely leave a contaminated zone or to protect themselves. Accordingly, highly selective sensitive sensors to CWA have to be the priAbbreviations: SMO, semiconductor metal oxide; DCP, 1,5dichloropentane; DMMP, dimethyl methylphosphonate; CW, chemical warfare; CWA, chemical warfare agent; IDLH, immediately dangerous for life and health concentration in air; IMS, ion mobility spectroscopy; SAW, surface acoustic wave; P-SAW, polymer coated SAW sensors; TIC, toxic industrial chemicals; PCA, principal component analysis; LDA, linear discriminant analysis ∗ Corresponding author. Tel.: +1 207 866 0100; fax: +1 207 866 2055. E-mail address: [email protected] (A.A. Tomchenko). 0925-4005/$ – see front matter © 2004 Elsevier B.V. All rights reserved. doi:10.1016/j.snb.2004.11.059

mary focus, i.e. they have to respond to low concentrations of agents—considerably lower than immediately dangerous to life or health concentrations (IDLH). They also have to discriminate CW agents from the other chemicals in the environment, and they have to identify CW agents on the other chemicals’ background. Modern analytical chemistry has instruments that can detect single molecules [1]. However, in general, these are delicate, stationary, and expensive equipment demanding complex and time-consuming sample preparation. At present, analyses of this kind are possible exclusively within the precincts of sophisticated research laboratories. “Fieldable” detectors of CWA and toxic industrial chemicals (TIC) have to be robust, portable, fast-acting, cheap, simple to operate, and, as discussed above, they have to be very sensitive and selective to the detected gases. There currently exist a wide variety of techniques to detect CW-agents [2,3]. Each, however, has drawbacks and limitations. Infrared spectrometers offer a limited level of standoff detection; they are expensive, complex and bulky. Raman spectroscopy that has success-

42

A.A. Tomchenko et al. / Sensors and Actuators B 108 (2005) 41–55

fully been employed for CWA detection in the laboratory conditions is not applicable for identification of the agents in the real-world environment due to the lack of robustness. Colorimetric indicators, commonly referred to as detector paper, is the fastest, cheapest, lightest, and easiest type of detector to use. However, they detect chemical agents in the liquid form only; they also suffer the high cross-sensitivity to smoke, acetone, gasoline and other interferents. Colorimetric tubes are applicable to both vapors and gases and provide a semi-quantitative indication of the amount of agent presented in the environment. The drawbacks of this technique are the low speed of the responses to agents and the size of instrumentation (a stable air pump is necessary for the operation). Ion mobility spectrometers can detect CWA in real-time, are portable and sensitive. But, this technique requires a radioactive source, has poor selectivity in the presence of interferents, is costly, and the instruments are too large for some applications. Mass spectrometers combined with gas chromatographs (GC–MS) are the most sensitive and most reliable of today’s instruments for CWA detection. However, they are bulky, expensive, require sample preparation and need technically trained personnel. Similar limitations are inherent in flame photometers. Polymer coated SAW devices (P-SAW) perform the real-time detection; they are selective and relatively inexpensive. But, previous efforts have reported concerns about their sensitivity and robustness, poisoning with hydrofluoric acid has been observed. Arrays of chemically sensitive micro resistors produced from semiconductor metal oxides (SMO) are considered as one of the most promising basic technologies for CWAdetection [4]. SMO sensors offer a wide variety of advantages over other analytical instruments such as low cost, short response time, easy manufacturing, and small size. SMO chemoresistors also can be successfully used in combination with other types of CWA detectors (with IMS or P-SAW detectors for example), and as demonstrated lately, the combinatorial approach makes it possible to produce more sensitive and reliable CWA-detection equipment [5,6]. A common concern about the SMO sensor technology is the lack of selectivity. The current state-of-the-art has identified four general approaches to improve this parameter. They are: (a) use of catalysts and promoters, (b) optimization of sensors’ operating temperature, (c) use of surface additives promoting the specific adsorption, and (d) use of molecular filters [7,8]. The implementation of SMO sensor arrays combined with appropriate pattern recognition tools is also considered as a promising approach to compensate for this lack of selectivity and to provide coverage for multiple types of agents [9]. The authors of this paper have recently demonstrated the feasibility of a SMO sensor array to discriminate and recognize various constituents of a combustion gas [10]. Principal component analysis along with several classification methods was successfully used to identify nitrogen oxides, ammonia, sulfur dioxide, and hydrogen sulfide. It has also been demonstrated the viability of SMO thick-film gas sensors prepared using cheap commercial sensor platforms

and a very simple drop-coating technique accompanied with in situ annealing of the deposited films by integrated heaters. We continue the investigation of the SMO sensors prepared using the drop-coating method. In this paper we present comprehensive data for the porous SMO thick films considered as potential sensitive elements for the detection of CWA. Various thick-film compositions have been studied at different operating temperatures in gas flows containing CWA simulants, interferents, or chemical warfare agents. The tests using the CW agents, which were sarin, soman, or mustard gas, were carried out at a certified live agent facility at the Southern Research Institute (Birmingham, AL, USA). We present and discuss the performance of the SMO sensors in respect of sensitivity and selectivity towards CWA. A subset of the SMO sensors was selected for inclusion in the sensor array for chemical warfare detection. A primary objective of this study is to analyze the ability of the array consisting of five selected metal oxide sensors to identify the gases under tests by means of pattern recognition techniques.

2. Experimental 2.1. Fabrication of thick-film sensors Porous metal oxide thick films approximately 50 ␮m thick were fabricated using a drop-coating technique and an in situ annealing method. The films were deposited onto commercial UST sensor platforms (UST Umweltsensortechnik GmbH) which are 3 mm × 3 mm alumina substrates suspended by platinum leads in a TO-8 housing [10,11]. The substrates were equipped with integrated platinum heater and two platinum electrodes. To form the film, a drop of metal oxide paste was applied onto electrodes. Metal oxide pastes were prepared by mixing nanocrystalline oxide powder precursors with glass frit and an organic binder. The SMO powders (Table 1) were acquired from the NanoProducts Corporation (Longmont, CO, USA) [12] and the Nanophase Technologies Corporation (Romeoville, IL, USA) [13]. After the thick-film deposition the samples (sensors) were put into a test gas chamber and in situ annealed using the integrated heaters in airflow of 200 cm3 min−1 . A trapezoid temperature profile with heating and cooling ramps of 5 ◦ C min−1 and a 10 min exposure at a peak temperature was used for annealing. Prior to the start of each test the sensors were preheated for 60 min at the test temperature to allow the SMO films to thermally stabilize. 2.2. Gas delivery system and electrical sensing testing The experimental setup used for testing the thick-film sensors is shown in Fig. 1. It consists of the SMO test chamber in parallel with the SAW test chamber containing reference DMMP-selective SAW sensors, SRD gas delivery system, mass flow controller units, 10-channel heater unit and 10channel resistance measurement unit for the SMO sensors,


43

Table 1 Characteristics of the SMO powders Oxide

X-ray phase

Specific surface area (m2 /g)

Equivalent spherical diameter (nm)

Density (g/cm2 )

Manufacturer

SnO2 WO3 In2 O3 CuO Y2 O3

Tetragonal Orthorhombic, monoclinic Cubic Monoclinic Cubic

30 18.5 40 16.1 14.7

29.4 45.3 20.9 58.1 81.5

6.8 7.16 7.18 6.4 5.01

Nanophase Technologies Corporation NanoProducts Corporation

a frequency counter for the SAW sensors, and a computer interfaced to relevant equipments. The SMO sensors were operated at 200–650 ◦ C using the integrated heater with an applied dc voltage controlled by feedback circuitry [14]. Custom LabVIEW-based software was used to autonomously control the experimental setup and take measurement of the sensors. The sensors’ responses were displayed in real-time and saved to disk for off-line processing and analysis. 2.3. Simulants, interferents and agents The gas exposure protocol included hexane, diesel vapour, methanol, 1,5-dichloropentane (DCP), and dimethyl methylphosphonate (DMMP) (Fig. 2). DCP and DMMP are commonly considered as simulants of mustard gas (HD) and nerve agents (GD, GB, VX), respectively. Bubblers with the naturally evaporated liquids were used as the gases’ supplies: a constant airflow about 5 cm3 min−1 through the bubblers was maintained in order to keep the gases’ basic con-

centration stable for the whole test cycle. During the exposure the flow containing the gas of interest was switched to the system’s manifold and additionally diluted with air. Two concentrations of each simulant were delivered to the sensors for the experiments. The DCP concentrations were approximately 3 ppm (DCP L) and 10 ppm (DCP H) in air. DMMP was delivered at approximately 0.5 ppm (DMMP L) and 5 ppm (DMMP H). In order to reduce risk of the tubing contamination with the above gases, the platform of the gas delivery system was heated to 50 ◦ C permanently. The in-house DMMP-selective SAW sensors [15] in the parallel test-chamber (Fig. 1) were used as references to control stability of DMMP delivery and to monitor the system’s DMMPcontamination. Diesel oil vapour, methanol, and hexane are most probable and active interferents in the real environment. We exposed the sensors to about 80 ppm of hexane, to 80 ppm of methanol, and to approximately 25 ppm of diesel vapour. To deliver diesel vapour to the chambers, the same principle

Fig. 1. Experimental test station.

44


Fig. 2. Gas exposure protocol used for the tests with the simulants and interferents.

had been used as for DCP and DMMP delivery. Airflow of 15 cm3 min−1 was permanently pumped through the bubbler with diesel oil. When it necessary, the diesel containing flow was switched to manifold of the gas system and after an additional dilution with air was delivered to the test chambers. The gas exposure protocol consisted of two parts. First, we exposed the sensors to each of the aforementioned gases sequentially, and then we tested the sensors’ DCP and DMMP sensitivity on the background of methanol. Flows containing target gases and the purge were alternately switched to the test chamber with a fixed flow rate of 100 cm3 min−1 . Each gas exposure was 3 min long, followed by 12 min of purge. The gas delivery system was verified on a regular basis via GC–MS. DCP with 99% purity and DMMP with 97% purity were obtained from Aldrich (Table 2) and used as received. The methanol–air and hexane–air mixtures were acquired from BOC Gases (The BOC Group Inc., Murray Hill, NJ, USA). The ordinary diesel fuel was acquired from a local gas (petrol) station. The experiments with CW agents (sarin, soman, and mustard gas) were carried out at the certified live agent facility using the same type of experimental setup as for simulant testing (see Section 2.2). The gases were delivered into the

test chambers as mixtures with dry air in concentrations of 1, 10, 50, 100, 500 or 1000 ppb. 2.4. Signal processing To characterize the performance of the SMOs we deal entirely with the transient response in this paper, the type shown in Fig. 3 for example. The transient response consists of two parts: the hit, which is the initial deviation from the baseline due to the interaction of the sensing film with the gas, and the recovery, where the response recovers back to its initial baseline once the gas exposure has ended. Due to the exponential behaviour of the resistance of SMOs we plot the logarithm (base 10) of the response on a linear axes (bottom plots of Fig. 11 for example), which is the equivalent to plotting the raw data on a semi-log plot. The individual hits are compared by normalizing directly before the start of the hit, that is simply zeroing the baseline of the log resistance (see Fig. 3 for example). This is equivalent to dividing by the baseline and plotting on a semi-log plot. The normalization step means the responses are dimensionless, but we label the units as log (Ω/Ω) to avoid confusion as to what exactly has been plotted. Thus, a change from 0 to −1 is one decade

Table 2 Characteristics of the CWA simulants Simulant

Chemical formula

CAS number

Density (g/ml)

Molecular weight

Boiling point

DCP

Cl CH2 CH2 CH2 CH2 CH2 Cl

628-76-2

1.106

141.04

63–66 ◦ C/10 mm Hg

756-79-6

1.145

124.08

182 ◦ C/760 mm Hg

DMMP


45

Fig. 3. Normalized response of SMO sensors to DCP (10 ppm) and methanol (80 ppm) vs. sensors’ operating temperature.

in terms of a change in real resistance, like 100–10 k for example. The magnitude of the transient response (the peak deviation from the baseline) varies with the concentration of the delivered gas, everything else being equal. To eliminate concentration dependence from the response shapes, they are further normalized by fixing the magnitude to one. This second normalization relies heavily on the assumption that response shape is linearly related to the concentration, which may hold true for a limited range of concentrations, but certainly deteriorates for low concentrations where the shape loses its ‘crispness’. To use this response as a feature for a classification scheme the number of dimensions needs to be reduced. The dimensionality is essentially the number of data points of the feature, i.e. how many axes would be required to plot the feature as a single point. The response is initially down sampled by interpolating Nh points during the hit and Nr points during the first x minutes of recovery. Though this reduces the di-

mensionality by a great deal, the data is still highly correlated between dimensions. To reduce the dimensionality further the popular technique of principle component analysis (PCA) or linear discriminant analysis (LDA) can be used. The PCA algorithm transforms the data so that correlations between the variables are minimized. It does this by iteratively finding an orthogonal axis (to all the previously found axes) that has the maximum variance of the projected data. The net result is a list of axes, called the principle components, which are ranked in decreasing order of variance. Hence, the first principle component contains most of the information about the dataset. The PCA algorithm itself is lossless, but by keeping only the top d-axes (which is usually d = 3 for visualisation purposes) a high percentage of the information about the data is retained whilst dramatically reducing the dimensionality. However, though the PCA faithfully characterises the full dataset, it may not be optimal for feature extraction, especially when dealing with correlated data. For one reason, it

46


does not use any class information and thus is a blind algorithm in terms of classification. In terms of information, PCA does not utilize all available information, which hints that a better performing algorithm may be possible by using this information. For feature extraction we want to maximize the distance between classes while minimizing the variance of each class of the projected data—this is the LDA objective from which the algorithm is based. Similarly to PCA, the LDA provides a ranked list of orthogonal axes (or equivalently a transform matrix), where we can retain the top d dimensions. The LDA provides a more efficient method of dimension reduction than PCA, which leads to improved selectivity. It is also more immune to noise, in particular low signal-to-noise responses that are heavily amplified in the second normalization step. We have used LDA in preference to PCA unless otherwise noted. When considering only a single sensor, the output of the LDA is considered the feature for that sensor. Visually viewing these features (with d = 3, as shown in Fig. 8 for example) gives a good indication of the expected classification performance. Though classification was performed, the results do not add much extra relevant information than the LDA plots. To determine the performance of a sensor array the features of the individual sensors are simply concatenated together, thus the dimension of the feature for a sensor array will be nd, where n is the number of individual sensor types in the array. To visualize the feature space of the sensor array another LDA is taken, again with d = 3. This is purely for visualization; any classifications performed on the sensor array used all the nd dimensions.

3. Results and discussion 3.1. Gas sensitivity of the individual sensors First we scrutinized the thick-film sensors based on the pure nanocrystalline metal oxides, namely: SnO2 , WO3 , In2 O3 and CuO. The simulants and interferents were tested at different operating temperatures from 150 to 400 ◦ C. The sensors’ sensitivity and reversibility improved significantly with increasing temperature (Fig. 3). In Fig. 3 et seq. the resistances are normalized using log R/R0 , where R0 is a resistance in air immediately before the start of a gas hit, and R = R(t) is a variable resistance after the start of the hit. We used this representation because the one gives an easy estimate of a sensor’s sensitivity and greatly simplifies the comparison procedure between various responses generated by different sensors towards different analytes. Fig. 3 summarises data for selected gases and SMO types at operating temperatures 200, 300, and 400 ◦ C. Nevertheless, the plots express the overall tendency stated above, i.e. the same type behaviour was observed for all of the sensors and for all of the gases tested. The temperature 400 ◦ C was chosen as the basic operating temperature. The sensors were quite stable. Slight baseline resistance variations were observed during the long-term – up

to 64 h – experiments. However, the responses were reproducible in their magnitudes and shapes. Fig. 4 demonstrates the response reproducibility for WO3 -based sensors towards diesel vapour and DMMP and for SnO2 -based sensors towards DCP and DMMP for sequential trials. The responses for WO3 are almost identical. Similar stability is also shown by CuO sensors. However, the SnO2 and In2 O3 -based sensors demonstrated a decrease in sensitivity and a slight distortion in response shape. It is evident from Fig. 4 (bottom plots) that the amount of change in shapes between the trials decreases with time; the responses approach asymptotic shapes. Sensors based on WO3 , In2 O3 and CuO were tested with live CW agents. Fig. 5 shows the responses of In2 O3 to sarin (GB) and mustard gas (HD). As can be seen, the sensors produced reliable responses to 10 ppb both of the gases, or in other words, they can potentially be used as reliable detectors for the agents starting 10 ppb concentration. Moreover, the real CWA tests demonstrated that the SMO sensors were more active to the agents than to the simulants though the response shapes are comparable. For example, characteristic poisoning of the sensors was observed for GB exposures that correlated with the data obtained for DMMP and thus additionally verified the validity of DMMP as a nerve agent simulant. An interesting trend was observed in SMO sensitivity to DMMP and nerve agents: the sensitivity increases from DMMP to sarin and even more to soman, i.e. with increasing number of methyl groups in the molecules (see Fig. 6 with the gases’ molecular structures shown as insets in the corresponding plots). It is probable that the methyl groups, which are eliminated during an agent’s dissociative adsorption and then oxidized, play the crucial role in the SMO electrical response generation. We can then predict with some reliability that if a sensor responds to low concentration of DMMP then it is most likely to respond to live agents with a higher sensitivity. As an illustration, the good WO3 -sensors’ sensitivity to 5 ppm of DMMP become transformed to the excellent sensitivity of the sensors to 1 ppm of sarin and soman (extrapolated from Fig. 6). The above experiments demonstrate that the pure SMO sensors are sensitive and stable. We now consider the selectivity with respect to the simulants and interferents. The responses to methanol, DCP and DMMP for four different types of the sensors based on “traditional” SMO materials are presented in Fig. 7. The SMO sensors based on various oxides react to the gases of interest in similar manner. There were slow responses and slow recoveries towards DMMP—the sensors did not completely recover after 12 min post-exposure air purge. The responses towards DCP were also slow, however they were completely reversible. Also all of the DCP-response curves had the specific low-speed part in the beginning of the exposures. The methanol responses were much faster as compared to DCP, no pronounced inflection points were observed. Numerous experiments have verified that the presented responses are typical for the aforementioned materials. Using


47

Fig. 4. Stability of the sensors: normalized responses of WO3 and SnO2 sensor towards diesel vapour, DCP and DMMP (at 400 ◦ C).

any of the above sensors and analyzing simply the response curves, we can easily discriminate between the three gases of interest as shown in Fig. 8. Also note for the features shown in Fig. 8 that Nr = 0, that is the recovery portion of the response is ignored. This is a far more practical implementation, as we do not require the gas exposure to end at some specified time. From this result we can make some general comments. • all of the thick-film sensors based on the “traditional” nanocrystalline SMO sensor materials and operated in the temperature range of 200–400 ◦ C – where their sensitivity is satisfactory to produce gas-related electrical signals of a significant magnitude – have the similar mechanism of their electrical response to a specific gas; • the responses to any gas of interest have the specificity that is conditioned by molecular properties of the gas and can be used for the identification of the gas; • the difference in the response magnitudes between the above thick-film sensors prepared on different metal ox-

ides is a result of every sensor type’s uniqueness conditioned by physical–chemical properties of the respective basic oxide. Thus, every gas of interest produces its specific type of response. To comprehend this specificity, let’s consider the SMOs’ response mechanisms to the above gases in detail. Methanol is a very active reducing agent. Taking into account the well-known fact that semiconductor surface is populated with chemisorbed atomic oxygen the process of generation of SMO electrical response to methanol could approximately be described by the following reactions. • Response: CH3 OH(gas) ↔ CH3 OH(ads)

(1)

CH3 OH(ads) + 2O(ads) − → CO2 ↑ + H2 O ↑ + 2e−

(2)

48


Fig. 5. Normalized responses of SMO sensors towards sarin and sulphur mustard (at 400 ◦ C).

• Recovery:

• Process 1 o Response:

O2(gas) ↔ O2(ads) −

O2(ads) + 2e ↔ 2O(ads)

(3) −

(4)

where e− is a free electron in the conduction band and O(ads) − is a negatively charged oxygen atom, a complex of an adsorbed oxygen atom with a localized electron from the conduction band. Methanol molecules are comparatively small, volatile and easily penetrate into the porous thick film. The reactions (Eqs. (1)–(4)) occur quite intensively at the above conditions. As a result, the smooth and fast electrical responses are observed experimentally with resistance decrease for ntype SnO2 , WO3 , In2 O3 and the resistance increase for p-type CuO (Fig. 7). The action of DCP (Cl CH2 CH2 CH2 CH2 CH2 CH2 Cl) is more complex. At least two symmetrical processes could take place on the semiconductor surface after the molecules’ adsorption. The processes could be described as follows.

Cl2 (CH2 )5(ads) +2O(ads) − → HCl ↑ + CO2 ↑ + H2 O ↑ + 2e− ,

(5)

o Recovery: O2(gas) ↔ O2(ads)

(6)

O2(ads) +2e− ↔ 2O(ads) −

(7)

• Process 2 o Response: Cl2 (CH2 )5(ads) ↔ Cl(ads) + Cl(CH2 )5(ads)

(8)

Cl(ads) + e− ↔ Cl(ads) −

(9)

o Recovery: Cl(ads) − + Cl(ads) − → Cl2 ↑ + 2e− ,

(10)

where Cl(ads) − is a complex of an adsorbed chlorine atom with a localized electron from the conduction band.


49

Fig. 6. WO3 sensors’ normalized responses towards DMMP, sarin and soman. The sensors operated at 400 ◦ C.

The reactions (Eqs. (5)–(7)) dominate for the above “traditional” sensor semiconductors operating at 200–650 ◦ C. There are plenty of free carriers in these materials at these conditions. This predetermines high concentrations of active oxygen on the semiconductors’ surface and, correspondingly, the DCP-semiconductor reaction along the first of the above routes. During the DCP exposure the atomic oxygen reacts with the gas actively. As a result, some of the electrons are delocalized and return to the conduction band. We observe this as a more or less sudden decrease in the sensitive elements’ resistance for n-type semiconductors and the resistance increase for the p-type sensors. The initial “delay” of about 30 s, when the speed of the DCP-responses is lower, we attribute to the slowness of the large DCP molecules’ diffusion into the films. The reactions (Eqs. (8)–(10)) dominate for the “very wide band gap semiconductors” with Eg of 4–5 eV – like Eu2 O3 or Sm2 O3 – operating at elevated temperatures (600 ◦ C and above). This route will be discussed in detail later.

Adsorbed DMMP – (CH3 O)2 P(O)CH3 – also actively interacts with O(ads) − . The process could roughly be described as follows. • Response: (CH3 O)2 P(O)CH3(ads) +2O(ads) − → H3 PO4 ↑ + CO2 ↑ + H2 O ↑ + 2e− ,

(11)

• Recovery: O2(gas) ↔ O2(ads)

(12)

O2(ads) + 2e− ↔ 2O(ads) − ,

(13)

The reaction (Eq. (11)) has been assumed based on the data presented in [16,17]. Eq. (11) does not show intermediate stages of the DMMP-SMO surface reaction. In reality, the process is quite complex and occurs with formation of various intermediate products, such as methyl methylphosphonate, methylphosphonate, methanol, etc. [18–20]. It was demonstrated in [18] that after DMMP dealkylation the

50


Fig. 7. Normalized responses of SMO sensors operating at 400 ◦ C towards DCP (10 ppm), DMMP (5 ppm), and methanol (80 ppm).

stable adspecies – methyl methylphosphonate (MMP) and methylphosphonate (MP) – populated the semiconductor surface even at 400 ◦ C. Taking this fact into account, the DMMPresponse curves (in Fig. 8) could be interpreted as follows. The slowness of the SMOs’ electrical responses towards the gas arises from the comparatively slow evolution of the adsorbed DMMP molecules into adspecies interacting with adsorbed oxygen. Evidently MMP and MP are held at the semiconductor surface without free carriers’ localization (“week chemisorption” [8,21]). The very slow and incomplete recovery (i.e. “poisoning”) could be attributed to the stability of MMP and MP on the surface: they coat the surface and hinder oxygen re-adsorption process (i.e. the reactions Eqs. (12) and (13)). The recovery continues until the self-cleaning of the surface is complete, and apparently the process is time consuming at 400 ◦ C (see Fig. 7). All of the described processes proceed faster with increase in the sensors’ operating temperature (Fig. 9). Thus, the experimentally observed specificity in the SMO responses to any one of the three gases considered above is a reflection of the substantial difference in physical–chemical

processes taking place on the SMO surface during the gases’ adsorption. This specificity could be a basis for the gases’ identification: applying the appropriate pattern recognition techniques [10] to the data obtained from any one of the above thick-film sensors – i.e. using just one SMO sensor – it is possible discriminate between the three tested gases. Fig. 10 illustrates more complex situation when CWA simulants are presented in two concentrations, and the exhausts of two more interferents – hexane and diesel oil vapour – are possible (individual gases only, no gas mixtures are considered). The discrimination between DCP, DMMP and the group of the interferents is good, however identification of the individual interferents becomes quite difficult – though that is not an issue for a CWA detector. The problem of gas identification could be simplified significantly by application of a sensor array based on several SMO materials (will be discussed in detail in Section 3.2). Nevertheless, the development of selective SMO sensors – i.e. the sensors reacting to an individual gas or to a small group of gases – is highly desirable. Further efforts were focused on the search for materials and operating conditions that provide the highest level


Fig. 8. Response curves for SnO2 and WO3 sensors operating at 400 ◦ C and the corresponding LDA.

Fig. 9. Normalized responses of WO3 sensor operated at 400 ◦ C and at 600 ◦ C towards DCP (10 ppm) and DMMP (5 ppm).

51

52


Fig. 10. SMO sensors’ normalized responses to the gases of interest and corresponding LDA.

of selectivity. This search resulted in several promising gassensitive SMO materials presented and discussed below. 3.1.1. Methanol As mentioned above, methanol is the most active reducing agent among the gases of interest. For this reason it is natural to operate the sensors at comparatively low temperatures when their activity towards the other gases is negligible. We found that sensors based on CuO doped with 5 wt.% of Ga2 O3 (Aldrich, Product No 215066) and operated at 200 ◦ C had practically no response to hexane, diesel oil vapour, DCP and DMMP, yet were sensitive to methanol (Fig. 11, left column of plots). The role of Ga2 O3 admixture could be twofold. On the one hand, Ga could act as a donor and depress conductivity in p-type CuO that in turn reduces the material’s sensor activity. On the other hand, the extrinsic Ga2 O3

particles could loosen the CuO film, which makes it more permeable for methanol, and provide better response to this gas. Apparently, the superposition of the two factors imparts excellent methanol selectivity to the composition. 3.1.2. DCP Y2 O3 -based thick films operated at 600–650 ◦ C demonstrated remarkable selectivity to DCP. The sensors were nonresponsive to the interferents and DMMP, but were sensitive to DCP (Fig. 11, right column). We attribute the Y2 O3 sensors’ selectivity to the unique combination of the base materials’ electrical properties and the appropriate choice of operational temperature and film thickness. The compositions mainly consisted of the wide bandgap p-type Y2 O3 . The comparatively high resistance of the films (around 1 G at 600 ◦ C) suggests that very few free carriers exist inside


53

Fig. 11. Selective Y2 O3 and CuO: Ga2 O3 sensors: (a) the sensors’ normalized responses to all gases of interest and (b) the corresponding raw data recorded from four CuO:Ga2 O3 sensors (left plot) and from two Y2 O3 sensors (right plot). The last gas hits indicated as “+ Meths” are three DCP H hits followed by three DMMP H hits, when gas mixture air/methanol (80 ppm) was used as a purge and a carrier gas (see Fig. 2).

the material. Apparently, the concentration of the active oxygen on the surface is considerably reduced at these temperatures and at these carriers’ concentrations, and the process given by Eqs. (8)–(10) becomes dominating. The chlorine that is a product of DCP dissociative adsorption successfully competes with oxygen due to its higher electron affinity. As a result a part of free electrons are localized near the Cl atoms, the balance shifts towards the majority carriers (i.e. holes) and the conductivity increases. Obviously, the DCP reaction with surface oxygen (Eqs. (5)–(7)) also takes place but its contribution to the DCP response is negligible. The other gases react with surface oxygen to generate a very weak response as a slight conductivity decrease. We have to underline that the above interpretation is tentative to some extent, and the phenomena of the Y2 O3 thick-film sensors’ selectivity will be further investigated. We also realize that the sensors are not “absolutely selective” and foresee

that they will react to Cl2 , HCl vapour, and to some strong oxidizing gases, like NO2 or O3 . All of the gases as well as real CW agents will be tested in future experiments. Finally, it should be remarked that similar sensor properties were also noticed for some lanthanoid oxide based compositions operating at 600–700 ◦ C, namely: Sm2 O3 , Eu2 O3 , Gd2 O3 , Dy2 O3 , Ho2 O3 , Er2 O3 , and Yb2 O3 (based on). However, the films based on Y2 O3 had lower baseline resistance, better stability and the best selectivity among the films tested. In summary, the above experiments on the individual SMO gas sensors demonstrated the sensors’ applicability for the detection of CW agents. From 10 ppb and on, the sensors reliably detected sarin, soman, and mustard gas. There exists a good correlation between the results obtained from the tests with simulants and the tests with live agents that support the validity for the choice of simulants. We also showed that the

54


Fig. 12. Sensor array: LDA visualization of the feature space.

so-called non-selective sensors have potential to discriminate between gases using the transient response. This was due to the difference in molecular structure of the gases causing different surface–gas interaction mechanisms, which are converted into observable difference in response shapes. However, the sensors responding to an individual gas (or to a small group of gases) are very desirable. Two thick-film gas sensitive compositions exhibited improved selectivity to methanol (CuO:Ga2 O3 ) and to DCP (Y2 O3 ) were found. These sensors along with the sensors based on WO3 were further investigated as a sensor array. 3.2. Matrix of SMO sensors as a sensor array for CWA detection Due to the good selectivity offered by some of the individual sensors, only the CuO:Ga2 O3 , Y2 O3 and WO3 sensors were used in the array. When adding a sensor into the array it is important that it provides new information otherwise it is just a source of noise that dilutes the other sensors. Ideally we would have n orthogonal responding sensors for detecting n gases. An example of an orthogonal response would be sensors A only responds to gas 1, sensor B only responds to gas 2, and so on, i.e. all the vectors to each point in the response space are orthogonal. The LDA plot in Fig. 12(b) gives a visualization of the feature space of the sensor array. From this plot it is easy to see that the sensor array provides good selectivity between the test gases.

4. Conclusions Several types of thick-film sensors based on different nanocrystalline metal oxides, namely, SnO2 , WO3 , In2 O3 , CuO, and Y2 O3, were fabricated using commercial sensor platforms and a drop-coating technique. The sensors were thoroughly examined over a range of temperatures for sensitivity and selectivity towards CWA simulants. The sensors showed many positive characteristics in terms of sensitivity, stability, and response speed. The sensors based on WO3 , In2 O3 , and CuO operating at 400 ◦ C were additionally tested with CW agents at a certified live agent facility. The experiments supported the sensors’ potential applicability for the

detection of CW agents. From 10 ppb and higher the sensors reliably detected sarin, soman, and mustard gas. We have shown using an array of non-selective sensors that we can differentiate between several gases. This is due to the unique gas–semiconducting film interaction that is converted to an observable change in resistance. In some applications where the molecular structure of the gases is varied (methanol, DCP and DMMP for example), a single SMO sensor can discriminate the gases. However, the sensors responding to an individual gas (or to a small group of gases) are very desirable. Two thick-film gas-sensitive compositions exhibited improved selectivity to methanol and to DCP were found. These are based on CuO:Ga2 O3 and on Y2 O3 , correspondingly. CuO:Ga2 O3 -based thick film operating at 200 ◦ C reversibly changed their resistance during adsorption/desorption of methanol and did not react to hexane, diesel oil vapour, or the simulants. The Y2 O3 -based thick film revealed a remarkable selectivity towards DCP at 600–650 ◦ C. Using these two types of sensors and WO3 -based sensor as a sensor array all of the test gases were able to be detected and identified.

Acknowledgement This work was supported by Sensor Research and Development Corporation under contract #N00014-01-C-0132 from the Office of Naval Research, USA.

References [1] C. Zander, J.E. Enderlein, R.A. Keller (Eds.), Single-molecule Detection in Solution: Methods and Applications, Wiley-VCH, Hoboken, NJ, 2002. [2] M.E. Kosal, The basics of chemical and biological weapons detectors, CNS (Center for Nonproliferation Studies), Research story of the week, November 24, 2003, http://cns.miis.edu/ pubs/week/031124.htm. [3] C.K. Ho, M.T. Itamura, M. Kelley, R.C. Hughes, Review of Chemical Sensors for In-Situ Monitoring of Volatile Contaminations, Sandia Report, SAND2001-064, printed March 2001, http:// www.sandia.gov/sensor/SAND2001-0643.pdf. [4] Office of Naval Research, USA, Contract #N00014-01-C-0132, The Development and Prototyping of a Chemical Warfare Agent Detector using Solid State Sensor Technologies. [5] M. Utrainen, E. Kärpänoja, H. Paakkanen, Combining miniaturized ion mobility spectrometer and metal oxide gas sensor for the fast detection of toxic chemical vapors, Sens. Actuators B: Chem. 93 (2003) 17–24. [6] Office of Naval Research, USA, Media Release of 3/12/2003, Personal Chemical Agent Detector Ready for Trials, http://www.onr.navy.mil/media. [7] S. Roy Morrison, Selectivity in semiconductor gas sensors, Sens. Actuators 12 (1987) 425–440. [8] V.Ya. Sukharev, in: L.Yu. Kupriyanov (Ed.), Semiconductor Sensors in Physico-Chemical Studies, Elsevier, Amsterdam, 1996, pp. 5–101 (Chapter 1). [9] T.C. Pearce, S.S. Schiffman, H.T. Nagle, J.W. Gardner, Handbook of Machine Olfaction: Electronic Nose Technology, Wiley-VCH, Weinheim, 2003.

A.A. Tomchenko et al. / Sensors and Actuators B 108 (2005) 41–55 [10] A.A. Tomchenko, G.P. Harmer, B.T. Marquis, J.W. Allen, Semiconductor metal oxide sensor array for the selective detection of combustion gases, Sens. Actuators B: Chem. 93 (2003) 126–134. [11] Umwelt Sensor Technik, Gas sensors, http://www. umweltsensortechnik.de/index3.htm. [12] NanoProducts Corporation: the building block of innovations, http://www.nanoproducts.com/site/index.php. [13] Nanophase Technologies Corporation: transforming the boundaries of performance, http://www.nanophase.com/catalog/index.asp. [14] J.W. Allen, B.T. Marquis, D.J.R. Smith, Investigating the long-term heating and analyte exposure effects on tin oxide thick-film sensors, IEEE Sensors 2002, in: Proceedings of the First IEEE International Conference on Sensors, vol. 2, Orlando, FL, USA, June 11–14, 2002, pp. 1260–1265. [15] G. Harmer, C. Yang, B. Marquis, Detection of chemical warfare agents in the presence of interferents, IEEE Sensors 2004. In: Proceedings of the Third IEEE International Conference on Sensors, Vienna, Austria, October 24–27, in press. [16] J.G. Ekerdt, K.J. Klabunde, J.R. Shapley, J.M. White, J.T. Yates Jr., Surface chemistry of organophosphorous compounds, J. Phys. Chem. 92 (1988) 6182–6188. [17] I.A. Pushkin, V.V. Batyrev, E.B. Kozlovskaya, The current status of understanding of CWA burning in a fire flame, Chemistry in Russia (Bulletin of RCS) N 11 (1999), http://www.chemnet. ru/rus/journals/xr/ob.html (in Russian). [18] M.K. Templeton, W.H. Weinberg, Adsorption and decomposition of dimethyl methylphosphonate on an aluminium oxide surface, J. Am. Chem. Soc. 107 (1985) 97–108. [19] M.B. Mitchell, V.N. Sheinker, E.A. Mintz, Adsorption and decomposition of dimethyl methylphosphonate on metal oxides, J. Phys. Chem. 101 (1997) 11192–11203. [20] C.S. Kim, R.J. Lad, C.P. Tripp, Interaction of organophosphorous compounds with TiO2 and WO3 surfaces probed by vibrational spectroscopy, Sens. Actuators B: Chem. 76 (2001) 442– 448. [21] F.F. Wolkenstein, Physico-Chemistry of Semiconductor Surface, Nauka, Moscow, 1973, pp. 79–88, Chapter 2 (in Russian).

55

Biographies Alexey A. Tomchenko was born in 1958 in Minsk, Belarus. He received the MSc degree in electronic engineering in 1984 from Minsk Radio Engineering Institute and the PhD degree in electronic engineering in 1999 from the Institute of Electronics of the National Academy of Sciences of Belarus. In 1981–2000 he worked at the Physical Technical Institute of the Academy of Sciences of Belarus, Minsk, first as a Test Engineer and then as a Researcher of the staff. He has been with Sensor Research and Development Corporation since 2000 and is currently their Senior Research Scientist. His research interests are chemistry, physics and technology of oxide films, chemical gas sensors, and sensor arrays. Gregory P. Harmer received the BSc (Applied Maths and Computer Science) degree in 1996, the BE (Elec, Hons I) degree in 1997 and the PhD in 2001, all from the University of Adelaide. He was an invited speaker at UPoN’99, Adelaide, Australia. He currently works at Sensor Research and Development Corporation and is studying sensor noise and signal processing techniques for sensor arrays. Brent T. Marquis received his BS and MS degrees in Electrical Engineering from the University of Maine in 1996 and 2000, respectively. He joined SRD in 1995 as a research engineer and is now their Director of Research Engineering. He has over 10 years of SMO and SAW sensor research and development experience and has published several papers related to SMO sensors, platforms, and operational characterizations. Mr. Marquis has expertise in SMO platform development, sensor testing systems design, thin and thick film development, and data processing in support of on-board computational power for miniaturized sensors. Mr. Marquis has special technical expertise in thin film deposition and metal oxide film design and engineering for selective detection of chemical warfare agents, combustion gases, and halogenated hydrocarbon species, as well as extensive experience in the development of adaptable sensor platforms for a variety of harsh industrial environments. He has been SRDs principal investigator on several sensor programs for the Department of Defence, the Department of Energy, and the National Science Foundation.