Gas Analysis by Monitoring Molecular Diffusion in a Microfluidic Channel

Sep 8, 2010 - Electronic Materials Laboratory, Electrical Engineering Department, K. N. Toosi University of Technology,. Tehran 16315-1355, Iran. Desp...
2 downloads 0 Views 2MB Size
Download PDF
Anal. Chem. 2010, 82, 8349–8355

Gas Analysis by Monitoring Molecular Diffusion in a Microfluidic Channel Faramarz Hossein-Babaei* and Vahid Ghafarinia Electronic Materials Laboratory, Electrical Engineering Department, K. N. Toosi University of Technology, Tehran 16315-1355, Iran Despite the increasing industrial and domestic demand for gas analyzers, the unmanageable size and cost of the available devices prevent them from fulfilling this pervasive need. In this paper, we demonstrate that, monitored by a generic gas sensor, the progress rate of a gaseous analyte’s free diffusion through an air-filled centimeterlong microfluidic channel yields sufficient information for gas recognition. The installation of additional channels made from different materials provides more uncorrelated information, enabling more detailed gas analyses. This device classifies gases based on their physical features at room temperature, namely, diffusivity and interaction (physisorption/desorption) with channel walls. This process does not degrade the gas discriminating element of the instrument. Our prototype, featuring a 50 mm long 50 µm bore borosilicate glass channel, successfully differentiated between 24 analytes, including four butanol isomers, and estimated the compositions of their binary and ternary gas mixtures dispersed in air at different concentration levels. The findings presented here are directly applicable to the production of a new inexpensive, compact, portable, and durable generation of artificial olfaction system whose performance is almost entirely independent of the utilized gas sensor’s drift. Size and cost are widely recognized factors preventing odor and gas identification devices from fulfilling an enormous range of roles, from bacteria recognition1,2 and food quality assessment3,4 to safety5,6 and environmental monitoring.7,8 Despite numerous reported miniaturization efforts,9-12 bulky and expensive gas * To whom correspondence should be addressed. Phone: +98 21 88734172. Fax: +98 21 88768289. E-mail: [email protected], [email protected]. (1) Sauer, S.; Kliem, M. Nat. Rev. Microbiol. 2010, 8, 74–82. (2) Turner, A. P. F.; Magan, N. Nat. Rev. Microbiol. 2004, 2, 161–165. (3) Careri, M.; Bianchi, F.; Corradini, C. J. Chromatogr., A 2002, 970, 3–64. (4) Lehotay, S.; Hajslova, J. Trends Anal. Chem. 2002, 21, 686–697. (5) Janata, J. Annu. Rev. Anal. Chem. 2009, 2, 321–331. (6) Yinon, J. Anal. Chem. 2003, 75, 99A–105A. (7) Marle, L.; Greenway, G. M. Trends Anal. Chem. 2005, 24, 795–802. (8) Aragon, P.; Atienza, J.; Climent, M. D. Crit. Rev. Anal. Chem. 2000, 30, 121–151. (9) Shortt, B. J.; Darrach, M. R.; Holland, P. M.; Chutjian, A. J. Mass Spectrom. 2005, 40, 36–42. (10) Sinha, M. P.; Gutnikov, G. Anal. Chem. 1991, 63, 2012–2016. (11) Ouyang, Z.; Cooks, R. G. Annu. Rev. Anal. Chem. 2009, 2, 187–247. (12) Snyder, P.; Harden, C. S.; Brittain, A. H.; Kim, M. G.; Arnold, N. S.; Meuzellar, H. L. C. Anal. Chem. 1993, 65, 299–306. 10.1021/ac101767r  2010 American Chemical Society Published on Web 09/08/2010

chromatography/mass spectrometry systems9-14 fail to satisfy the requirements of common usage. “Electronic noses”15-19 are also costly and operate with sensor arrays whose cumbersome drifts necessitate frequent calibration and array replacement. These unavoidable drifts generally result from the chemical interaction between the analytes and array components, which in most of the cases, takes place at elevated temperatures. In gas chromatography, the diluted analyte is monitored for recognition after it is forced through meter-long columns.13,14 This paper demonstrates that, monitored by a generic gas sensor, the progress rate of a gaseous analyte’s free diffusion through an airfilled centimeter-long microfluidic channel20-24 yields sufficient information for gas recognition. The installation of additional channels made from different materials provides more uncorrelated information, enabling “higher order”25 gas analyses. This gas analyzer classifies gases based on their physical features at room temperature, namely, diffusivity26 and interaction (physisorption/desorption)27 with channel walls, rather than their chemical properties at elevated temperatures. As a result, the degradation of the gas discriminating part of the instrument, i.e., the channel, is eliminated. EXPERIMENTAL SECTION The microfluidic channel connects a small cavity that houses a generic gas sensor,28 such as a chemoresistive sensor,29 to an analyte-contaminated air chamber (Figure 1A) for a predetermined period (t ) 0 to te). Analyte molecules diffuse through the (13) Grob, R. L.; Barry, E. F. Modern Practice of Gas Chromatography; John Wiley & Sons: Hoboken, NJ, 2004. (14) McNair, H. M.; Miller, J. M. Basic Gas Chromatography; John Wiley & Sons: Hoboken, NJ, 2009. (15) Persaud, K.; Dodd, C. Nature 1982, 299, 352–355. (16) Lim, S. H.; Feng, L.; Kemling, J. W.; Musto, C. J.; Suslick, K. S. Nat. Chem. 2009, 1, 562–567. (17) Raman, B.; Hertz, J. L.; Benkstein, K. D.; Semancik, S. Anal. Chem. 2008, 80, 8364–8371. (18) Rock, F.; Barsan, N.; Weimar, U. Chem. Rev. 2008, 108, 705–725. (19) Pearce, T. C.; Schiffman, S. S.; Nagle, H. T.; Gardner, J. W. Handbook of Machine Olfaction; Wiley-VCH: Weinheim, Germany, 2003. (20) Whitesides, G. M. Nature 2006, 422, 368–373. (21) deMello, J. Nature 2006, 422, 394–402. (22) Weigl, H.; Yager, P. Science 1999, 283, 346–347. (23) Kelly, R. T.; Woolley, A. T. Anal. Chem. 2005, 77, 96A–102A. (24) Hossein-Babaei, F.; Shakerpour, S. J. Appl. Phys. 2006, 100, 124917:1-9. (25) Hierlemann, A.; Gutierrez-Osuna, R. Chem. Rev. 2008, 108, 563–613. (26) Crank, J. The Mathematics of Diffusion; Oxford University Press: London, 1975. (27) Do, D. D. Adsorption Analysis: Equilibria and Kinetics; Imperial College Press: London, 1998. (28) James, D.; Scott, S. M.; Ali, Z.; O’Hare, W. T. Microchim. Acta 2005, 149, 1–17. (29) Yamazoe, N.; Sakai, G.; Shimanoe, K. Catal. Surv. Asia 2003, 7, 63–75.

Analytical Chemistry, Vol. 82, No. 19, October 1, 2010

8349

Figure 1. The prototype analyzer, its processes and responses. (A) Device structure. The large pallet area of the generic gas sensor necessitates using a group of parallel identical microfluidic channels. (B) Analyte diffusion along one channel. Physisorption to the walls reduces the effective diffusion rate. The red, green, violet, and blue arrows illustrate the diffuse-in, physisorption, desorption, and diffuseout processes, respectively. (C) The same as in part B when the channel is reconnected to the clean air reservoir. The pace of analyte depletion is determined by the desorption rate of the molecules and its diffusivity in air. (D) An experimental normalized response profile and its “line diagram”. The line diagram is specified by determining tr and tm′, moments where the response profile reaches 5% and 95% of its maximum level, respectively, and Rf, the dimensionless magnitude of the profile at tf.

channel (Figure 1B), reach the cavity, interact with the sensor’s pallet, and generate a mounting response. By reconnecting the channel to the clean air reservoir at te, the analyte molecules diffuse out from the cavity (Figure 1C) causing response decay which is monitored until tf. The resulting response profile is normalized according to its peak which occurs at tm (Figure 8350

Analytical Chemistry, Vol. 82, No. 19, October 1, 2010

1D). For a fixed channel structure and temperature, the configuration of a normalized transient response profile is almost independent of the fluctuations in both the sensor’s sensitivity and analyte concentration and contains information on the analyte’s molecular diffusivity in air, its propensity to physical adsorption (physisorption) to the channel walls, and the desorption rate of the physisorbed molecules. These profiles are insensitive to the number of identical channels simultaneously utilized for diffusion monitoring, suggesting that the large pallet area and the high analyte consumption rate of generic sensors can be compensated for by using channel bundles instead of a single one (Figure 1A). Details regarding prototype preparation and the measurement scheme are available in the Supporting Information. We rendered a simplified linear representation of a normalized response profile by neglecting its curved details (Figure 1D). These easy to read diagrams contain almost all of the useful information content of the original profile (see below). For a fixed channel, this method can graphically represent any gas or gas mixture with a pair of intersecting straight line segments in the time domain. The “lines” of an analyte are created by determining a set of three independent figures, tr, tm′, and Rf as defined in Figure 1D, from the response profile. Hence, this set of figures, forming the components of a 3-D “feature vector”, θ(tr, tm′, Rf), of the analyte, contains the entirety of the useful information in the analyte lines and can rightfully represent the analyte in the “feature space” (see below). We have used the analyte lines for the graphic representation of the experimental response profiles and the related θs for the analyte classification in the feature space. The raw response profiles obtained for various concentrations of methanol are given in Figure 2A. The output level before t ) 0 indicates the sensor’s baseline (its conductance in clean air.) The exposure of the open end of the channel to the methanol contaminated air at t ) 0 starts the “diffuse in” process which gradually allows the gas to the sensor cavity and increases the response level. At t ) 40 s, the channel is reintroduced to clean atmosphere, which initiates the “diffuse out” process. At t > 40 s, a combination of these two processes are in effect. As a result, the sensor’s output reaches a maximum level, which is higher for higher methanol concentrations and decends, gradually, toward the baseline. The maximum analyte concentration experienced by the sensor is much less than that in the gas chamber, as the channel is disconnected from the gas chamber at t ) 40 s, well before the completion of the diffusion process. Recording is terminated at t ) 160 s. Similar test results obtained for 1-propanol are also presented in Figure 2A. These two sets of results are truncated to cover the 0-120 s time interval and normalized within the [0-1] range. In the final configuration, presented in Figure 2B, regardless of the difference the concentrations levels examined, the response profiles related to the same analyte have appeared alike, while those of the two different analytes are clearly distinguishable. With the use these profiles, the lines presenting these analytes were produced (Figure 2C) based on the above given definitions. This resulted in θ (14.1, 49.7, 0.38) and θ (26.2, 80.0, 0.94) for methanol and 1-propanol, respectively. RESULTS AND DISCUSSION The experimental response profile was mathematically connected to the analyte’s parameters by inserting the quantitative

Figure 2. (A) The raw response profiles obtained for methanol and 1-propanol at the stated analyte concentrations in the gas chamber. (B) The same results after truncation and normalization; the inserts depict changes in the order caused by normalization. (C) Determination of the analytes’ simplified plots (lines) and θ vector components. Thicker methanol lines are due to the repeat experiments at different concentration levels in the wider concentration range of 500-1500 ppm (see the inserts).

description of the physisorption/desorption process, determined by the Langmuir model27 (valid for low analyte concentrations at which the probability of double layer formation on the channel wall surface is negligible), into the “diffusion-physisorption equation”.24 The mathematical description of the instument’s operation is presented in the Supporting Information. The outcome analytically validated our initial results and confirmed the advantages of a microfluidic channel over its large bore counterpart. Large bore channels yield inadequate response profiles containing information solely on the diffusivity of the analytes, differentiating between gases significantly different in diffusivity.30,31 The quantitative analysis also confirmed that, although the information derived from geometrically different channels made from the same fabric (30) Hossein-Babaei, F.; Orvatinia, M. IEEE Sens. J. 2004, 4, 802–806. (31) Hossein-Babaei, F.; Hemmati, M.; Dehmobed, M. Sens. Actuators, B 2005, 107, 461–467.

were correlated, decreasing the channel diameter increased the significance and clarity of the other analyte-related parameters (its physisorption/desorption parameters) in the produced responses; a 2 µm bore channel, approximately 10 mm in length, would provide as much information as our prototype. The results also suggest that the instrument can be compacted by channel design adjustments (e.g., zigzag or spiral.) The device (Figure 1A), then, can be redesigned to be mass-produced with silicon technology32 that will facilitate the integration of the channel and the sensor cavity with the preprocessing circuitry. For analyte recognition, θ was simply marked in the feature space precalibrated according to the θs of gases of interest. While the lines of different analytes were visually distinguishable (Figure 3A-D), those of the same analyte examined at different concentrations in the 500-1500 ppm range were alike and resulted in practically identical θs (see methanol lines and the inserts in Figure 2C). Data classification and mapping techniques, such as principal component analysis,33 could be utilized to map the θs onto a 2- or 1-D feature space if necessary. To gauge the adequacy of the θs as replacements for the response profiles, the original responses, defined as high-dimensional feature vectors, were mapped onto appropriately selected 3-D spaces. Comparisons between the resulting class segregations showed that the diagnostic value of an analyte’s θ and, thus of its lines, was hardly inferior to that of its complicated response profile (Figure S-22 in the Supporting Information). It was interesting to observe the θs of various pure alcohol vapors (except that of tert-butanol) line up on a simple “path” in the feature space (Figure 3E,F) while those of nonalcohols fell outside this track. A similar sense of order was seen in other target gas categories, such as ketones and alkanes, as well (Figure 3F). Accordingly, the location of a known gas in the feature space could be approximated prior to the experiment. Similarly, an unknown pure analyte outside the listed targets could be categorized based on its position in relation to the recognized paths. Sensor degradation and replacement are major problems in the multisensor systems, as variations in sensor parameters distort the aquired discriminative information and necessitate the recalibration of the processing unit. The effect of the normal degradation and replacement of the sensor on the lines of different analytes was investigated by replacing the working sensor of the prototype with randomly selected new and used sensors of the same type. The results, presented in Figure S-9 in the Supporting Information, display practically no change in the analyte lines, and the prototype with the replaced sensor could perform based on the original calibration data. After 13 months of experimental service, our borosilicate glass channels show no signs of degradation. Fluctuations in ambient humidity, temperature, pressure, and air composition impose unwanted variations in the configuration of the responses. At this early stage, the optimum solution for the laboratory applications is to perform the experiment in controlled conditions. All the results reported here have been obtained in air with temperature and relative humidity of 25.0 ± 0.1 °C and 20 ± 2%, respectively. For the field applications, however, we believe the solution is to provide different sets of calibration data, with respect to a number of typical ambient (32) Wise, K. D.; Najafi, K. Science 1991, 254, 1335–1342. (33) Wold, S.; Esbensen, K.; Geladi, P. Chemom. Intell. Lab. Syst. 1987, 2, 37– 52.

Analytical Chemistry, Vol. 82, No. 19, October 1, 2010

8351

Figure 3. Identification of pure analytes. (A-D) The simplified responses (lines) experimentally obtained using a 50 mm long 50 µm bore borosilicate glass channel at 25.0 ( 0.1 °C and relative humidity of 20 ( 2%, for hydrogen (1), carbon monoxide (2), argon (3), oxygen (4), methanol (5), ethanol (6), isopropanol (7), 1-propanol (8), tert-butanol (9), 2-butanol (10), iso-butanol (11), 1-butanol (12), methane (13), n-butane (14), n-pentane (15), acetone (16), butanone (17), 2-pentanone (18), methyl isobutyl ketone (19), chloroform (20), toluene (21), benzene (22), carbon tetrachloride (23), and ammonia (24). Five repeat experiments were carried out for each analyte, except O2 and Ar, at different concentrations in the 900-1100 ppm range. The background gas in all the experiments was clean air. Tests on O2 and Ar were repeated at five concentrations in the 80-95% range. (E) θ vectors marked in the feature space. (F) The sense of order in the locations of alcohols, ketones, and alkanes in the feature space.

conditions; the system would switch over to the optimum set based on the prevailing ambient conditions. For smaller ambient fluctuations occurring around these typical states, however, we intend to utilize available mathematical methods to compensate for the “driftlike” terms. The subject is currently under investigation in our laboratory.34 The idea is to measure fluctuations in environmental parameters and use the obtained data to modify the readouts or the calibrations in the feature space to further reduce the dependencies of the produced discriminative information on the ambient conditions. Assuming an accuracy of ±1% in manufacturing channel dimensions, a single calibration database can be universally employed for all similar channels, as the components of θ, and thus the analyte’s position in the feature space, will not be (34) Hossein-Babaei, F.; Ghafarinia, V. Sens. Actuators, B 2010, 143, 641–648.

8352

Analytical Chemistry, Vol. 82, No. 19, October 1, 2010

displaced more than 2.5%. The finer adjustments of the calibration data can be carried out for each instrument by a single-point standard test (say, at 1500 ppm ethanol in air). The obtained displacement vector is, then, applied to all the θs in the database. Our present design uses bundles of channels instead of a single one; this helps to average out some of the dimensional inaccuracies. The experimental results reported here have been measured at analyte concentrations in the range of 900-1100 ppm. To clarify the effect of analyte concentration on the configuration of the analyte lines and, thus, the position of θ in the feature space, methanol results were reported for the 500-1500 ppm range. An analytical discussion on the effect of gas concentration on the instrument’s performance is given on page S-18 in the Supporting Information, where we show that the θ positions vary minimally with the analyte concentration due to the deviation of the sensor

Figure 4. Estimating composition in two-component gas mixtures. (A) The lines of 1-propanol-methanol mixtures at the stated compositions. (B) The θs of the same gas mixtures marked in the feature space. (C) Same as part B of 2-butanol-acetone mixtures. The insert depicts the cluster resulted from eight repeat experiments. (D) Demonstration of parts B and C in the universal (tr, tm′, Rf) feature space of our prototype (Figure 3E). Note that the “path” of 1-propanol-methanol mixtures (brown) interestingly curves away from that of the pure alcohols (green) providing an easy way to distinguish between mixed and pure substances.

element from linearity (see page S-8 in the Supporting Information). Experiments encountered technical problems below ∼200 ppm due to the low signal levels which made the recorded patterns fuzzy and unreliable. Above 5000 ppm, the sensor’s behavior is substantially nonlinear (see Figure S-6 in the Supporting Information) and can shift θs to undefined locations in the feature space. However, a functioning commercial instrument would use a number of different “analyte exposure time modes” (this was 40 s in all the experiments reported here) which would be selected automatically upon receiving a signal from the reference gas sensor that monitors the analyte concentration level in the gas chamber. Shorter exposure times prevent concentration build up at the sensor cavity from increasing beyond a safe operation limit (say, 1500 ppm). As a result, the instrument would have practically no upper limit for the analyte concentration. The lines of two-component gas mixtures generally vary between those of their pure constituents (Figure 4A) depending on component proportion; the mixture composition can be estimated by a quantitative comparison of their θs in a feature space. The compositions of all the examined chemically stable two-component mixtures were successfully determined with a single microfluidic channel (Figure 4B,C). The precision of the estimations depends on the Euclidean distance between the pure constituents in the feature space. θs of a binary mixture system line up according to composition on a “path” connecting the pure components in the feature space. Interestingly, these “mixture paths” curve away from those related to the pure substances, providing a valuable means of discrimination between mixed and

pure substances (Figure 4D). Because of this feature, no mixture of, say, methanol and isopropanol can overlap with pure ethanol in the feature space. The database produced for the pure analytes is utilized as the calibration basis for the mixtures as well (Figure 4D). Consequently, an instrument assigned to test 1-propanolmethanol mixtures, for instance, can also identify a third pure gas introduced as the analyte (Figure 4B,D, also see Figures S-16-S20 in the Supporting Information). The information content of the θs obtained experimentally for three-component gas mixtures proved sufficient for estimating their compositions (Figure 5A-D, see also Figure S-21 in the Supporting Information). The θ positions related to various compositions of a ternary mixture system occupied a smoothly curved surface in the feature space. The edges of this surface were defined by the paths related to the binary gas mixtures formed by the three components of the mixture. As a result, the locus of the θs of an ABxCy mixture system, where x and y vary independently to form all the possible mixtures between A, B, and C gases, could be determined and approximately calibrated by establishing the ABx, ACx, and BCx paths along with the central ABC point in the feature space. Figure 5D shows three such surfaces obtained for three different ternary gas mixtures in the feature space. The surfaces obtained for the θs of methanol-ethanol-acetone and methanol-ethanol-methane mixtures share the path related to the methanol-ethanol binary mixture as their common edge (Figure 5D). The composition of a mixture of three known components can be determined by plotting its θ in the feature space. The position of θ on the Analytical Chemistry, Vol. 82, No. 19, October 1, 2010

8353

Figure 5. Estimating the composition in three-component gas mixtures. (A) θs of methanol-ethanol-acetone and methane-methanol-ethanol mixtures at the stated compositions marked in the feature space projected to the Rf-tm′ plane for easy reading. (B) The same as part A for tert-butanol-toluene-2-pentanone mixtures projected onto the Rf-tr plane. (C) Classification of 1-butanol-isobutanol-2-butanol mixtures. The six-dimensional feature vectors that resulted from two different material channels are mapped onto a three-dimensional feature space. (D) Demonstration of parts A and B in the universal feature space of our prototype (Figure 3C).

calibrated locus related to that ternary system (Figure 5) will determine the analyte’s composition. θ positions outside the locus would label the analyte as impure containing gases other than the assumed components. In the exceptional cases of three-alcohol (or three-ketone) mixtures, the closeness of the edge paths, related to the involved binary mixtures, allowed only a limitted locus area for the θ vectors in the feature space and impeded correct composition estimation. This was overcome by combining the information obtained from two different channels made of different materials. 1-Butanol-isobutanol-2-butanol mixtures were analyzed by combining the data obtained from two channels made of borosilicate glass and polyvinylidene floride. The 6-D feature vector, formed by combining the components of the two 3-D θs, was mapped onto a 3-D feature space for classification (Figure 5C). Response profiles obtained using different material channels contain uncorrelated information and, when used in conjunction, provide more discriminative information on complex gas mixtures: The analyte diffusion process is independent of the channel material and only depends on the gas type. The physisorption/ desorption process, however, depends both on the gas type and the channel material and can provide uncorrelated information when monitored at different material channels. (For example, such interactions with a hydrophilic and a hydrophobic surface would depend on different features of the analyte molecule and can provide uncorrelated information.) Ideally, each different material channel introduces two independent parameters, and n such channels can, in theory, facilitate the composition estimation of a 8354

Analytical Chemistry, Vol. 82, No. 19, October 1, 2010

(2n + 1)-component gas mixture. This has been analytically described in the Supporting Information, Figure S-13. Each channel has its own calibration database. When using two or more different channels in conjunction, the dimensions of the θ and the feature space increases to 6 or more (which can later be mapped onto a 3-D space for visualization, as carried out in Figure 5C.) The calibration of this high-D multichannel feature space, however, can result from the calibration databases of the standalone channels. Channel materials (and coatings) can be chosen to provide more uncorrelated information. Considering the vast background literature available on GC columns, further work should quickly prove fruitful. In the case of multicomponent gas mixtures and complex odors, our single channel analyzer can still function as an artificial olfaction system for numerous quality and process control applications. The operation is based on comparing the experimental θ of the analyte with those available from the previous experiences. Detecting binary and ternary gas mixtures was the intended extent of the present report. Much work is required on multicomponent gas mixtures. CONCLUSIONS We have introduced a key principle that the recorded rate of the free diffusion of a gaseous analyte along a microfludic channel yields sufficient information on its physical features to facilitate its identification. The analyzer designed based on this percept is inexpensive, small, mass producible via various existing technologies, including silicon technology, and durable because the gas

discriminating channel suffers no degradation in the analyte examination process. Additionally, our graphic method for extracting information from the recorded responses of the instrument drastically reduces the memory and computation capacity required for the analyses. This foreshadows a portable analyzer with an integrated processing unit. The success of our prototype in identifying 24 gaseous analytes and estimating the composition of their stable two- and threecomponet mixtures in 120 s per analyte was demonstrated. The concept allows for much flexibility in terms of design: custommade small pallet area sensors, for example, would allow using finer channels, enhancing the gas discrimination ability while decreasing the overall size of the device. Also, a combination of multiple devices enables analyses of more complex gas mixtures. The method and instrument described here are suitable for a large

number of applications, and their new practical features promise to add momentum to the already exciting field of gas diagnosis. ACKNOWLEDGMENT The authors thank B. Bahraminejad and A. H. Alibeigi of KNTU for their assistance at the early stages of the experimental work and M. Hossein-Babaei for proofreading the manuscript. SUPPORTING INFORMATION AVAILABLE Aditional information as noted in text. This material is available free of charge via the Internet at http://pubs.acs.org. Received for review July 4, 2010. Accepted August 28, 2010. AC101767R

Analytical Chemistry, Vol. 82, No. 19, October 1, 2010

8355