7928
Langmuir 2006, 22, 7928-7935
Vapor Sensing Using Polymer/Carbon Black Composites in the Percolative Conduction Regime Brian C. Sisk and Nathan S. Lewis* DiVision of Chemistry and Chemical Engineering, Noyes Laboratory, 127-72, California Institute of Technology, Pasadena, California 91125 ReceiVed December 5, 2005. In Final Form: April 5, 2006 To investigate the behavior of chemiresistive vapor sensors operating below or around the percolation threshold, chemiresistors have been formed from composites of insulating organic polymers and low mass fractions of conductive carbon black (CB, 1-12% w/w). Such sensors produced extremely large relative differential resistance changes above certain threshold vapor concentrations. At high analyte partial pressures, these sensors exhibited better signal/noise characteristics and were typically less mutually correlated in their vapor response properties than composites formed using higher mass fractions of CB in the same set of polymer sorption layers. The responses of the low-mass-fraction CB sensors were, however, less repeatable, and their nonlinear response as a function of analyte concentration required more complicated calibration schemes to identify and quantify analyte vapors to compensate for drift of a sensor array and to compensate for variability in response between sensor arrays. Because of their much larger response signals, the low-mass-fraction CB sensors might be especially well suited for use with low-precision analog-to-digital signal readout electronics. These sensors serve well as a complement to composites formed from higher mass fractions of CB and have yielded insight into the tradeoffs of signal-to-noise improvements vs complexity of signal processing algorithms necessitated by the use of nonlinearly responding detectors in array-based sensing schemes.
I. Introduction Arrays of broadly cross-reactive sensors have received significant attention for their possible use in the detection and classification of analyte vapors. Many signal transduction modalities, including polymer-coated quartz-crystal microbalances (QCMs) or surface-acoustic wave (SAW) devices,1-3 glass beads or optical fibers coated with dye-impregnated polymers,4-8 conducting polymer9-11 or polymer composite12-14 chemically sensitive resistors, polymer-coated micromachined cantilevers,15 polymer-based capacitors and field-effect transistors,16,17 and metal oxide chemiresistors,18-21 have been used in such sensor arrays. Work in our laboratory has focused on the development of conducting composite films consisting of insulating organic * To whom correspondence should be addressed. (1) Ballantine, D. S.; Rose, S. L.; Grate, J. W.; Wohltjen, H. Anal. Chem. 1986, 58, 3058-3066. (2) Patrash, S. J.; Zellers, E. T. Anal. Chem. 1993, 65, 2055-2066. (3) Rose-Pehrsson, S. L.; Grate, J. W.; Ballantine, D. S.; Jurs, P. C. Anal. Chem. 1988, 60, 2801-2811. (4) Dickinson, T. A.; Michael, K. L.; Kauer, J. S.; Walt, D. R. Anal. Chem. 1999, 71, 2192-2198. (5) Dickinson, T. A.; White, J.; Kauer, J. S.; Walt, D. R. Nature 1996, 382, 697-700. (6) Rakow, N. A.; Suslick, K. S. Nature 2000, 406, 710-713. (7) Ronot, C.; Archenault, M.; Gagnaire, H.; Goure, J. P.; Jaffrezicrenault, N.; Pichery, T. Sens. Actuators, B 1993, 11, 375-381. (8) White, J.; Kauer, J. S.; Dickinson, T. A.; Walt, D. R. Anal. Chem. 1996, 68, 2191-2202. (9) Bartlett, P. N.; Archer, P. B. M.; Ling-Chung, S. K. Sens. Actuators 1989, 19, 125-140. (10) Gardner, J. W.; Pike, A.; Derooij, N. F.; Koudelkahep, M.; Clerc, P. A.; Hierlemann, A.; Gopel, W. Sens. Actuators, B 1995, 26, 135-139. (11) Shurmer, H. V.; Gardner, J. W.; Corcoran, P. Sens. Actuators, B 1990, 1, 256-260. (12) Freund, M. S.; Lewis, N. S. Proc. Natl. Acad. Sci. U.S.A. 1995, 92, 26522656. (13) Lonergan, M. C.; Severin, E. J.; Doleman, B. J.; Beaber, S. A.; Grubbs, R. H.; Lewis, N. S. Chem. Mater. 1996, 8, 2298-2312. (14) Severin, E. J.; Doleman, B. J.; Lewis, N. S. Anal. Chem. 2000, 72, 658668. (15) Lang, H. P.; Baller, M. K.; Berger, R.; Gerber, C.; Gimzewski, J. K.; Battiston, F. M.; Fornaro, P.; Ramseyer, J. P.; Meyer, E.; Guntherodt, H. J. Anal. Chim. Acta 1999, 393, 59-65.
sorptive phases filled with conductive carbon black (CB).22 In these systems, current passes through continuous pathways of the conductive CB that traverse the gap between the pair of contacting electrodes. The sorption of a chemical vapor leads to swelling of the film, which breaks some of the continuous CB pathways and increases the dc resistance of the composite.23 At relatively high mass fractions of CB, such sensors have been shown to produce a linear response to the concentration of analyte, facilitating a straightforward analysis of the data and enabling quantification of analytes and analyte mixtures using simple linear signal processing methods.24-26 Given the relatively low noise exhibited by these sensors of approximately 1 part in 90 000 of the mean baseline resistance value,27 as well as their high degree of repeatability,28 relatively small responses can still be recorded with a high degree of fidelity. In this work, we have explored the behavior of composites that have been deliberately filled with low mass fractions of CB. Such sensors ought to have a larger response to a given analyte because of the reduced number of pathways for conduction in the film but additionally ought to exhibit a nonlinearity in response as a function of analyte concentration as the sorption of analyte(16) Cornila, C.; Hierlemann, A.; Lenggenhager, R.; Malcovati, P.; Baltes, H.; Noetzel, G.; Weimar, U.; Gopel, W. Sens. Actuators, B 1995, 25, 357-361. (17) Torsi, L.; Dodabalapur, L.; Sabbatini, L.; Zambonin, G. Sens. Actuators, B 2000, 67, 312-316. (18) Corcoran, P.; Shurmer, H. V.; Gardner, J. W. Sens. Actuators, B 1993, 15, 32-37. (19) Gardner, J. W.; Shurmer, H. V.; Corcoran, P. Sens. Actuators, B 1991, 4, 117-121. (20) Watson, J. Sens. Actuators 1984, 5, 29-42. (21) Yamazoe, N. Sens. Actuators, B 1991, 5, 7-19. (22) Lewis, N. S. Acc. Chem. Res. 2004, 37, 663-672. (23) Blythe, A. R. Electrical Properties of Polymers; Cambridge University Press: Cambridge, UK, 1979. (24) Severin, E. J.; Lewis, N. S. Anal. Chem. 2000, 72, 2008-2015. (25) Sotzing, G. A.; Phend, J. N.; Grubbs, R. H.; Lewis, N. S. Chem. Mater. 2000, 12, 595-595. (26) Tillman, E. S.; Koscho, M. E.; Grubbs, R. H.; Lewis, N. S. Anal. Chem. 2003, 75, 1748-1753. (27) Briglin, S. M.; Freund, M. S.; Tokumaru, P.; Lewis, N. S. Sens. Actuators, B 2002, 82, 54-74. (28) Sisk, B. C.; Lewis, N. S. Sens. Actuators, B 2005, 104, 249-268.
10.1021/la053287s CCC: $33.50 © 2006 American Chemical Society Published on Web 07/26/2006
Vapor Sensing Using Polymer/Carbon Black Composites
Langmuir, Vol. 22, No. 18, 2006 7929
Table 1. Summary of Experiment Runsa
a
run
since last run
analyte 1
analyte 2
analyte 3
analyte 4
1 2 3 4 5 6
N/A 2 weeks 1 day 3 months 1 day 1 day
iso-butyl acetate iso-butyl acetate iso-butyl acetate iso-butyl acetate 1,1,1-trichloroethane dichloromethane
1-chlorobutane 1-chlorobutane 1-chlorobutane 1-chlorobutane pyridine isopropyl benzene
ethanol ethanol ethanol ethanol decane methyl acetate
water water water water n-octanol methanol
All experiments also used isooctane, toluene, THF, and chloroform.
induced film-swelling drives the composite through the percolation threshold.13,23 This larger response will produce improved signal levels and will therefore require fewer bits of precision in any accompanying read-out electronics but will also require nonlinear algorithms to analyze the resulting sensor array response data. Use of such sensors therefore allows exploration of the tradeoffs involved in using nonlinearly responding sensors in array-based vapor detection systems relative to arrays that have linearly responding sensors. Questions of interest concern the ability of such arrays to robustly classify and quantify analytes, the signal/noise levels of such sensors relative to linearly responding composite chemiresistors, the robustness of the signal profiles to drift, and to varying levels of interferents in the background ambient. II. Experimental Section A. Sensor Fabrication. Sensors were fabricated from either poly(ethylene oxide) (PEO), molecular weight (MW) ) 100 000 Da; poly(ethylene-co-propylene) (PEP), 40% propylene; poly(ethyleneco-vinyl acetate) (PEVA), 40% vinyl acetate; or poly(vinyl stearate) (PVS). The polymer of interest was dissolved in a compatible solvent (CHCl3 for PEO; xylene at 50 °C for PEP; THF for PEVA and PVS) and CB (Black Pearls 2000, Cabot Co.) was then added to produce the desired mass ratio of polymer-CB. The total mass of polymer and CB was 200 mg in 20 mL of solvent. For a given polymer, sensors were formed using both low- and high-mass fractions of CB. The high (40% w/w)-mass fraction CB-polymer composite sensors have been studied previously and generally exhibit a linear response vs analyte concentration.13,29 Low-mass-fraction (2%, 12%, 7%, and 1% for PEO, PEP, PEVA, and PVS, respectively) CBpolymer composites were prepared using the smallest quantity of CB that produced a viable (measurable baseline resistance and measurable response to analytes) sensor for that polymer type. The polymer-CB suspension was sonicated for 30 min to break up the agglomerated CB particles, producing nanoparticles ≈20 nm in radius.30 Electrodes were formed from interdigitated metal lines having gaps of 10 µm and a total interfacial contact distance of 2 cm. Sensor films were sprayed (with an airbrush) onto these electrodes.13,27,28 The resistances of the resulting sensors were 0.20-0.30, but the film/vapor partition coefficient monotonically increased with further increases in the analyte partial pressure. The ∆meq/mb vs P/P0 behavior of a 10% CBPVS composite film was very similar (i.e., varied by less than 10%) to that of the 1% CB-PVS composite. Hence, the mass uptake properties of these films are determined by the polymer/ vapor sorption process and not by the CB. Figure 1b shows the ∆Req/Rb response data for the same PVS1% CB composite film, and Figure 2a shows the data for a 40% by mass CB-PVS composite exposed to THF. The ∆Req/Rb signal was proportional to P/P0 for the 40% CB film but not for the 1% CB film. As shown in Figure 1c, for the PVS-1% CB composite, the value of ∆meq/mb was proportional to ∆Req/Rb for P/P0 < 0.10 (Figure 1c), at which point the relative differential resistance response increased significantly with respect to further increases in ∆meq/mb. This is consistent with the behavior expected when analyte-induced swelling of the composite drives the film conduction mechanism through the percolation threshold.13,35 Similar behavior was observed for the other low-mass-fraction CB sensors investigated in this work. For example, Figure 2a-d compare the ∆Req/Rb response of high- and low-mass-fraction CB composites of PVS and PEP, respectively, as a function of the partial pressure of isooctane, THF, and chloroform. In all cases, the low-mass-fraction CB sensors exhibited a linear response of ∆Req/Rb vs P/P0 at low analyte concentrations and then displayed much larger, nonlinear, responses at higher analyte concentrations. B. Determination of Signal-to-Noise Characteristics. Signalto-noise (S/N) values were derived from each exposure to all analytes of the first data collection period. The signal was taken from the ∆Req/Rb values, and the noise level was taken to be three times the detrended standard deviation of the baseline resistance value.36 The sensing film area was approximately 14 mm2 for the high-mass-fraction CB sensors, while the low-massfraction CB sensors had an area of approximately 1 mm2. The S/N scales as the square root of the sensing area when the sensor thickness is held constant and the sensor dimensions are larger than the correlation length of the physical process that dominates the sensor noise. This is the case for CB composite sensors, for which the noise at low frequencies is 1/f in character and scales with the square root of the sensor area (at constant film thickness),27 so the raw S/N values for the low-mass-fraction CB composite sensors should be scaled by a factor of ∼3.7 for consistent comparison with the S/N values measured for the high-mass-fraction CB composite sensors in this work. As would be expected, the high-mass-fraction CB-polymer composite sensors showed an approximately linear dependence (33) Burl, M. C.; Sisk, B. C.; Vaid, T. P.; Lewis, N. S. Sens. Actuators, B 2002, 87, 130-149. (34) Vaid, T. P.; Burl, M. C.; Lewis, N. S. Anal. Chem. 2001, 73, 321-331. (35) Tillman, E. S.; Lewis, N. S. Sens. Actuators, B 2003, 96, 329-342. (36) Skoog, D. A.; Holler, F. J.; Nieman, T. A. Principles of Instrumental Analysis, 5th ed.; Harcourt-Brace: Philadelphia, 1998.
Figure 1. Response vs concentration profiles for a poly(vinyl stearate) 1% CB sensor exposed to tetrahydrofuran: (a) ∆m/mb vs concentration and (b) ∆Req/Rb vs concentration. Data were recorded from a “sawtooth” concentration vs time profile. (c) ∆Req/Rb vs ∆m/mb.
of S/N on analyte concentration, while the low-mass-fraction CB-polymer sensors showed a generally exponential increase in S/N as analyte concentration increased. At low analyte partial pressures, the high-mass-fraction CB composites showed a higher average S/N than the low-mass-fraction composites, while the low-mass-fraction CB films showed higher S/N values at high
Vapor Sensing Using Polymer/Carbon Black Composites
Langmuir, Vol. 22, No. 18, 2006 7931
Figure 2. ∆Req/Rb response vs concentration for isooctane (×), tetrahydrofuran (O), and chloroform (0) exposed to composites of (a) poly(vinyl stearate), 40% CB; (b) poly(vinyl stearate), 1% CB; (c) poly(ethylene-co-propylene), 40% CB; and (d) poly(ethylene-co-propylene), 12% CB. Data were recorded from a series of “boxwave” concentration vs time profiles.
analyte partial pressures. The high- and low-mass-fraction CB sensors yielded similar results near P/P0 ) 0.060 before scaling the S/N values to account for the differences in electrode area. Upon scaling, the low- and high-mass-fraction CB composite sensors yielded similar S/N values at concentrations as low as P/P0 ) 0.020, whereas the low-mass-fraction CB composites yielded an approximately 8-fold increase in S/N at P/P0 ) 0.16 for the analytes investigated. C. Sensor Array Response Data. Figure 3 displays the first principal component, which contained the bulk of the variance from the data, as a function of P/P0 for selected analytes presented to arrays of (a) the high-mass-fraction CB sensors and (b) the low-mass-fraction CB sensor sets. The first principal component was monotonic and approximately linear with P/P0 for the highmass-fraction CB sensors, whereas it increased approximately exponentially for the low-mass-fraction CB sensors. Consideration of further principal components allowed for recognition of percolative effects earlier than may be possible on a simple sensor by sensor basis. For example, 1-chlorobutane and isopropyl benzene were easily recognized as displaying significant nonlinear behavior in the fourth, and least significant, principal component at P/P0 ) 0.06, while the first principal component did not clearly display such behavior until P/P0 ) 0.10. To evaluate the degree of orthogonality provided by sensors having the same polymeric component but different mass fractions of CB, correlation matrixes were derived from the first run of the high- and low-mass-fraction CB PEO, PEP, PEVA, and PVS composite sensors upon exposure to isooctane, toluene, THF, isobutyl acetate, 1-chlorobutane, chloroform, ethanol, and water. Table 2 displays this correlation matrix derived from the lowest concentrations (P/P0 < 0.06) of each analyte (averaged over the eight analytes used) and presents data derived from the highest
concentrations (P/P0 > 0.10). The low-mass-fraction CB sensors were in general less mutually correlated than the high-massfraction CB sensors, particularly at the higher analyte concentrations where percolation behavior was evident. The high- and low-mass-fraction CB sensor arrays were also compared on the basis of their ability to statistically separate various pairs of analyte responses into distinct sensor response feature vectors. FLD was used to distinguish pairwise among the eight different analytes of the first data collection run, with a resolution factor (RF) evaluated for each of these binary separation tasks.32 The RF value was derived from the mean and variance of the two populations of concern:
RF )
σ21
δ + σ22
(2)
where δ is the difference in the mean vector responses of the two populations and σ1 and σ2 are the standard deviations of the two populations. All possible tasks of distinguishing between two analytes at the same concentration were evaluated, and the process was repeated for each of the concentrations at which data were collected. For 31.7% of the FLD separation tasks involving analytes having P/P0 > 0.06, the low-mass-fraction CB composite sensor array yielded greater RF values than the array that had only high-mass-fraction CB composite sensors. Specifically, the lowmass-fraction CB detector array yielded RF values at least 50% higher than the high-mass-fraction CB sensor array for 14.8% of the tasks and yielded RF values at least 100% higher for 6.2% of such separation tasks. The respective ratios for separation tasks at analyte partial pressures with P/P0 < 0.06 were 3.0%, 0.67%, and 0.17%, respectively. Thus, at high analyte concentrations, the low-mass-fraction CB sensor array often yielded
7932 Langmuir, Vol. 22, No. 18, 2006
Figure 3. Most significant principal component of a data matrix derived from exposure of (a) the four sensors containing high CB loadings and (b) the four sensors containing low CB loadings to each of the 16 analytes tested. Highlighted are water (0), chloroform ()), isopropyl benzene (O), and methyl acetate (3).
more robust separation between analyte clusters than the highmass-fraction CB composite sensor array, but at low analyte concentration, the high-mass-fraction CB sensor array generally exhibited superior separation between analytes. While these results do not indicate that low-mass-fraction CB sensors could fully replace high-mass-fraction CB sensors, they indicate that the sensor types are complementary in their response properties. To evaluate the analyte classification performance of both types of arrays, a nearest-neighbor approach32 was used in which each of the 80 analyte/concentration combinations was treated as a separate analyte. The Euclidean distance was calculated between the means of each of the 80 analyte/concentration clusters and the feature vector produced by each analyte exposure. For unnormalized data, the classification performance overall was poor (20 bits of resolution. In handheld devices, such capabilities are rarely available, and digital noise will then be significant compared to the noise of the sensors. In such cases, the sensor that produces the largest signal will be preferred. Considering the most extreme case, one might seek a sensor that can yield meaningful results when using even a 1-bit conversion, assuming only that one has the sensor in a circuit that allows it to have a variable “turn-on” resistance. In this case, the high-mass-fraction CB sensors would be difficult to use, as their linear response to analyte concentration would provide no convenient point (such as a percolation threshold) at which to designate the “turn-on” point. In contrast, for low-mass-fraction CB sensors, the enhanced response and nature of the percolation threshold could allow low-mass-fraction CB sensors to be treated as either “on” or “off.” Another potential benefit of considering the highly nonlinear percolative sensors from a binary on/off standpoint is as a mimic of the mammalian olfaction system, which (at the lowest level) consists of receptor neurons that either do or do not fire.37 Taken together, a large array of percolative sensors would then generate, in effect, a bit vector for any analyte exposure, similar to how the response set is generated by olfactory receptors. In this way, (37) Buck, L.; Axel, R. Cell 1991, 65, 175-187.
Vapor Sensing Using Polymer/Carbon Black Composites
an array of sensors functioning in a binary capacity could simulate saturation of receptors in the nasal epithelium, particularly if a variety of CB loadings are used, which would represent various receptors of a given type that are more or less sensitive.38
V. Conclusions Low-mass-fraction CB sensors present a variety of advantages when used in conjunction with high-mass-fraction CB sensors. The low-mass-fraction composites generate larger signals, often better resolve analytes, and are generally more sensitive above concentrations of P/P0 ≈ 0.06. These advantages are tempered, however, by the more significant drift, lack of linearity in response vs analyte concentration, and lesser response reproducibility. Even including such considerations, sensor arrays derived from low mass fractions of CB typically performed nearly as well as high-mass-fraction CB sensor arrays for all but the lowest analyte concentrations and often performed better. While low-mass(38) Johnson, B. A.; Leon, M. J. Comput. Neurol. 2000, 422, 496-509.
Langmuir, Vol. 22, No. 18, 2006 7935
fraction CB sensors often afforded greater signal-to-noise ratios as compared to high-mass-fraction CB analogues given ideal signal conversion electronics, less-capable electronics may be unusable with high CB sensors, as digital noise may overwhelm the responses of the sensors in such cases. In these situations, lower-mass-fraction CB sensors may be preferred. Given the additional information they add and new use modes they present, low-mass-fraction CB sensors show promise for use in arrays also containing high-mass-fraction CB sensors. The limitations of low-mass-fraction CB sensors will prevent them from fully replacing high-mass-fraction CB sensors in sensor arrays, but their benefits allow them to complement the function of highmass-fraction CB sensors. Arrays of low-mass-fraction CB sensors also offer unique ways to mimic biological olfaction phenomena such as response saturation. Acknowledgment. We acknowledge the ARO, NIH, HSARPA, and DARPA for support of this work. LA053287S
