NANO LETTERS

Capacitance and Conductance of Single-Walled Carbon Nanotubes in the Presence of Chemical Vapors
Eric S. Snow* and F. Keith Perkins
NaVal Research Laboratory, Washington, D.C. 20375
Received August 22, 2005

2005 Vol. 5, No. 12 2414-2417

ABSTRACT
Simultaneous conductance and capacitance measurements on a single-walled carbon nanotube (SWNT) network are used to extract an intrinsic property of molecular adsorbates. Adsorbates from dilute chemical vapors produce a rapid response in both the capacitance and the conductance of the SWNT network. These responses are caused by a combination of two distinct physiochemical properties of the adsorbates: charge transfer and polarizability. We find that the ratio of the conductance response to the capacitance response is a concentration-independent intrinsic property of a chemical vapor that can assist in its identification.

Introduction. The unique structural and electrical properties of single-walled carbon nanotubes (SWNTs) have inspired researchers to investigate and develop SWNT-based chemical and biological sensors. Initial work in this area has shown that the conductance of SWNTs changes in response to the presence of molecular adsorbates on the SWNT surface.1,2 Subsequently, such SWNT-based chemiresistors have been used to detect a wide variety of gases and chemical vapors.3-8 In addition, by using random networks of SWNTs as the active sensing material, such sensors can be fabricated with high yield using conventional microlithographic processing.8,9 More recently, we have demonstrated that the capacitance of such SWNT networks is sensitive to a broad spectrum of molecular adsorbates and that this surface-enhanced capacitance effect provides a new, sensitive transduction mechanism for SWNT-based sensors.10 In this Letter, we report the results of simultaneous capacitance and conductance measurements of SWNT network sensors in the presence of dilute chemical vapors. Previous work has shown that the conductance response, G, is caused by charge transfer from analyte molecules adsorbed on the SWNT surface11,12 and that such adsorbates also form a polarizable layer that increases the SWNT capacitance.10 Here we report that this capacitance response, C, also contains an additional contribution from changes to the SWNT quantum capacitance, CQ, which is small for most vapors. However, for analytes such as ammonia that desorb slowly and produce a large charge transfer, we find that CQ can dominate C. Additionally, we find that at low to moderate vapor concentrations the G/C ratio is a constant, independent of concentration, and that this ratio varies from
* Corresponding author. E-mail: snow@bloch.nrl.navy.mil. 10.1021/nl051669c CCC: $30.25 Published on Web 10/26/2005 2005 American Chemical Society

analyte to analyte. Consequently, G/C provides a concentration-independent intrinsic molecular signature that can assist in the identification of an unknown analyte. Experimental Section. The SWNT sensors were fabricated by using chemical vapor deposition to grow a SWNT network on a 30-nm-thick thermal oxide on a 0.001 cm silicon substrate.9 For each sensor, a 2 2 mm2 interdigitated array of Ti/Au electrodes was deposited on top of the SWNT network by using photolithography and lift-off. The region inside the array was protected by photoresist, and the unprotected SWNTs were removed by an oxygen plasma etch. The photoresist was then removed, exposing the SWNT network to the ambient environment. A schematic of the device is shown in Figure 1. The sensor geometry is equivalent to that of a large-area transistor with an effective channel length of 0.34 mm and a channel width of 10 mm. The interdigitated electrodes provide contacts for the measurement of the conductance of the SWNT network, and the substrate bias, VS, serves as a gate voltage that can be used to accumulate or deplete the network of charge. For the capacitance, the SWNT network forms an array of nanoscale electrodes that serves as one plate of the capacitor with the other electrode formed by the silicon substrate. We measured the capacitance by applying a 30 kHz, 0.1 V ac voltage between the substrate and top electrodes and detecting the out-of-phase ac current with a lock-in amplifier. We measured the conductance by applying a 311 Hz, 0.1 V ac voltage between the interdigitated electrodes and measuring the in-phase ac current with a second lock-in amplifier. An additional resistor (capacitor) was added in series (parallel) to the SWNT network to

Figure 1. Schematic of the device, which was designed for simultaneous measurement of the SWNT network capacitance and conductance.

improve the data collection efficiency. Using this approach, we could measure the capacitance and conductance simultaneously and independently. The isolation of the two measurements was tested by changing the ac voltage for the capacitance (conductance) measurement while monitoring changes in the conductance (capacitance). We observed no cross talk under the measurement conditions. The device size and interdigitation were chosen to reduce the level of 1/f noise, which is minimized by using largearea, low resistance devices.13 These large-area SWNT networks also facilitate accurate measurements of the SWNT capacitance by producing a device capacitance (1 nF) that is much larger than the capacitance of an individual SWNT (20 aF). Using these devices, we can detect relative changes of C/C and G/G < 10-4 reliably. Dilute chemical vapors were delivered to the sensors by bubbling a low flow of dry air through the liquid analyte and mixing this saturated vapor with a high flow of dry air. The exception to this was ammonia where the air was bubbled through a 30% aqueous solution of ammonium hydroxide. The vapor pressure, P, reported in the data is listed as a fraction of the saturated vapor pressure, P0. Results and Discussion. Before exploring the response to chemical vapors, it is instructive to examine the electrical properties of the SWNT network. The network consists of a combination of electrically interconnected metallic and semiconducting SWNTs and behaves similar to a semiconducting thin film in many ways.9 The device capacitance can be modeled as two capacitors in series: the gate capacitance, CG, which depends on the oxide thickness and the density of SWNTs, and the quantum capacitance, CQ, which is a function of the Fermi energy EF in the SWNTs.14 Thus, C ) (1/CG + 1/CQ)-1. At zero temperature, CQ ) e2g(EF) where g(E) is the SWNT network density of states (for finite temperature thermal broadening effects have to be taken into account to calculate CQ15). For the present device geometry and for EF positioned in the valence band of the semiconducting SWNTs, we estimate that CQ is 10 CG.15,16 In this case, CQ makes only a small contribution to the overall device capacitance, and C CG. However, as the semiconducting SWNTs are depleted of charge (e.g., by an applied gate voltage) the Fermi energy of the network shifts into the band gap of the semiconducting SWNTs where
Nano Lett., Vol. 5, No. 12, 2005

Figure 2. Dependence of the network capacitance (red) and conductance (green) on the substrate voltage, VS. The network capacitance is approximately 1/4 the value of the capacitance for a parallel-plate capacitor with an equivalent area and oxide thickness.

there is a lower overall network density of states. In this case, CQ can be reduced so that it becomes a significant contribution to the total device capacitance. To observe this effect experimentally, in Figure 2 we plot the network capacitance and conductance as a function of VS, where the substrate bias serves as a gate to accumulate and deplete charge in the network. In this experiment, CG remains fixed and the bias-dependence of the capacitance arises from variations in CQ. We find that the conductance decreases with increasing VS as the bias depletes the p-type SWNTs of holes. Likewise, the device capacitance also drops as the substrate bias shifts the Fermi energy into a region with a lower density of states, thereby reducing CQ. As the bias depletes the network of charge, CQ becomes a larger contribution of the total capacitance. For negative gate bias, the SWNTs accumulate additional holes causing the conductance to rise. In this voltage range, C is relatively constant because CQ is already large in comparison to CG and because the density of states does not vary as rapidly in this energy range (i.e., the valence band of the semiconducting SWNTs). We can use the above information to better understand the charge-transfer effects of chemical vapors. In Figure 3 we plot simultaneous measurements of C and G of a device in response to repeated 20-s doses of acetone (P ) 0.01P0) and ammonia (P ) 0.0002P0). For reasons that will become apparent later, we plot the normalized responses C (G) in units of VS that represent the change in substrate voltage required to produce an equivalent change in C (G), that is, C ) C/(dC/dVS) and C ) G/(dG/dVS) where the derivatives are evaluated at VS ) 0. The response observed for acetone is typical of most of the vapors that we have tested. Most vapors weakly physisorb on the SWNT surface and quickly desorb once the ambient vapor is removed. The rapid response and recovery of C and G upon delivery and removal of the analyte is evidence of this rapid desorption. Such weakly interacting vapors increase the measured device capacitance (negative C) while the polarity of G and the relative amplitude, G/ C, vary from analyte to analyte. For most vapors, the
2415

Figure 3. Normalized capacitance (red) and conductance (green) response to 20-s doses of (a) acetone (P ) 0.01P0) and (b) ammonia (P ) 0.0002P0).

magnitude of C is 10 to 100 times larger than the magnitude of G. Previous work has established that adsorbates on the SWNTs form a polarizable layer that increases the capacitance.10 On the basis of the data in Figure 2, we expect that the capacitance response will also contain an additional contribution from charge effects via the quantum capacitance, that is, C = C/ + C/Q Q where the first term represents the dielectric effects of the absorbate and the second term arises from the charge-transfer response via CQ. In terms of the normalized capacitance, C = C/ + VS where we have used the fact that Q ) CVS. If we assume that the primary effect of charge transfer on the conductance is to change the number of charge carriers in the SWNTs,11,12,17 then G = VS. Consequently, G is equal to the charge transfer response in C, and the G/C ratio represents the relative amount that charge transfer contributes to the total capacitance response. In the case of acetone, we observe that G/C ) 3%. Indeed, for most analytes we find that G/ C< 10%, which indicates that the polarizability of the adsorbate dominates the capacitance response for this device geometry. An exception to the above rule is ammonia, for which changes in the quantum capacitance dominate the capacitance response. The sensor response to doses of ammonia is shown in Figure 3b. Ammonia bonds more strongly to the SWNTs,
2416

Figure 4. (a) Normalized capacitance (red) and conductance (green) response to 20-s doses of DMMP at concentrations ranging from P ) 0.0002P0 to 0.02P0. (b) Magnitude of the capacitance and conductance response as a function of P/P0 where both C and G are normalized to 1 at P/P0 ) 2%.

causing G and C to persist upon removal of the vapor. In addition, ammonia is the only analyte we have tested to produce a capacitance decrease (positive C), and further more G/C 1. This unusual capacitance response is caused by the large binding energy and strong electrondonating property of ammonia, which produces a persistent decrease in the hole concentration of the p-type SWNTs.12 Ammonia adsorbates donate about 0.04 electrons per molecule to the SWNT, which decreases the quantum capacitance.11 In addition, the long desorption time produces an atypically small dielectric response, relative to the large dipole moment of NH3 (1.47 D). Earlier numerical simulations indicate that for many polar adsorbates the main contribution to the adsorbate polarizability is the electric field dependence of the binding energy, which produces a fielddependent population of surface dipoles.10 For ammonia, this dielectric response will be strongly suppressed because the long desorption time prohibits the adsorbate population from responding to the 30 kHz measurement frequency. Consequently, the combination of a weak dielectric response and a large charge transfer effect results in a capacitance response dominated by changes in the quantum capacitance. Note that in this case ideally C should equal G. However, for such
Nano Lett., Vol. 5, No. 12, 2005

Figure 5. Normalized capacitance (red) and conductance (green) response to 20-s doses of DMMP and DMP at concentrations ranging from P ) 0.0004P0 to 0.003P0. Table 1. Values of G/C Measured for Various Chemical Vapors
chemical vapor dinitrotoluene dichloropentane nitrobenzene water hexane toluene benzene 2-propanol acetone tetrahydrofuran DMMP G/C 0.20 0.10 0.080 0.045 0.043 0.025 0.013 -0.027 -0.03 -0.10 -0.12

of two chemically similar organophosphonates, DMMP, (CH3O)2P(O)CH3, and dimethyl phosphite (DMP), (CH3O)2P(O)H. These nerve-agent simulants differ only by the replacement of a methyl group with an H atom. However, note that at all concentrations G/C is consistently about 3 times larger for DMMP than for DMP. Thus, the G/C ratio, which can be measured simultaneously in a single sensor, can be used to distinguish the vapors of these chemically similar molecules. Summary. We have shown that both the capacitance and the conductance of the SWNT network respond to the presence of dilute concentrations of a wide range of chemical vapors. For most vapors, the dielectric effect of the molecular adsorbate dominates the capacitance response, whereas charge transfer from the adsorbate controls the conductance response and also perturbs the capacitance via changes in the SWNT quantum capacitance. We observe that the ratio of the capacitance response to the conductance response is a concentration-independent intrinsic molecular property. SWNT network sensors offer the potential for specific identification of an unknown vapor challenge by combining this ratio with other class-specific chemoselective approaches. Acknowledgment. We gratefully acknowledge the financial support of the Office of Naval Research and the Naval Research Laboratory Nanoscience Institute. References
(1) Kong, J.; Franklin, N. R.; Zhou, C.; Chapline, M. G.; Peng, S.; Cho, K.; Dai, H. Science 2000, 87, 622. (2) Collins, P. G.; Bradley, K.; Ishigami, M.; Zettl, A. Science 2000, 287, 1801. (3) Bekyarova, E.; Davis, M.; Burch, T.; Itkis, M. E.; Zhao, B.; Sunshine, S.; Haddon, R. C. J. Phys. Chem. B 2004, 108, 19717. (4) Someya, T.; Small, J.; Kim, P.; Nuckolls, C.; Yardley, J. T. Nano Lett. 2003, 3, 877. (5) Li, J.; Lu, Y.; Ye, Q.; Cinke, M.; Han, J.; Meyyappan, M. Nano Lett. 2003, 3, 929. (6) Goldoni, A.; Larciprete, R.; Petaccia, L.; Lizzit, S. J. Am. Chem. Soc. 2003, 125, 11329. (7) Qi, P.; Vermesh, O.; Grecu, M.; Javey, A.; Wang, Q.; Dai, H.; Peng, S.; Cho, K. Nano Lett. 2003, 3, 347. (8) Novak, J. P.; Snow, E. S.; Houser, E. J.; Park, D.; Stepnowoski, J. L.; McGill, R. A. Appl. Phys. Lett. 2003, 83, 4026. (9) Snow, E. S.; Novak, J. P.; Campbell, P. M.; Park, D. Appl. Phys. Lett. 2003, 82, 2145. (10) Snow, E. S.; Perkins, F. K.; Houser E. H. Science 2005, 307, 1942. (11) Bradley, K.; Christophe, J.-C.; Gabriel, P.; Briman, M.; Star, A.; Gruner, G. Phys. ReV. Lett. 2003, 91, 218301. (12) Kong, J.; Dai, H. J. Phys. Chem. B 2001, 105, 2890. (13) Snow, E. S.; Novak, J. P.; Lay, M. D.; Perkins, F. K. Appl. Phys. Lett. 2004, 85, 4172. (14) Rosenblatt, S.; Yaish, Y.; Park, J.; Gore, J.; Sazonova, V.; McEuen, P. L. Nano Lett. 2002, 2, 869. (15) Guo, J.; Goasguen, S.; Lundstrom, M.; Datta, S. Appl. Phys. Lett. 2002, 81, 1486. (16) Snow, E. S.; Campbell, P. M.; Ancona, M. G.; Novak, J. P. Appl. Phys. Lett. 2005, 86, 033105. (17) The adsorbates may also perturb the conductance by affecting the carrier mobility and/or contact resistance, although we have not observed any evidence that would suggest these factors play a significant role here. (18) This statement is valid only while the charge-induced VS is in a linear region of C(VS) and G(VS). At higher voltage shifts, the nonlinear effects would have to be taken into account to normalize G and C correctly. In addition, the ratio may also change at high concentrations if the adsorbates form multilayers, which would have different effects on the capacitance and the conductance.

large values of VS the simple normalization procedure is inaccurate and should be adjusted to take into account the nonlinearities in C(VS) and G(VS). Having better understood the sensor response, it is instructive to examine the concentration dependence of C and G. Responses to varying concentrations of the nerve agent simulant dimethylmethylphosphonate (DMMP) are shown in Figure 4a. In Figure 4b, the magnitudes of C and G are plotted as functions of the DMMP concentration, which was varied from 0.0002P0 to 0.02P0. In this Figure, the data were normalized to 1 at the highest concentration of 0.02 P0. At these moderately high levels, saturation effects are observed as C and G display a sublinear dependence on concentration, which is an indication that the number of available adsorption sites is limiting the rate of adsorption. Note, however, that over both the linear and the sublinear regions, the ratio of G/C is independent of concentration. Thus, this ratio provides a concentration-independent quantity that depends on the relative strength of the adsorbate charge transfer and polarizability effects.18 We list in Table 1 values of G/C for a few representa tive chemical vapors. As noted above, G is typically a few percent of C, and the ratio takes both positive and negative values. To demonstrate the utility of this quantity, we plot in Figure 5 C and G measured in response to 20-s doses
Nano Lett., Vol. 5, No. 12, 2005

NL051669C
2417