Chemical Detection with a Single-Walled Carbon Nanotube Capacitor

See allHide authors and affiliations

Science  25 Mar 2005:
Vol. 307, Issue 5717, pp. 1942-1945
DOI: 10.1126/science.1109128


We show that the capacitance of single-walled carbon nanotubes (SWNTs) is highly sensitive to a broad class of chemical vapors and that this transduction mechanism can form the basis for a fast, low-power sorption-based chemical sensor. In the presence of a dilute chemical vapor, molecular adsorbates are polarized by the fringing electric fields radiating from the surface of a SWNT electrode, which causes an increase in its capacitance. We use this effect to construct a high-performance chemical sensor by thinly coating the SWNTs with chemoselective materials that provide a large, class-specific gain to the capacitance response. Such SWNT chemicapacitors are fast, highly sensitive, and completely reversible.

Sorption-based microsensors are currently a leading candidate for low-power, compact chemical vapor detection for defense, homeland security, and environmental-monitoring applications (19). Such sensors combine a nonselective transducer with chemoselective materials that serve as a vapor concentrator, resulting in a highly sensitive detector that responds selectively to a particular class of chemical vapor. An array of such sensors, each coated with a different chemoselective material, produces a response fingerprint that can detect and identify an analyte (13). Sorption-based sensors provide sensitive detection for vapors ranging from volatile organic compounds to semivolatile chemical nerve agents, although low–vapor pressure materials such as explosives are challenging because they do not produce a sufficiently high concentration of vapor (4).

The transducer elements for such sensor arrays need to be small, low-power, and compatible with conventional microprocessing technology. Among the choice of transducers are mechanical oscillators that respond to changes in mass (1, 2), chemicapacitors that detect changes in dielectric properties (46), and chemiresistors that monitor the resistance of a polymer laced with conductive particles (79). Of these transducers, chemicapacitors (4) and chemiresistors (7, 8) are the best suited for low-power sensor arrays. Chemiresistors are simple to implement, but instability of the conductive particle/polymer interface can be a disadvantage. Chemicapacitors are more stable but can take minutes to respond and recover (4). This slow response is limited by the time necessary to load and then remove the analyte from the relatively thick layers of chemoselective dielectric (∼1 μm) that are typically used.

We describe a chemicapacitor constructed from single-walled carbon nanotube (SWNT) electrodes that combines the features of stability, high sensitivity to a broad range of analytes, and fast response time. The capacitance response of the SWNT chemicapacitor is dominated by surface adsorbates, which allows us to use very thin layers of chemoselective material down to, and including, a single molecular monolayer. By achieving chemical selectivity with such a monolayer, we eliminate the time required to load and refresh a thick, chemoselective dielectric and can perform sensitive, real-time sensing.

The surface capacitance effect is caused by the large electric-field gradient radiating from the ∼1-nm-diameter SWNT electrodes. This transduction mechanism is quite general and can be used to detect both volatile organics and low–vapor pressure explosives. We demonstrate the compatibility of this transducer with conventional chemoselective polymers by using a hydrogen-bonding polymer to achieve a minimum detectable level (MDL) of 0.5 parts per billion (ppb) for dimethylmethylphosphonate (DMMP), a simulant for the chemical nerve agent sarin.

To improve the response time, we replaced the polymer layer with a hydrogen-bonding molecular monolayer. In this case, we achieve a MDL of 50 ppb for DMMP with a 90% recovery time, t90, < 4 s. By combining 1-nm-diameter electrodes with molecular-scale functionalization, we achieve a sorption-based chemicapacitor that offers stability, real-time sensing, and a high sensitivity to a wide spectrum of chemical vapors ranging from volatile organics to low–vapor pressure explosives.

We fabricated the SWNT chemicapacitors by using chemical vapor deposition to grow a SWNT network on a 250-nm-thick thermal oxide on a degeneratively doped silicon substrate (10). For each sensor, a 2 mm by 2 mm interdigitated array of Pd electrodes was deposited on top of the SWNT network by using photolithography and lift-off. The interdigitated electrodes provide contacts for the simultaneous measurement of both the capacitance and the resistance of the SWNT network. The region inside the array was protected by photoresist, and the unprotected SWNTs were removed from the substrate by a CO2 snowjet. The photoresist was then removed, which left the SWNT network exposed to the ambient environment. We prepared the chemical vapors by mixing saturated vapors of the analyte with dry air at 25°C.

The SWNT network forms an array of nanoscale electrodes that serves as one plate of the capacitor, with the other electrode formed by the heavily doped Si substrate (Fig. 1). We measured the capacitance by applying a 30-kHz, 0.1-V ac voltage between the SWNTs and the substrate and detecting the out-of-phase ac current with a lock-in amplifier. The measured capacitance, ∼ 10 nF/cm2, is close to the parallel-plate value corresponding to a 250-nm-thick SiO2 gate dielectric and is the expected value for an inter-SWNT spacing less than the SiO2 thickness (11).

Fig. 1.

Optical micrograph of a SWNT chemicapacitor. The region between the electrodes is covered with an optically transparent but electrically continuous network of SWNTs (shown in the inset atomic force microscope image). The capacitance was measured by applying an ac bias between this top surface and the underlying conducting Si substrate. The electrodes were interdigitated to allow simultaneous measurement of the network resistance, but were electrically shorted for data collected here.

Under an applied bias, fringing electric fields (∼105 to 106 V/cm, for a 0.1-V bias) radiate outward from the SWNTs. These fringing fields are strongest at the SWNT surface and produce a net polarization of the adsorbates that we detect as an increase in capacitance. The relative capacitance change, ΔC/C, of one such device in response to repeated 20-s doses of N,N-dimethylformamide (DMF) at varying vapor concentrations (Fig. 2) shows that the observed response is rapid (limited by the 4-s response time of our vapor-delivery system), proportional to the analyte concentration, and completely reversible. Of the chemical vapors that we have tested (Table 1), we observe a similar, rapid capacitance response that is completely reversible upon removal of the vapor. We also note that ΔC is independent of the applied voltage for Vac < 1 V, which indicates that the polarization is a linear function of the electric field.

Fig. 2.

Measured relative capacitance change, ΔC/C, of a SWNT chemicapacitor in response to repeated 20-s doses of dimethyl formamide (DMF) at varying concentrations noted in the figure.

Table 1.

Capacitance response to various chemical vapors. Listed are the measured values of ΔC/C corresponding to P/P0 = 1%. Also listed are the values of the dipole moment, μ, the equilibrium vapor pressure, P0, at 25°C, and the vapor concentration, P, in parts per million.

Chemical vapor P0 (mbar) at 25°C P (ppm) at 1% P/P0 μ (D) ΔC/C × 10-3 at 1% P/P0
Benzene 127 1290 0 0.3 ± 0.1
Hexane 200 2030 0 0.4
Heptane 61 618 0 0.2
Toluene 38 385 0.38 0.5
Trichloroethylene 91 922 0.8 0.6
Chloroform 257 2600 1.04 0.8
Trichloroethane 38 385 1.4 0.8
Isopropyl alcohol 108 1093 1.58 3.8
Ethanol 78 792 1.69 3.0
Chlorobenzene 16 162 1.69 0.4
Methyl alcohol 168 1702 1.7 2.7
Tetrahydrofuran 215 2180 1.75 5.9
Ethyl acetate 127 1290 1.78 3.1
Water 32 324 1.85 0.5
Dichlorobenzene 2 20.3 2.5 0.4
Acetone 304 3080 2.88 6.1
Dimethylmethylphosphonate 1.6 16.2 3.62 10.2
N,N-dimethylformamide 5 50.7 3.82 9.3
Dinitrotoluene 0.0028 0.028 4.39 0.5

In Table 1, we list values of ΔC/C for a number of chemical vapors, each measured at a fixed fraction, P/P0 = 1%, of the equilibrium vapor pressure P0. We also list literature values of P0 (12), the vapor concentration, P, in parts per million (ppm) at P/P0 = 1%, and the molecular dipole moment, μ (12, 13). In Fig. 3, we plot the values of ΔC/C reported in Table 1 for each of the analytes versus their respective dipole moments.

Fig. 3.

Measured capacitance response to P/P0 = 1% doses of various chemical vapors plotted as a function of their molecular dipole moment. The capacitance response generally increases with dipole moment; however, large deviations from this trend are observed.

In Fig. 3, we observe that for several analytes, the magnitude of the capacitance response correlates with the value of its dipole moment. Nonpolar molecules such as hexane and benzene produce a small response, whereas relatively polar molecules like DMMP and DMF produce a large capacitance response. This correlation with dipole moment holds under the condition that the vapors are each delivered at a constant value of P/P0, and not for a constant value of P. For example, acetone (μ = 2.88 D) and DMMP (μ = 3.62 D) produce a comparable capacitance response when both are delivered at P/P0 = 1% even though this condition corresponds to vapor concentrations of 3080 ppm and 16 ppm, respectively. Several analytes such as chlorobenzene, 1,2-dichlorobenzene, 2,4-dinitrotoluene, and water (represented by squares in Fig. 3) produce a small capacitance response even though they have a relatively large dipole moment. These data indicate that the magnitude of the capacitance response is strongly modified by surface interactions.

The polarizability of a free vapor molecule is given by Embedded Image (14), where the first term arises from the intrinsic molecular polarizability, γmol, and the second term arises from the field-induced alignment of the otherwise randomly oriented molecular dipole moment. From the Clausius-Mossotti equation, this polarizability is related to the dielectric constant, ϵ, by Embedded Image where N is the number of molecules, which is proportional to the vapor pressure, P. Thus, for a dilute vapor, the capacitance response should scale as Pμ2. Instead, we observe that Embedded Image and that there are large deviations from simple μ2 behavior.

For surface adsorbates, the polarization will be proportional to the number of the adsorbates, which is proportional to Embedded Image (15), where Ei is the analyte mutual interaction energy and Eb is the binding energy to the SWNT surface (approximately the thermal energy, kT, for physisorbed molecules). Thus, for adsorbates, we expect the capacitance to scale as P/P0 with significant analyte-to-analyte variations caused by the differences in binding and interaction energies.

Additionally, surface interactions will preferentially orient the molecular dipole moment. For example, the low response of chlorobenzene, 1,2-dichlorobenzene, and 2,4-dinitrotoluene can be understood in the context of surface interactions. These analytes indicate that the capacitance response scales with the component of the dipole moment oriented perpendicular to the SWNT surface. Each of these molecules has a dipole moment that is oriented in the plane of an aromatic ring. Our density functional calculations (16) indicate that the lowest energy configuration corresponds to the ring lying flat on the SWNT surface. In this orientation, the dipole moment lies perpendicular to the radial electric field, which minimizes the polarization and results in a small capacitance effect. The cause of the small water response is not clear. However, it has been suggested that the dipole moment of water also aligns tangentially to the SWNT surface in its lowest energy configuration (17).

Our initial density functional calculations (16) indicate that for some analytes such as acetone the primary polarization effect derives from the field dependence of the binding energy, which causes a change in the number of adsorbates by a factor ∼ΔEb/kT. However, further study is needed to understand the precise polarization mechanism, which will differ for different analytes.

The rapid, completely reversible capacitance response that we observe contrasts with the behavior of SWNT chemiresistors (1825). SWNT chemiresistors respond to a narrower range of analytes, and typically the resistance recovers very slowly after exposure. Part of the reason for this difference is that the SWNT chemiresistors detect charge transfer from analytes, whereas the SWNT chemicapacitors operate via a different transduction mechanism, the polarization of surface adsorbates (26). Our experience measuring both effects simultaneously on the same device indicates that, for most vapors, the capacitance response is more sensitive, recovers much faster, and applies to a broader range of analytes (27).

Because most chemical vapors, ranging from volatile organics such as acetone to low–vapor pressure solids such as 2,4-dinitrotoluene, produce an easily measured capacitance response, this transduction mechanism can be used to detect a broad spectrum of molecular analytes. To explore this possibility, we coated our sensors with a chemoselective polymer, HC, that we designed for preferential absorption of chemical nerve agents. HC is an acidic, strong–hydrogen-bonding polycarbosilane (25). We coated a sensor with a thin layer (∼100 nm) of HC and tested the response to several analytes. The response of this polymer-coated sensor to repeated 10-s doses of acetone ranging from 60 to 540 ppm is shown in Fig. 4. The acetone produces a large, rapid response that is ∼100 times larger than the response measured in the same sensor before the HC deposition. The HC concentrates the acetone vapor in the vicinity of the SWNTs, which increases the response while maintaining a rapid response time. The response to a single 200-s dose of DMMP delivered at 320 ppb shows that the measured gain for DMMP relative to the uncoated sensor is about 500 (Fig. 4). Note that the low diffusion rate of DMMP in the HC causes a slower recovery rate, t90 = 370 s. For water and chloroform, the polymer coating produces much lower response gains of 1 and 10, respectively. Thus, the HC provides a large chemically selective gain, demonstrating the feasibility of SWNT sorption-based chemical sensing.

Fig. 4.

Blue curve: Response to 10-s doses of acetone of a SWNT chemicapacitor coated with the polymer, HC. The concentration was set at 60, 180, and 540 ppm. Green curve: Response of the same HC-coated sensor to a single, 200-s dose of DMMP. Note the increased response time caused by the slower diffusion of the DMMP. Red curve: Response (×10) of a SAM-coated sensor to 10-s doses of DMMP. The concentration was set at 320 ppb, 960 ppb, and 2.9 ppm. Note the improvement in response time relative to the HC-coated sensor.

These sensor characteristics compare favorably with those of commercial chemicapacitors. Using a signal-to-noise ratio of 3:1 as a detection criterion, we estimate that MDL = 0.5 ppm and t90 < 4 s for acetone and MDL = 0.5 ppb and t90 = 370 s for DMMP. For these same analytes, the commercial sensor achieves a MDL = 2 ppm and t90 = 228 s for acetone and MDL = 2 ppb and t90 = 3084 s for DMMP (4). We attribute our faster response and recovery times to the use of a much thinner layer of chemoselective material. For HC, the minimum layer thickness was limited by the tendency of the HC to form a discontinuous film below ∼100 nm.

Our initial polymer-coated SWNT sensors achieve both higher sensitivity and faster response times than do current chemicapacitors. However, both of these properties can be substantially improved with a few design modifications. The sensitivity is currently limited by the small series capacitance of the thick SiO2 layer. By thinning the SiO2 layer or replacing it with a high–dielectric constant insulator (28), we estimate that we can increase the series capacitance by about a factor of 10, which should produce a comparable increase in response.

The response time for analytes such as DMMP is limited by diffusion through the layer of HC. Because the SWNT capacitor is based on a surface effect, we can improve the response time and still achieve chemical gain by using extremely thin layers of chemoselective material down to, and including, a single molecular monolayer.

To explore this limit of a chemoselective monolayer, we coated the SiO2 surface with a self-assembled monolayer (SAM) of allyltrichlorosilane. We then reacted the terminal alkenes with hexafluoroacetone to produce a monolayer of hexafluoroisopropanol that partially covers the SWNTs with fluoroalchohol groups. The response of this SAM-coated sensor to repeated 10-s doses of DMMP ranging from 320 ppb to 2.9 ppm is shown in Fig. 4. For this sensor, the response tracks our vapor-delivery system, indicating that t90 < 4 s, and we measured a MDL = 50 ppb. Notably, with the SAM coating, the capacitance response of DMMP relative to that of water is increased by a factor of 40, indicating that we achieved substantial chemically selective gain. These initial promising results indicate that optimization of the chemoselective monolayers to better cover the SWNTs, combined with improved sensor design, can result in a new class of sorption-based sensors that combine the features of low power, high sensitivity, and fast response time.

Supporting Online Material

SOM Text

Fig. S1

References and Notes

View Abstract

Navigate This Article