Roles of inter-SWCNT junctions in resistive humidity response

SWCNT networks are grown on quartz substrate in a thermal-chemical vapor deposition (CVD) reactor at 925 °C using ferritin as the catalyst precursor [11]. The morphologies of as-grown SWCNTs are first characterized by scanning electron microscopy (SEM). Figures 1(b) and (c) show the as-grown SWCNT network with low density and high density of SWCNTs, respectively. The density of the SWCNT network is estimated by counting the number of SWCNTs per unit area in atomic force microscopy (AFM) images. The densities of the as-grown SWCNT networks used in this work are between ∼1 μm⁻² and ∼13 μm⁻². SWCNT networks with nanotube density less than ∼1 μm⁻² were generally found to have higher electrical noise, which was probably due to the poor percolation of SWCNTs.

Zoom In Zoom Out Reset image size

The properties of the as-grown SWCNT networks are then characterized using Raman spectroscopy as well as AFM. Similar to our previous results [11], a mixture of metallic and semiconducting SWCNTs with a diameter distribution of ∼1.0–2.2 nm is confirmed by Raman scattering with the incident laser wavelength of 532 nm and 633 nm (figures 2(b) and (c)). AFM measurement, shown in figure 2(a), further confirms the diameter distribution of the as-grown SWCNTs.

Zoom In Zoom Out Reset image size

The electrical conductance of as-grown SWCNT networks is also characterized and their conductance versus density curves are consistent with standard percolation theory with appropriate fitting parameters [12]. The total conductance of an as-grown SWCNT random network could be modeled by the expression $\sigma =k{\left(N-{N}_{C}\right)}^{\alpha }$ [12], where $\sigma$ is the total conductance of the network, $k$ is a proportional constant, $N$ is the volume loading of the nanotubes in mg l⁻¹, ${N}_{C}$ is the critical volume loading of nanotubes (mg l⁻¹) corresponding to the percolation threshold, and $\alpha$ is the fitting exponent. The critical volume loading is determined by the length scale of the nanotubes with the expression ${N}_{C}=1/\pi {\left(4.236/L\right)}^{2}$ mg l⁻¹ [12], where $L$ ${\ }$is the average length of the nanotubes in μm. By assigning the proportional constant $k=1.7\times {10}^{-7},$ the critical fitting exponent $\alpha =1.94$ [12, 13], the effective nanotube length $L=50\;\mu {\rm{m}}$ and the conversion ratio of mass per unit volume (mg l⁻¹) over nanotube loading per unit area (μm⁻²) equals 1, the average conductances measured for the as-grown networks are in good agreement with the standard percolation theory (see figure 3). The standard deviations of the resistances are within 0.5% (see figure 3 inset) when sampling the resistance every 1 s with a constant dc voltage of 1 V applied to the electrodes for 2 min.

Zoom In Zoom Out Reset image size

As illustrated in figure 1(a), humidity devices are fabricated by depositing a pair of Ti/Au at both ends of the SWCNT networks. The contacts between the nanotubes and the electrodes were covered by 300 nm thick silicon nitride to prevent humidity from affecting the contact resistance between the electrodes and nanotubes. The resistive humidity test was conducted in a test chamber, whose humidity levels were varied by controlling the relative flow ratio between dry air and humid air. The normalized resistances of four SWCNT networks with different densities are plotted versus their humidity levels. As shown in figure 4(a), when humidity increased from 10% to 85%, the total resistance of the SWCNT networks first decreased and then increased. Therefore, for each curve of the resistance versus humidity, there existed a 'critical humidity' (as labeled in figure 4(a)) beyond which the SWCNT network resistance started to increase. The 'critical humidity' levels for 12 SWCNT networks are plotted versus their densities in figure 4(b). It can be seen that the 'critical humidity' was between 30% and 45% for SWCNT networks with nanotube loading larger than 6.8 μm⁻². For SWCNT networks with nanotube loading less than 6.8 μm⁻², the 'critical humidity' levels became higher. The influence of SWCNT network density on the 'critical humidity' level is explained in supporting information. In this work, we use 'monotonic' response to describe the case where the SWCNT random network resistance continuously increases or decreases as humidity increases from 10% to 85%. In contrast, a 'non-monotonic' response refers to the case where the resistance does not change in one direction as humidity increases from 10% to 85%. Due to the non-monotonic resistance versus humidity, as-grown SWCNT networks were therefore not suitable for humidity sensing applications.

Zoom In Zoom Out Reset image size

In order to understand the non-monotonic resistive humidity responses of the above SWCNT networks, it is important for us to study the roles of the nanotubes themselves and their junctions. Since a random SWCNT network consists of SWCNTs with a large variety of chiralities, it is difficult to identify the role of a particular SWCNT. However, it is possible for us to determine the collective humidity response of SWCNTs in a horizontally aligned array in which the roles of inter-tube junctions should not be considered. As shown in figure 5(a), the humidity sensing device was fabricated on a quartz substrate with a horizontally aligned SWCNT array, and no inter-tube junction could be observed under SEM. The resistance of the SWCNTs parallel to the conduction channel decreased monotonically as humidity increased from 10% to 85% and no 'critical humidity' was observed, which is in sharp contrast to SWCNT random networks. As the horizontally aligned SWCNTs grew under the same conditions as those random SWCNT networks, they had very similar SWCNT chirality distributions, as confirmed by Raman scattering. Thus, the only difference in the humidity responses must originate from the inter-tube junctions. As a result, the increase of SWCNT network resistance beyond the 'critical humidity' must be caused by the inter-tube junctions.

Zoom In Zoom Out Reset image size

To further verify the hypothesis, two horizontally aligned SWCNT array devices with low and high degrees of crosslinking SWCNTs were also compared with respect to their humidity responses. As shown in figures 5(b) and (c), the 'critical humidity' appeared on both devices due to the crosslinking. In addition, the high crosslinking area circled in figure 5(c) showed a much larger resistance increase and a much lower 'critical humidity'. Therefore, we could suggest that, as humidity increases, the overall resistive response of an as-grown SWCNT network is actually a competition between two mechanisms, i.e. the resistance drop caused by all the SWCNTs and the resistance increase due to all the inter-tube junctions. The 'critical humidity' thus represents the relative humidity at which the two mechanisms are balanced.

The finding that inter-tube junction resistance increases with increasing humidity is consistent with previous theoretical predictions in drop-casted SWCNT films, where the origin was explained through the suppression of inter-tube carrier hopping by water molecules [5]. As humidity increases, the average separation between SWCNTs at the tube–tube contact will also be increased so that the thermally activated carrier hopping at inter-tube junctions is suppressed, resulting in an increase in the inter-tube resistance [5]. However, the resistive humidity response of a horizontally aligned SWCNTs array without inter-tube junctions has been little studied. It is well understood that the resistance decrease of an SWCNT could be achieved by either increasing carrier densities or reducing carrier scatterings. Carrier concentration could be increased through electron doping from water molecules to semiconducting SWCNTs [14–17]. However, the doping mechanism is not sufficient to explain the monotonic drop of SWCNT array resistance over the entire humidity range. This is because the electrons transferred from water molecules to semiconducting SWCNTs would first decrease hole concentration and then increase electron concentration due to the inherent p-type nature of semiconducting SWCNTs in air [4]. Therefore, in order to explain the monotonic drop of SWCNT resistance, there must be other mechanism(s) in which the carrier scattering in SWCNTs could be reduced by water molecules. Actually, there have been studies [18–21] suggesting that the response of SWCNTs to chemical vapors is mainly contributed through the defect sites that are typically low energy adsorption centers, and the trapped molecules could contribute to the electrical conductance via the Poole–Frenkel (PF) conduction. Thus, in the presence of defects, water molecules could easily attach to the SWCNTs and decrease the resistances of both metallic and semiconducting SWCNTs via the PF conduction.

In order to use as-grown SWCNT random networks as a humidity sensor element, the influences from the inter-tube junctions must be eliminated. For this purpose, the SWCNTs are modified through covalent functionalization. This has two apparent effects on the sensing behaviors of the SWCNT networks. On one hand, it could create additional defects on SWCNTs and increase their resistance by a few orders of magnitude so that the total resistance is largely limited by the SWCNTs rather than the inter-tube junctions. Thus, the resistance changes of the SWCNTs by adsorbing water molecules dominate the overall resistive response of the SWCNT network so that the detection range of the humidity sensor could be largely extended. On the other hand, the created defects could enhance the hydrophilicity of the SWCNTs and, at the same time, increase the influences via the PF conduction mechanism [18–21].

An SWCNT random network with nanotube density of ∼6.8 μm⁻² was subjected to low-degree and high-degree covalent functionalizations by soaking the sample in a 10 μM four-bromobenzene diazonium tetrafluoroborate (4-BBDT) solution at room temperature for 1 min and 5 min, respectively [22]. The device was subsequently washed thoroughly in deionized water and annealed at 120 °C in air for 2 min to fully remove the chemical residual. The SWCNTs were successfully functionalized as manifested by the Raman D band (∼1340 cm⁻¹) to G band (∼1580 cm⁻¹) ratio (I_D/I_G) increase as shown in figures 6(c) and (e). (The Raman spectra are calculated by averaging all the 100 × 100 points in a Raman mapping on a 10 μm × 10 μm square.) figure 6(c) shows a higher I_D/I_G than figure 6(b), indicating that higher concentration defects are created under a longer reaction time.

Zoom In Zoom Out Reset image size

As seen in figures 6(b), (d) and (f), the 'critical humidity' level shifted from 30% to 45% after 1 min functionalization and further shifted to 75% after 5 min functionalization. This finding suggests that at a higher defects concentration, the humidity response originating from the SWCNTs dominates over the humidity response from the inter-tube junctions due to the significant increase in SWCNT resistance. As shown in figure 2, the resistance is increased by more than two orders after 5 min of the functionalization. Therefore, the total resistance of the SWCNT network is now significantly limited by the functionalized SWCNTs, and the change in the SWCNT resistance should also have a much larger effect on the overall resistance. As a result, SWCNT resistance becomes more strongly dominant over the inter-tube junctions and thus shifts the 'critical humidity' towards a higher value. In addition, the addend groups also significantly increase the hydrophilicity of the SWCNTs, resulting in a huge increase of sensitivity from less than 5% to more than 25%. The shift of 'critical humidity' and increase in sensitivity by 4-BBDT functionalization is also observed in SWCNT networks with lower and higher densities as shown in supporting information figure S3.

The sensor responses of the SWCNT network at different defect concentrations are also compared with a relative humidity between 10% and 30%, and between 45% and 75% at time intervals of 30 s. As seen in figure 7(a), the sensitivity of the SWCNT network subject to 10% and 30% humidity is much increased by the functionalization. In figure 7(b), it is shown that the pristine network had its resistance slightly increased when the relative humidity was increased from 45% to 75%. In contrast, after the functionalization, the SWCNT network had its resistance decreased when the relative humidity was increased from 45% to 75%. In addition, the sensitivity of the SWCNT network was also increased by the functionalization due to the increased hydrophilicity of the SWCNTs by attaching covalent molecules.

Zoom In Zoom Out Reset image size

It has been noticed that significant drifts of response curve baseline were observed in both figures 7(a) and (b). In figure 7(a), there is a down-drift of the resistance as humidity switches between 10% and 30%. Referring to figure 6, the resistance of the pristine, 1 min and 5 min functionalized samples tends to decrease when humidity is increased from 10% to 30%. However, the resistance values shown in figure 6 are taken at stabilized states, and it requires about 20 min for the resistance value to be stabilized after humidity is switched from 10% to 30%. In contrast, the humidity switching in figure 7(a) is carried out in 30 s short intervals, so that the unsaturated adsorption (when humidity switched from 10% to 30% humidity) and incomplete desorption (when humidity switched from 30% back to 10%) of water molecules to and from the SWCNTs are responsible for the down-drift.

In figure 7(b), the humidity is switched between 45% and 75%, and there is an up-drift for the pristine and 1 min functionalized samples, but a down-drift of the 5 min functionalized sample. Referring to the response curves in figures 6(b) and (d), for the pristine and 1 min functionalized samples, when humidity is switched from 45% to 75%, their total resistance will increase due to the inter-tube junctions. Therefore, the up-drift of the pristine and 1 min functionalized samples in figure 7(b) is due to the unsaturated adsorption and incomplete desorption of water molecules at the inter-tube junctions in the 30 s short intervals. In contrast, as shown in figure 6(f), for the 5 min functionalized sample, when humidity is switched from 45% to 75%, the total resistance decreases due to the SWCNT resistance drop. Therefore, the down-drift of the 5 min functionalized sample in figure 7(b) is due to the unsaturated adsorption and incomplete desorption of water molecules to the SWCNTs in the 30 s short intervals.

It is worth noting the response curve of the 1 min functionalized sample in figure 7(b). The overall trend of the total resistance is an up-drift when humidity is changed from 45% to 75% at 30 s intervals. This overall up-drift trend is consistent with the stabilized resistance curve in figure 6(b) where increasing humidity from 45% causes an increase in resistance. However, the transient resistive response at the switching is opposite to the up-drift trend (i.e. at the moment when humidity is switched from 45% to 75%, there is always a resistance drop rather than increase). Therefore, there must be different mechanisms governing the up-drift trend and the transient resistance decrease when humidity switched from 45% to 75%. As discussed earlier, the resistance decrease of the SWCNT networks is caused by the SWCNTs, while the resistance increase is caused by the inter-tube junctions. Therefore, this abnormal response curve in figure 7(b) further suggests that the overall resistive humidity response of the SWCNT networks is a competition between the SWCNT resistance change and inter-tube junction resistance change. In addition, the opposite transient response also implies that the response from the SWCNTs could be much faster than the responses from the inter-tube junctions.

Thermal CVD growth of SWCNT networks on quartz substrate was conducted in a quartz tube furnace at 925 °C with ferritin as the catalyst precursor and ethanol as the carbon source [16]. The densities of the SWCNT networks were controlled by varying the catalyst densities. Humidity sensors were fabricated by depositing pairs of Ti/Au electrodes onto the SWCNT networks on quartz substrates (∼1 cm × 1 cm) through shadow masks. In order to exclude possible influences of the contacts between the electrodes and SWCNT network on sensing responses, a 300 nm thick silicon nitride passivation layer was deposited onto the contact areas, leaving only the SWCNT network exposed to the environment.

Resistive humidity responses of the SWCNT networks were measured in a sensor chamber where the relative humidity could be changed by changing the relative flow ratio of dry air and humidly air (through a deionized water bubbler). The total flow rate was set to 1000 sccm so that the entire volume of the chamber (∼300 cm³) could be effectively filled by the test gases in less than 1 min. The resistance of the SWCNT network device was sampled every 1 s with a constant 1 V dc voltage applied. The SWCNT network resistance values at different humidity levels were recorded after the resistance was stabilized at each humidity level.

Raman spectroscopy was used to determine the diameter distribution of the pristine SWCNTs as well as to determine the degree of defects in the functionalized SWCNTs. Raman scattering was performed with excitation laser lines of 532 nm and 633 nm using a confocal Raman system in air and calibrated with the silicon peak ∼520 cm⁻¹. Raman mappings of SWCNT networks were performed on random 10 μm × 10 μm squares with 100 × 100 points resolution.