Introduction

The advent of nanotechnology opens up the potential for miniaturizing electronic components thereby paving the way for the fabrication of portable, flexible yet high-performance electronic devices that has many applications such as in electric or hybrid vehicles, printed electronics, and also in the aerospace industry. However, one of the major challenges is in fabricating suitable power sources that are compatible with these devices. Therefore, state-of-the-art energy storage devices, such as highly efficient batteries and capacitors, are the focus of substantial research efforts. Conventional supercapacitors are characterized by their high power density but have a relatively low energy density. This is because they function solely through charging the surface of the electrode forming an electric double layer (EDL). Therefore, electrodes based on conducting nanomaterials having higher surface area in contact with the electrolyte would result in higher capacitance. Another type of supercapacitors, termed pseudocapacitors, combines the EDL effect with charge stored through reversible redox reactions occurring at the surface of the electrodes. This is usually achieved by combining redox materials, e.g., conductive polymers [1–5] and transition metal oxides [6–10], with the conducting electrodes. Pseudocapacitors have attracted much attention due to their ability to combine the charge storage capability of batteries and the high power density of conventional capacitors.

The discovery of carbon nanotubes (CNTs) provided the impetus for current advancements in various fields including charge storage technologies [11, 12]. Specifically, single-walled CNTs (SWCNTs) exhibit very high theoretical specific surface area of 1315 m2/g [13] as compared with commercially available activated carbon (660 m2/g [14]), making it a conductive material that is highly desired in supercapacitors. CNTs are also known for high tensile strength [15], which makes them viable as a material for flexible integrated electronics.

CNTs have been extensively used for assembling supercapacitors. This area has been reviewed by Bose [16] and Pan [17], which includes a wide variety of materials combined with different types of CNTs, i.e., SWCNTs and multi-walled CNTs (MWCNTs). However, the specific capacitance reported of similar systems span over a wide range of values, which indicates of the uncertainty of measurement that is mostly due to the error in weighing the active materials as regular lab balances do not have high enough resolution. At the same time, for practical applications, the areal capacitance, which is often neglected in the reported data, should also be considered.

Among the metal oxides to be coupled with CNTs, RuO2 and MnO2 are considered to be the most promising. The former has fairly high conductivity, redox reversibility, and stability in various electrolytes [18–22], and the latter has low cost and relatively comparable properties. In a recent study, RuO2 and MnO2 were deposited on carbon fabric from Ru(VII) and Mn(VII) solutions, respectively [18]. The deposition was accomplished by soaking in solutions of KRuO4 and KMnO4 for several hours. The morphology of the resultant oxide films showed very little cracks, especially for the MnO2 samples. This report also showed that RuO2 and MnO2 possess high capacities of ca. 824 F/g in 1 M H2SO4 and 1080 F/g in 2 M LiOH. Furthermore, their ability to be coupled with carbon fabric reveals their possible compatibility with CNTs. However, the total capacitance of this device depended mostly on the reversible redox reactions of the oxides as it covers most parts of the carbon fibers therefore resulting in a device with lower power capabilities on EDL as a charge storage mechanism. Reviewing studies like these reveals that the method of electrode assembly, namely, the approach for mixing CNTs and metal oxides, is still to be optimized. Metal oxides are usually deposited in the CNT network without morphological control in the form of nanoparticles [19, 20, 23], bulk electrochemically deposited films [7, 24, 25], or simply by mixing or forming the metal oxide nanoparticles in the presence of the CNTs [10, 26, 27]. The interfacial assembly and properties of the intimate contact between the two components, i.e., the conductive and the redox materials, are of utmost importance and will affect significantly the total capacitance. However, by coupling these oxides with CNTs, high electrical resistance is introduced and this impedes the charge transfer from the oxides to the CNTs and the current collectors [27]. More importantly, the uncontrolled deposition of the CNTs and the redox materials also reduces the area in which the active material is in contact with the electrolyte. Some strategies explored to mitigate this problem would be the use of electrophoresis [28, 29], dip coating [30], and air-spraying to deposit the active materials [5]. However, most of these techniques are only able to deposit the active material on the surface of the current collector, and thereby, the effective surface area that is available for charge collection is limited to the size of the current collector, and thickness and the porosity of the film. However, these challenges are not reflected if only the specific capacitance values are reported as in most cases. Therefore, a three-dimensional porous structure would be necessary to store a practical amount of charge without the need for a large current collector surface area as the bulk of the material can then be utilized as well.

In this work, a method for assembling of CNTs/metal oxides on a glass fiber felt was demonstrated. The latter was used as a flexible, stable 3D support with high surface area, which is superior as compared with most of the conventional flexible materials such as cellulose and polyester. A three-dimensional hierarchical layered structure was fabricated by sequentially depositing CNTs and metal oxides, i.e., RuO2 or MnO2. It was found that the layered hierarchical structure has the potential to enhance the utilization of both the EDL and the pseudocapacitive components as seen in the improvement of the specific and areal capacitance as compared with the bulk architecture.

Experimental

Carboxylic acid-functionalized SWCNTs (P3, 0.5–3 µm in length and 1.4 nm in diameter) were procured from Carbon Solutions, Inc. and used as received. Glass fiber felt (20 µm in diameter of single fiber) was obtained from Hollingsworth and Vose (USA). Ru(III) and Mn(VII) solutions were prepared [18, 31] from RuCl3·xH2O (Aldrich, ReagentPlus®) and KMnO4 (Merck, EMSURE®ACS), respectively. Ru(III) solution consisted of 5 mM RuCl3·xH2O, 0.1 M NH4Cl, and 0.01 M HCl, while the Mn(VII) solution contained 20 mM KMnO4, both dissolved in deionized water. Electrodeposition and all electrochemical measurements were carried out in a 3-electrode system using a Pt counter electrode of 10 × 20 × 0.2 mm3 and an Ag/AgCl reference electrode. The working electrode is the glass felt soaked with SWCNTs, having apparent area of 4.5 cm2. The depositions of RuO2 and MnO2 were both performed using cyclic voltammetry (CV). RuO2 was deposited from the RuCl3 solution by CV upon scanning the potential between 0.4 and 1.1 V at 50 mV/s. Similarly, MnO2 was deposited from the KMnO4 solution by cycling the potential between 0.4 and −1.0 V at the same scan rate. Homogeneous aqueous SWCNT suspension of 1 mg/mL was prepared by sonication for 2 h using a probe sonicator. The glass fiber felts were cut into 1.5 × 3 cm2 rectangles and immersed in the SWCNT solution for a few seconds and then dried at 120 °C for 6 min. The immersion and drying steps are denoted as a single SWCNT deposition cycle.

Fabrication of the layer-deposited samples comprised an initial stage of 15 SWCNT deposition cycles to assure good conductivity. The weight and conductivity of the SWCNT layer were measured after each cycle. The 15 SWCNT deposition cycles were followed by repeated alternating deposition via 50 CV cycles of oxide (either MnO2 or RuO2) and 5 soaking cycles of SWCNT until 150 CV cycles of oxide and 25 deposition cycles of SWCNT in total (15 SWCNT/50 oxide/5 SWCNT/50 oxide/5 SWCNT/50 oxide) as shown in Fig. 1 (from (b) to (g)) were obtained. Fabrication of bulk deposited samples comprised 25 deposition cycles of SWCNTs followed by 150 CV cycles of oxide (25 SWCNT/150 oxide).

Fig. 1
figure 1

FESEM image of bare glass fiber and schematic of the layered and bulk deposited structures

Full size image

Resistivity measurements was obtained using a Jandel-RM3000 test unit and multi-height probe with solid tungsten carbide probe needles at 1 mm linear spacing and 100 g loads. The electrochemical characterization of the samples was carried out by CV in 1 M H2SO4 for RuO2–SWCNT and 1 M Na2SO4 for MnO2–SWCNT samples at 5 mV/s. The electrochemical impedance spectroscopy was measured in 1 M H2SO4 solution at open circuit potential in the frequency range of 105–10−2 Hz, with a sine wave of 10 mV amplitude. Cyclic stability was measured by means of galvanostatic charge/discharge at a current density of 1 A/g from 0 to 0.7 V for all cells. All electrochemical measurements were performed at room temperature (~25 °C) on a Solartron 1400A potentiostat/galvanostat. The morphology of the samples was characterized using the JEOL 7600F field-emission scanning electron microscope (FESEM).

Results and discussion

Before constructing the SWCNT–metal oxide layered assembly, the deposition of multi-layers of SWCNT was studied in order to optimize the weight of SWCNTs to be loaded. The optimal electrode should possess the lowest possible weight and resistivity. Figure 2 shows the effect of loading cycles (up to 30) of SWCNTs on the weight, resistivity, and capacitance of the electrodes. As seen from Fig. 2a, the weight of the SWCNTs increases linearly with the loading cycles, implying an almost constant loading per cycle. Below 10 loading cycles, high resistivity was recorded due to the insufficient conduction pathways. As the SWCNT loading increases, the resistivity decreases. However, as the loading cycles increase beyond 15, the decrease in resistivity is insignificant. This is due to the addition of CNTs which increases the conduction pathway (decreases resistivity) but also adds to the number of nanotube–nanotube junctions (increases resistivity). Figure 2b shows the effect of SWCNT loading on the capacitance of the electrodes. The capacitance of the sample with 5 loading cycles could not be measured due to the poor conductivity of the electrode. Figure 2b shows very clearly that the specific capacitance increases with loading. Considering the specific capacitance, weight, and resistance, the rest of the experiments were carried out applying 15 cycles of SWCNT loaded onto the glass fiber felt as the starting material for the electrodeposition of the oxides because it shows a good compromise between the weight, resistivity, and capacitance of the loaded SWCNTs.

Fig. 2
figure 2

a Resistivity of the electrode and weight of SWCNT as a function of the number of SWCNT loading cycles. b Specific capacitances versus the SWCNT loading cycles

Full size image

Next, 7 samples of RuO2 or MnO2 electrodeposited on the SWCNT were made. From previous reports, it was noted that the performance of MnO2-based pseudocapacitors does not differ drastically from the RuO2-based pseudocapacitors at low oxide loading [18, 32, 33], and MnO2 has lower cost and is more environmental friendly as compared with RuO2 [32, 34]. Figure 3a shows the specific capacitance of SWCNT–oxide layered electrodes with different loading. Specifically, SWCNT and oxide layers were sequentially deposited as shown in Fig. 1(b)–(g). It is evident that while the specific capacitance of the SWCNT–RuO2 increases with the loading, that of the SWCNT–MnO2 levels off after the first oxide deposition (seen in Fig. 3a). This alludes to the difference between the microstructures of the SWCNT–oxides vide infra.

Fig. 3
figure 3

a Specific and b areal capacitance of (b)–(g) layered samples as illustrated in Fig. 1. The SE of the mean was obtained by dividing the mean with the square root of the sample size (n = 7)

Full size image

Figure 3b depicts the areal capacitance determined by dividing the actual capacitance with the geometric area (~4.5 cm2). Due to the uncertainty in weight measurement as well as the concern for practical usage, the areal capacitance is a more reasonable parameter for comparison. There are three notable observations in Fig. 3b: first, the areal capacitance increases as a result of the first electrodeposition of the oxide as compared with bare SWCNT electrode; second, the capacitance also increases with each of the oxide deposition cycle; third, the difference in capacitance between the two oxides is relatively small. These findings indicate that the electrolyte penetrates across the entire layered structure leading to the increase of capacitance with each layer of oxide deposition. The mean specific capacitances for the best configuration studied [25 layers of SWCNT and 150 electrodeposition cycles of oxide, Fig. 1(g)] for the SWCNT–MnO2 and SWCNT–RuO2 are 72 and 98 F/g, whereas the corresponding areal capacitances are 0.07 and 0.09 F/cm2, respectively.

Further characterization of the electrodes was carried out by CV. Figure 4 shows the CV of SWCNT–oxides. The capacitance of the pseudocapacitors was calculated using the relationship, \( C = \frac{\Delta Q}{\Delta V} = \frac{{\int\limits_{\text{Lowerhalf}} {I{\text{d}}t} }}{\Delta V} \), where C is the total capacitance, \( \Delta Q \) is the amount of charge stored, \( \Delta V \) is the potential window for discharge, and \( \int\limits_{\text{Lowerhalf}} {I{\text{d}}t} \) is the integral of the discharge current with respect to time. The supercapacitors for which the CVs are shown in Fig. 4 were loaded up to 40 wt% of the active material (SWCNTs and oxide) for the MnO2 samples and about 20 wt% of the active material for the RuO2 samples at 25 layers of SWCNT and 150 cycles of the respective oxides. It can be seen that the specific current for the SWCNT–RuO2 samples was higher than that for the SWCNT–MnO2 samples. Furthermore, SWCNT–MnO2 discharges at a lower rate as compared to the SWCNT–RuO2 (seen from the slope of the CV upon reversing the scan). The latter is presumably attributed to the higher conductivity of RuO2 [23, 32].

Fig. 4
figure 4

CVs of SWCNT–RuO2 (in black) and SWCNT–MnO2 (in red). Scan rate: 5 mV/s (Color figure online)

Full size image

Better insight to the SWCNT–MOx-electrolyte interfaces can be obtained from EIS. Figure 5a shows the Nyquist diagrams of the SWCNT-only, SWCNT–RuO2, and SWCNT–MnO2 electrodes. Four segments, i.e., inductance, a semi-circle, Warburg diffusion and an incomplete arc, can be seen in each curve from high to low frequency. The high-frequency inductance is likely to be related to the porous nature of the electrodes (felts). The semi-circle at high frequency domain reflects the charge transfer resistance and the double-layer capacitance at the electrode/electrolyte interface. The Warburg diffusion in mid-frequency range is attributed to the diffusion of electrolyte in the active materials. The low-frequency incomplete arc may arise from the parallel of mass capacitance and leakage resistance [35, 36]. The data were fitted using the equivalent circuit illustrated in Fig. 5b, and selected parameters are shown in Table 1. The elements involved include the R s, which mainly reflects the DC resistance of the film, C dl which is the capacitance of the double layer, R ct which is the charge transfer resistance, W, the Warburg element, R L the leakage resistance, and C L is the actual capacitance of the device. It can be seen that the R s of the electrodes increases upon introducing the metal oxide. The resistance increase is likely due to the coverage of SWCNTs by metal oxides. The R s is higher in the SWCNT–MnO2 electrode than in the SWCNT–RuO2, which is in agreement with the CV results (slope of charging/discharging). Furthermore, the R ct and C dl values for the SWCNT and SWCNT–RuO2 samples are similar. This suggests that loading of RuO2, unexpectedly, does not significantly affect electron transfer, which could be attributed to the increase of the surface area due to the porosity. The latter is confirmed by the morphology of the electrode surface as shown in FESEM (Fig. 6). On the other hand, the SWCNT–MnO2 samples have higher R ct and lower C dl, probably related to the compactness of the MnO2 films (Fig. 6). Moreover, the SWCNT–RuO2 electrode has higher mass capacitance as compared with the SWCNT-only electrode, which is in agreement with the trend seen from CV results. Unfortunately reliable R L and C L values for the SWCNT–MnO2 electrode could not be derived due to insufficient low-frequency data.

Fig. 5
figure 5

a Electrochemical impedance spectroscopy of SWCNT-only, SWCNT–RuO2, and SWCNT–MnO2 devices measured at OCP in 1 M H2SO4 solution and b the equivalent circuit for fitting the data

Full size image
Table 1 Selected EIS fitting results for the SWCNT-only, SWCNT–RuO2, and SWCNT–MnO2 samples
Full size table
Fig. 6
figure 6

FESEM showing microstructure of layered SWCNT–RuO2 (a) and SWCNT–MnO2 (b)

Full size image

All these findings show that integrating the oxides with SWCNTs improves the capacitive performance of the electrodes, and in this context, the RuO2 coatings perform better than the MnO2. It is evident from Fig. 6 that the RuO2 surface is smoother and more homogeneous than that of the MnO2 but the typical pore sizes are much larger for RuO2 as compared to the MnO2 surfaces. Hence, the higher capacitance of the former may be attributed to the higher permeability of the RuO2 layers.

To evaluate the cyclic capability of the pseudocapacitors, the capacitance retention was measured up to 1000 cycles. Figure 7 shows the change in the capacitance as a function of the number of charge/discharge cycles. The inset presents the individual charge/discharge at the 3rd, 500th, and 1000th cycle. While the capacitance of the SWCNT–MnO2 shows a single significant decrease at approximately 300 cycles and remains more or less constant, that of SWCNT–RuO2 decreased after 100 cycles and it declines continuously. After 1000 cycles, the SWCNT–MnO2 sample retained 93 % of its original capacitance, while the SWCNT–RuO2 sample retained only 86 %. This indicates that the SWCNT–MnO2 electrodes have a higher cyclic stability as compared with the SWCNT–RuO2 samples.

Fig. 7
figure 7

Potential cycling showing capacitance retention of optimal a SWCNT–MnO2 and b SWCNT–RuO2 samples after 1000 cycles

Full size image

A comparison was also made between the layered (refer to (g) in Fig. 1) and the bulk deposited (refer to (i) in Fig. 1) samples. The latter was prepared by first soaking the glass fiber felt in the SWCNT solution (for 25 cycles) followed by electrodepositing the respective oxide (for 150 cycles). From Fig. 8, it can be seen that the layered samples improved more in terms of both specific and areal capacitances as compared to the bulk samples using the SWCNT-only samples as the baseline. Layering the deposition of SWCNTs and the oxides should increase the surface area of the oxide that is in contact with the SWCNTs hence maximizing the surface area of SWCNTs and oxides that is in contact with the electrolyte.

Fig. 8
figure 8

Chart showing the improvement from SWCNT-only samples in both specific and areal capacitance for the layered and bulk samples after oxide deposition. The mean specific and areal capacitances of the CNT-oxide-layered samples were taken as a ratio to the mean specific and areal capacitances of the SWCNT-only samples to give the capacitance enhancement showing that the CNT-oxide-layered samples showed more specific and areal capacitance enhancement as compared to the bulk samples

Full size image

The performance of energy storage devices is commonly compared on a Ragone chart [37] where the specific energy is plotted versus the specific power. Figure 9 shows the Ragone plot where the performances of our electrodes are compared with commercial devices and some selected reports. This approach shows great promise in terms of the specific energy (4.14 and 4.25 Wh/kg for MnO2–SWCNT and RuO2–SWCNT devices, respectively, as compared to 2.35 and 3.60 Wh/kg for supercapacitors from Maxwell and Nesscap, respectively) taking into account that our system has not been fully optimized. Moreover, it should be noted that specific configuration of the glass fiber felt in our work results in high active surface area leading to high areal capacitance of the electrodes as compared to many other systems. This offers advantages in terms of practical applications such as printed flexible electronics and miniaturized capacitors.

Fig. 9
figure 9

Ragone plot of some commercial supercapacitors, recent reported values and the layered SWCNT–RuO2 and SWCNT–MnO2 samples [38, 39]

Full size image

Conclusions

In conclusion, the approach demonstrated in this study stems from the combination of an inert three-dimensional substrate that is coated by a conducting carbonaceous material and a redox oxide to form a layered structure. The inert substrate provides a low-cost porous network whereas the SWCNT provides the high conductivity and double-layer capacity which is well coupled with the high charge storage ability of the reversible redox oxides. Further improvement and optimization of this concept can be achieved by better control of the deposition parameters. This work has the potential to incorporate carbon nanomaterials into fabrics to be used as portable, wearable, and high-performance pseudocapacitors. The pseudocapacitors based on glass fiber felts can also be used to power electronic devices in harsh environments that involve high temperatures.