Solid-state supercapacitors with rationally designed heterogeneous electrodes fabricated by large area spray processing for wearable energy storage applications

Supercapacitors are in demand for short-term electrical charge and discharge applications. Unlike conventional supercapacitors, solid-state versions have no liquid electrolyte and do not require robust, rigid packaging for containment. Consequently they can be thinner, lighter and more flexible. However, solid-state supercapacitors suffer from lower power density and where new materials have been developed to improve performance, there remains a gap between promising laboratory results that usually require nano-structured materials and fine-scale processing approaches, and current manufacturing technology that operates at large scale. We demonstrate a new, scalable capability to produce discrete, multi-layered electrodes with a different material and/or morphology in each layer, and where each layer plays a different, critical role in enhancing the dynamics of charge/discharge. This layered structure allows efficient utilisation of each material and enables conservative use of hard-to-obtain materials. The layered electrode shows amongst the highest combinations of energy and power densities for solid-state supercapacitors. Our functional design and spray manufacturing approach to heterogeneous electrodes provide a new way forward for improved energy storage devices.

become thicker because of decreasing material utilisation through the electrode thickness, especially when ionic diffusion is limited such as the case in solid-state supercapacitors; but thicker electrodes are desirable to reduce the overall current collector area and increase capacitance per area 14 . The focus here is to maximise the utilisation of active materials in relatively thick, structured electrodes for solid-state supercapacitors.
The electrodes were manufactured using a scalable spray technique that enabled a discrete, heterogeneous three-layer electrode structure to be manufactured over large areas. Because of the discrete layered structure and the different intrinsic response of each layer, a transition in the electrode charge storage dynamics, from charge storage primarily due to redox reaction at slow scan rates (< 500 mV s −1 ) to solely high electric double layer (EDL) capacitance at fast scan rates (500 mV s −1 ) was clearly resolved.

Design rationale for the electrode structure
Laboratory synthesised meso-porous anatase TiO 2 particles (p-TiO 2 ) have recently been studied for energy conversion and storage applications, such as solar cells and supercapacitors 15,16 , but this is the first time that p-TiO 2 has been applied for solid-state supercapacitors. These meso-porous TiO 2 particles made by a template-based route are potentially attractive because of their high surface area, but are available currently only in comparatively small quantities. We mixed p-TiO 2 with multi-wall carbon nanotubes (MWNTs) in a stable suspension that was sprayed directly onto a 1M H 2 SO 4 treated H + ion conducting Nafion membrane to form the first, 650 nm thick layer (Layer 1) of a three-layer electrode, as shown schematically in Fig. 1(a). The rationale was to exploit optimally the high surface redox reactivity of p-TiO 2 at the membrane interface where there was the highest concentration of H + ions, and so to use this hard-to-obtain material both efficiently and conservatively.
A much thicker (Layer 2, 32 μm) of smaller and non-porous, readily available commercial anatase TiO 2 nanoparticles (c-TiO 2 , 20 nm in size), again mixed with MWNTs, was then sprayed on top of Layer 1. The rationale was to provide an inter-connected MWNT network decorated with the c-TiO 2 nanoparticles that exploited both the redox reaction possibilities of TiO 2 and the EDL capacitance of the inter-connected MWNT scaffold decorated with c-TiO 2 nanoparticles (since MWNTs are well-known to provide EDL capacitance in the presence of H + 17 ). A critical aspect was that for Layers 1 and 2, ion conducting ionomer was co-sprayed along with the TiO 2 and MWNTs so that during in-situ drying at deposition, the TiO 2 attached to the MWNT scaffold, and both were coated with ionomer 12 . The ionomer coating was intended to promote ion mobility throughout the electrode in the solid-state, while the inter-connected MWNT network provided high electrical conductivity.
Finally, Layer 3 (~200 nm in thickness) comprising low-defect few-layer graphene sheets (400 S cm −1 18 ) made by shear exfoliation of graphite in deionised water, was sprayed on top of Layer 2 with the intent to decrease the contact resistance between the electrode and the subsequently added current collector. Some graphene sheets moved into the porous MWNT scaffold to connect to the MWNTs, and the edge planes exposed by the graphene sheet also contributed some further EDL capacitance [19][20][21][22] . Cu current collectors were then pressed onto Layer 3 after the electrode was dried, so the layer of graphene sat at the interface of Layer 2 of the electrode and the current collector itself. To check for any Cu reaction or corrosion effects, indium tin oxide (ITO) coated glass current collectors in an identical arrangement were also studied, and the consistent electrochemical results between Cu and ITO coated glass current collectors confirmed that any underlying Cu reaction with the electrode or electrolyte could be neglected, and as previously shown 17 . Spray processing. Fabrication of the symmetric solid-state supercapacitor with electrodes each comprising of up to three discrete layers was realised by spray atomisation and deposition of up to three different suspensions in sequence, in a single operation.
The three types of aqueous-based suspensions of electrode materials were prepared by sonication at 600 W and 20 kHz for 30 min. Fig. 1(b) shows the spray apparatus where multi-nozzles sprayed consecutively the three types of aqueous-based suspensions of electrode materials into the three layers of the electrode, onto an H + ion conducting Nafion membrane, maintained at 100 °C on a heated vacuum stage. The nozzles moved in a pre-programmed zig-zag pattern along X and Y directions at 20 mm s −1 to spray an area of up to 20 cm × 20 cm. Any fugitive water from the spray suspension evaporated continuously on the heated vacuum-chuck stage as the electrodes formed. The membrane with one side covered with the three-layer sprayed electrode was then flipped and the other side sprayed using the identical procedure, to directly form a solid-state supercapacitor with no need for any subsequent re-immersion in liquid electrolytes. No binders were needed for any of the layers.
To allow comparison with electrodes of similar materials 23 , the average loading mass of the electrodes was determined as 1.06 mg cm −2 (with an error estimation of ± 1.1% based on measurements of 50 samples), with coin cell test area of 1.13 cm 2 and 1.2 cm diameter. Larger electrodes of 4.5 cm × 4 cm were also investigated to show scalability, and gave consistent areal capacitances to within ± 2.7% variation of the smaller cells. Figs. 1(c) and S1 in the Supplementary Information (SI) show an even larger 16 cm × 9.5 cm solid-state supercapacitor fabricated by the same spray method. The spray process is capable, so far, of making 200 nm-70 μm thick and up to 1 m × 20 cm large area electrodes using a drum coater variant 24 .

Results and Discussion
Morphology, electrode structure and surface area characterisation. In order to quantify the benefits of the layering approach ( Fig. 1(a)), various other layered and non-layered electrodes that contained the same materials but in different, but comparable arrangements were fabricated identically, with the same thickness to within a ± 2.4% variation, and are shown in Tables 1 and 2.  The surface area of the p-TiO 2 powder was measured by the Brunauer-Emmett-Teller (BET) method as 233 ± 1.2 m 2 g −1 compared with 50 ± 1 m 2 g −1 for the c-TiO 2 powder. Here, we were also able to measure the BET surface areas of the as-sprayed free-standing electrodes of [p-TiO 2 + MWNT] and [c-TiO 2 + MWNT]. Fig. S4 shows an image of a 2.8 cm × 2.3 cm free-standing electrode of [c-TiO 2 + MWNT] peeled carefully from the substrate for the BET measurement. For a fair comparison, the electrode thickness was kept the same and the weight ratio between TiO 2 and MWNTs in both films was kept the same at 1:2. The electrode specific surface areas were 143 m 2 g −1 and 98 m 2 g −1 , the pore volumes were 0.44 cm 3 g −1 and 0.26 cm 3 g −1 , and the average adsorption/ desorption pore sizes were 12 nm and 9 nm for the [p-TiO 2 + MWNT] and [c-TiO 2 + MWNT] free-standing electrodes respectively. Therefore, p-TiO 2 contributed more strongly towards an overall higher electrode surface area than c-TiO 2 .
To characterise the graphene sheets in Layer 3, Fig. S5 is a Raman spectrum of a drop of aqueous suspension of the exfoliated graphene sheets, showing a relatively weak narrow D band, a dominant narrow G band indicative of few defect graphene made by exfoliation, and a 2D band suggestive of 4 to 7 layers of graphene 18 . Holey-carbon transmission electron microscopy (TEM) grids were sprayed simultaneously with Layer 3, and Fig. 2(e) is a wide-field TEM image showing full coverage. The multi-layer graphene sheets (red circle) were typically ~360 nm × 200 nm, consistent with the literature 18 . The graphene sheets (green circle in the top right corner in Fig. 2(e)) were magnified in Fig. 2(f), showing straight edges with ordered fringes, which were further magnified in Fig. 2(g) and manual counting suggested 4-17 layers after spraying 25 .
To characterise together Layers 2 and 3 of the three-layer electrode (E3 in Table 1), X-ray photoelectron spectroscopy (XPS) depth profiling using Ar + ions sputtering through the top Layer 3 was used to investigate chemical composition changes through Layer 3, down into Layer 2. Fig. S6 shows the atomic % profiles of C, O and Ti versus etch depth. As expected, the profiles showed that the C concentration decreased and the Ti concentration increased with increasing depth, supporting the intent of a multi-layer structure where the layers had different compositions. The C concentration was stable at ~200 nm, agreeing with the estimate of ~200 nm thick graphene Layer 3 measured by a stylus profilometer. Since the top graphene layer may not be entirely continuous and can be permeable to incident X-rays, some Ti signal was inevitably detected even before all of the graphene was etched away, but Figure S7 shows that as expected the Ti 2p peak intensity increased as etch depth further increased.
Electrochemical characterisation. Fig. 3(a) shows the cyclic voltammetry (CV) curves of the solid-state supercapacitor using the three-layer electrodes E3 up to 500 mV s −1 . The main pair of redox reaction peaks between 0.13 and 0.67 V shown in Fig. 3(a) was due to the reversible redox reaction between -OH functional groups of TiO 2 and H + ions from the H 2 SO 4 treated Nafion membrane, and residual H 2 SO 4 from the sprayed suspension, according to [26][27][28]12 : where X + indicates the protons and/or alkali metal cations (e.g. Na + , Li + and K + ) in the electrolyte 28 .  Fig. 3(b) shows the same CV data with current normalised by the square root of scan rate as a function of voltage for the same electrode, which can be used to better resolve low scan rate behaviour when a wide range of scan rates is used 17 . The small peak at ~1.3 V at 5 mV s −1 in Fig. 3(b) arose from the interaction of the mobile H + ions (in the ionomer coating in the electrode) and the negatively charged F − species in the ionomer 29 . This peak became negligible at scan rates faster than 5 mV s −1 , which represents a more typical range used for supercapacitors 30 . Fig. 3(b) also shows a pair of redox reaction peaks between 0.13 and 0.67 V associated with the -OH functional group of TiO 2 as shown in Eq. (1). To confirm the proposed redox reaction, Fig. 3(c,d) show the detailed XPS Ti 2p spectra of the pristine TiO 2 nanoparticles and the electrode E3 after one CV cycle at 5 mV s −1 respectively. Both spectra show peaks at 459 and 465 eV corresponding to Ti 4+ (Ti 2p3/2 ) and (Ti 2p1/2 ) respectively 31 . The electrode after a single CV cycle showed a peak at 455 eV corresponding to Ti 3+ 31 , showing that preparation of the aqueous-based solution for spraying and/or the redox reaction during charge and discharge increased the hydroxyl group concentration on the TiO 2 surface 12 . A similar change in oxidation state has also been shown by in-situ X-ray absorption spectroscopy (XAS) in titanium carbide in a supercapacitor electrode 32 . Additionally, Figure S8 shows the detailed XPS O 1s spectrum for the same electrode after the CV cycle. The peaks at 532.5 and Fig. S8 534.0 eV corresponded to Ti-O-Ti and Ti-OH, respectively 33 , confirming the presence of -OH groups on Ti. Fig. 3(e) shows the areal capacitance in relation to maximum cell voltage for the solid-state supercapacitor using the three-layer electrodes E3, which is another method to investigate the presence of any irreversible reactions within the cell 34 . The areal capacitance continued to increase almost linearly with increasing cell voltage from 0.3 to 1.5 V, indicating negligible irreversible reactions over this range 34,35 . To confirm further, electrode current as a function of time was also monitored at a constant voltage of 1.5 V, and as shown in Fig. 3(f), the current density was small and relatively stable at ~0.1 A g −1 over 2 hrs, consistent with negligible H 2 evolution or additional parasitic reactions 32 .
To study the benefits of the layered electrode, the electrochemical properties of solid-state supercapacitor cells using the different types of electrode arrangement were compared, as summarised in Tables 1 and 2. All electrodes were of the same thickness ± 2.4%. Fig. 4(a) shows the CV curves of the solid-state supercapacitors using Here, the capacitance of the Ec electrode arose due to (i) the wetted, intimate interface between the TiO 2 nanoparticles and the electron conducting MWNT scaffold, and with the ion-conducting ionomer that coated them both 12 ; and (ii) the supply of H + ions from the dilute H 2 SO 4 used for spraying and the redox reaction of the hydroxyl groups on TiO 2 nanoparticles, shown by the XPS spectrum in Fig. S8. The mobility of H + ions was provided by the negatively charged SO 3 − end groups in the ionomer coating. The mobile H + ions were charge compensated by the relatively high electron mobility in the inter-connected MWNT that formed the electrode 38 , reflected by the relatively high current densities in Fig. 4(a) 39 . Fig. 4(b) shows the CV curves of the solid-state supercapacitors using [p-TiO 2 + MWNT] electrodes (Ep). The mass ratio of TiO 2 : MWNT was kept the same for both electrodes in Figs. 4(a,b). The CV curves exhibited more prominent redox reaction peaks than electrode Ec, showing that the porous p-TiO 2 with a higher surface area contributed proportionally more redox reactions to charge storage than the non-porous c-TiO 2 40 . However, the absolute current densities for electrode Ep in Fig. 4(b) were lower than for electrode Ec in Fig. 4(a). Four point probe measurements of electrodes Ep and Ec gave electrical conductivities of 0.22 S cm −1 and 3.4 S cm −1 respectively, suggesting that even though the p-TiO 2 had a higher specific surface area than c-TiO 2 , the nearly one order of magnitude larger p-TiO 2 particles inhibited the electrical connectivity of the MWNT network 41 . Fig. 4(c) shows the CV cur ves of a solid-state supercapacitor using randomly mixed [p-TiO 2 + c-TiO 2 + MWNT] electrodes (Epc). Comparing the CV curves for the different electrodes shows that the three-layer electrode E3 in Fig. 3(a) exhibited more prominent redox peaks than the randomly mixed electrode Epc, despite comprising largely the same electrochemically active materials. To understand how electrode structure gave such distinct differences, Trasatti's method 42 for each electrode was used to deconvolute surface and diffusion-controlled contributions to capacitance. The method relies on the assumption that the surface and diffusion-controlled contributions are governed by different kinetics, and respond differently to increasing scan rates 43 . Fig. 5(a) shows 1/C (C = areal capacitance) for the electrode E3 as a function of the square root of the scan rate v in the 5-100 mV s −1 region, where both surface and diffusion-controlled contributions were significant 43 . The intercept of the linear region of the plot with the 1/C axis estimated the total capacitance possible for E3 at an infinitely slow rate as 333.3 mF cm −2 . Then, to extract the surface-controlled contribution to capacitance, Fig. 5(b) shows C as a function of v −1/2 . The intercept of the linear region of the plot with the C axis estimated the surface charge at a scan rate of infinity as 62.5 mF cm −2 . In comparison, Figs. 5(c,d) estimated the total capacitance for Epc as 74.2 mF cm −2 , the surface charge at a scan rate of infinity as 34.1 mF cm −2 . The lower surface charge of Epc than E3 likely arose from a lower active surface area because the smaller c-TiO 2 may have blocked some of the pores in the larger p-TiO 2 (the size of the c-TiO 2 was 20 nm and the pores in p-TiO 2 was 20-50 nm). These negative synergistic interactions between the two types of TiO 2 were avoided in the discrete layered electrode structure. Tables 1 and 2    for a ~33 μm thick electrode) because, as previously shown, the comparatively large p-TiO 2 particles (200 nm) prevented the formation of a well-inter-connected MWNT network throughout the electrode. This is because the MWNTs after sonication and spraying were typically ~500 nm in length, while the p-TiO 2 particle size was ~200 nm: when the particle and MWNT length scales were similar, the percolating MWNT network was more restricted, as shown by the electrical conductivity measurements. Consequently, a proportion of the p-TiO 2 particles remained electrically isolated within the electrode and could not contribute to charge storage.
On the other hand if the same electrode consisted entirely of [c-TiO 2 + MWNT] only, the specific areal and gravimetric capacitance of the electrode might be expected to increase (91.7 mF cm −2 and 86.3 F g −1 at 5 mV s −1 for the same electrode thickness) because these smaller TiO 2 nanoparticles (20 nm) were more readily electrically connected into the MWNT network. However, while this may be the case, as the early CV curves showed, these non-porous c-TiO 2 particles did not contribute as much pseudo-capacitance as the high surface area porous p-TiO 2 40 . Instead, by placing the p-TiO 2 at the interface between the H + treated Nafion membrane and the rest of the electrode only (Layer 1 in the schematic diagram in Fig. 1(a)), with a lower fraction of MWNTs, the higher redox reactivity of the p-TiO 2 could be exploited, without undermining connectivity in the majority of the rest of the electrode. Then using the c-TiO 2 nanoparticles and a higher fraction of MWNTs (Layer 2 in Fig. 1(a)) for the rest of the electrode ensured a high surface area, inter-connected MWNT network and a relatively high conductivity pathway from the p-TiO 2 to the current collector. For both layers, the ionomer coating facilitated H + movement and EDL contributions to capacitance. Critically, the high redox reactivity in Layer 1 and the generation of additional H + ions to promote EDL capacitance combined to give an overall high capacitance of 247.6 mF cm −2 (237.4 F g −1 ) at 5 mV s −1 for the same electrode thickness. Capacitance was further increased after adding a third layer of graphene, as now described below.
Overall, the variations in capacitance measurements for each electrode type in Tables 1 and 2 were in the range ± 3-6%, and were thus significantly smaller than the differences in capacitance among the different electrodes. We note that if the differences in electrochemical behavior were controlled primarily by underlying additional reactions, the results between the five different electrode arrangements of the same materials would not be so marked since all contained the same various materials. The strong differences between the electrode arrangements show that the arrangement of materials and electrode structure are the dominant effects on electrochemical response.
To further assess the proposed positive synergistic effects in the layered arrangement and to estimate the utilisation of the active materials in the solid-state supercapacitor arrangement, the electrodes were also tested in a standard three-electrode configuration using a liquid 1 M H 2 SO 4 electrolyte, Pt counter electrode and Ag/AgCl reference electrode 44 . Fig. 6(a) shows the CV curves of the [c-TiO 2 + MWNT] electrode Ec with a parallelogram shape, indicating fast charge/discharge kinetics typical for supercapacitors using a liquid electrolyte. The estimated capacitance for the electrode Ec in a liquid electrolyte from Fig. 6(a) was 92.5 mF cm −2 (87.1 F g −1 ) at 5 mV s −1 and 64.1 mF cm −2 (60.4 F g −1 ) at 100 mV s −1 . By comparison with capacitances of 91.7 mF cm −2 (86.3 F g −1 ) at 5 mV s −1 and 46.3 mF cm −2 (43.7 F g −1 ) at 100 mV s −1 for the same electrode Ec in the solid-state supercapacitor configuration, the average utilisation of active materials in the solid-state for the composite electrode Ec was estimated as 86%.
The three-layer electrode E3 was also tested using the same three-electrode configuration. Fig. 6(b) shows the corresponding CV curves for E3 with more obvious redox reaction peaks again confirming the better performance of the layered electrode, and also a parallelogram shape showing fast charge/discharge kinetics. The capacitance of E3 estimated from Fig. 6(b) was 277.8 mF cm −2 (270.8 F g −1 ) at 5 mV s −1 and 99.4 mF cm −2 (94.6 F g −1 ) at 100 mV s −1 . Again by comparison with capacitances of 272.5 mF cm −2 (265.9 F g −1 ) at 5 mV s −1 and 77.1 mF cm −2 (73.3 F g −1 ) at 100 mV s −1 for the same electrode E3 in the solid-state configuration, the average utilisation of active materials in the solid-state was estimated similarly as 88%. Fig. 3(a) also showed that the CV curve of the solid-state supercapacitor using the three-layer electrodes E3 became more parallelogram-shaped at a relatively fast scan rate of 500 mV s −1 , suggesting redox reactions no longer had time to occur 27 , leaving only residual EDL capacitive behaviour at the interface between the MWNTs, TiO 2 and the ionomer coating. Fig. 7(a) shows the CV curves of the solid-state supercapacitor using electrodes E3 up to extremely fast scan rates of 2000 mV s −1 , with the curves maintaining an approximate parallelogram shape. Similar fast charging behaviour has been shown in exfoliated graphene electrodes in a liquid electrolyte of 1-butyl-3-methyl-imidazolium tetrafluoroborate in acetonitrile 45 . In contrast, Fig. 7(b) shows a more distorted CV curve, even at 500 mV s −1 , for the solid-state supercapacitor using the two-layer electrodes E2 (i.e. without the graphene layer, otherwise identical to E3, as summarised in Tables 1 and 2) indicative of restricted charge/discharge kinetics at fast scan rates for E2 46 . The charging and discharging kinetics of electrode E3 were improved because some few-layer graphene sheets infiltrated into the MWNT scaffold during fabrication before the fugitive carrier (dilute H 2 SO 4 ) evaporated completely. The few-layer graphene enhanced electrical connectivity of the MWNT scaffold and its connection to the current collector, as shown schematically in Fig. 8. The many few-layer graphene sheets also likely made a contribution to EDL capacitance since their exposed edge planes can provide up to an order of magnitude higher EDL capacitance than the basal planes provided by the graphene surface and MWNTs [19][20][21][22]47 .
The dependence of capacitance on scan rate for both the three-layer electrode E3 and two-layer electrode E2 is shown in Fig. 7(c). Capacitance was reduced but relatively stable at scan rates above 500 mV s −1 because there was no redox reaction energy storage contribution, only residual EDL capacitance. There was a 42% decrease in capacitance for E3 compared with a 70% decrease for E2 as the scan rate increased from 500 mV s −1 to 2000 mV s −1 (summarised in Tables 1 and 2), noting that a scan rate of 2000 mV s −1 is amongst the highest used for supercapacitors [48][49][50] , showing that the three-layer electrode E3 with a graphene layer had a greater ability to maintain residual EDL capacitance than the two-layer electrode E2 without the graphene layer, again indicating that the few-layer graphene contributed to the connectivity of the MWNT scaffold, and to its efficient connection to the current collector.
For the three-layer electrode E3 in particular, there was a resolvable transition from redox active behaviour at slow scan rates below 500 mV s −1 , dominated by p-TiO 2 , to EDL capacitive behaviour above 500 mV s −1 , which was particularly enabled by the graphene. Conventional randomly mixtures of electrode materials integrate together the different storage contributions, masking the intrinsic behaviour of the constituent materials (typically metal oxides and C-based materials). Capacitance in these electrodes decreases comparatively quickly as 2000 mV s −1 is approached because the insulating effect of the metal oxides on electrical conductivity becomes exposed as their appreciable contribution to energy storage fades 51 . However, in E3 the discrete graphene Layer 3 was able to facilitate a distinct EDL component (through exposed edge planes and by efficient connections to the MWNT scaffold in Layer 2) even at the fastest scan rates. Similar effects have been reported in both hybrid nanoporous gold/MnO 2 films of 100 nm thick, where MnO 2 contributed redox-based pseudo-capacitance and nanoporous gold provided EDL capacitance 13 , and also in a pseudocapacitive Mo x N < 100 nm thick film coated on a Ti substrate with a H 4 SiW 12 O 40 -H 3 PO 4 -poly(vinyl alcohol) (SiWA-H 3 PO 4 -PVA) solid-polymer electrolyte 52 . Here, we have shown similar effects but in much thicker electrodes comprising discrete layers.
To further assess the effect of the graphene Layer 3 in the electrode, galvanostatic charge/discharge and electrochemical impedance spectroscopy (EIS) were used. Fig. 7(d) shows the galvanostatic charge/discharge curves of the solid-state supercapacitor using three-layer electrodes E3 at a current density of 1 mA cm −2 , with the non-linear response again due to the redox reaction (which strictly is non-capacitive in nature) associated with TiO 2 12,26 . Fig. 7(e) shows the performance of electrode E3 at a higher current density of 3 mA cm −2 . The more linear response exposes again the underlying EDL behaviour of electrode E3 at high charge/discharge current densities, as previously discussed. The estimated capacitance per electrode for E3 from the linear part of the discharge curve was 246.2 mF cm −2 (231.8 F g −1 ) at 1 mA cm −2 and 64.6 mF cm −2 (60.8 F g −1 ) at 3 mA cm −2 . The IR drop is related to the internal resistance i.e. the sum of the electrolyte ionic resistance, electrode resistance and interfacial resistance 3,45 . The IR drop in Fig. 7(e) was 0.036 V, lower than 0.08-0.2 V commonly reported for solid-state supercapacitors tested in similar conditions 53 . Fig. 7(f) shows a Nyquist plot from a solid-state supercapacitor using the three-layer electrodes E3. The intersection point of the best-fit curve to the data with the real axis at high frequency represented the series resistance (R s ), which includes the contact resistance between the electrode material and current collector, and the resistance of the electrolyte 54 . R s was estimated as 3.5 Ω for the three-layer electrode E3, lower than 28 Ω for the two-layer electrode E2 in Fig. S9, and also lower than R s for other solid-state supercapacitors (e.g. 4 Ω for an electrode of MoS 2 on C cloth with a LiCl-poly(vinyl alcohol) (PVA) gel electrolyte 54 ), showing that as intended the graphene layer reduced the contact resistance between the electrode material and current collector.
The semi-circle diameter of the Nyquist plot at high frequency represented the charge transfer resistance (R CT ) of the electrode 55 , and was estimated at 8.5 Ω for the three-layer electrode E3, lower than 42 Ω for the two-layer electrode E2 in Fig. S9, and also lower than R CT for other solid-state supercapacitors (e.g. 10.7 Ω for an activated carbon-TiO 2 hybrid electrode with a H 3 PO 4 -PVA gel electrolyte 56 ). Therefore, adding a small amount of graphene as Layer 3 effectively reduced R s (contact resistance of the interface between electrode and current collector) by 88%, and also reduced R CT (through-plane electrode resistance) by ~80%, supporting the idea that some of the sprayed few-layer graphene mixed into the meso-porous electrode structure immediately on deposition. Furthermore, R s for electrode E3 without the current collector, measured by careful electrical connection directly to the graphene layer, was ~7 Ω, suggesting that if made more robust, the graphene layer could act directly as the current collector.
The volumetric energy and power densities of the three-layer electrode E3 were 23.3 mWh cm −3 and 380 mW cm −3 at 1 mA cm −2 , and 6.1 mWh cm −3 and 1400 mW cm −3 at 3 mA cm −2 , respectively. The energy and power densities compare favorably with the literature 57-60 as shown in Fig. 9(a). For example, although a reduced graphene oxide (RGO)/polypyrrole (PPy) interdigital electrode using a PVA-H 2 SO 4 gel electrolyte exhibited a higher power density of ~10000 mW cm −3 at a similar energy density of ~8 mWh cm −3 at a similar high current density, the electrode exhibited a lower maximum energy density of 13.2 mWh cm −3 compared with 23.3 mWh cm −3 for E3 at a low current density of ~1 mA cm −2 60 .
The gravimetric energy and power densities of the three-layer electrode E3 were 45.7 Wh kg −1 and 1.1 kW kg −1 at 1 mA cm −2 , and 19.0 Wh kg −1 and 4.2 kW kg −1 at 3 mA cm −2 , respectively. Fig. 9(b) shows the Ragone plot of gravimetric energy and power densities per electrode, with the three-layer electrode E3 again providing a competitive performance to other solid-state supercapacitor electrodes [61][62][63] .
However, gravimetric energy and power densities per electrode frequently do not give a realistic indication of the performance of an assembled cell, since full cells also contain current collectors, separators etc., and often volumetric normalisation is preferred 23 . In this study, the total volume of the solid-state supercapacitor cell including current collectors, electrodes and treated Nafion membrane was 0.0199 cm 3 . The volumetric energy and power densities of the cell were then estimated as 2.2 mWh cm −3 and 35.8 mW cm −3 at 1 mA cm −2 , and 0.6 mWh cm −3 and 127.5 mW cm −3 at 3 mA cm −2 , respectively. Again comparing with the literature, at a similar current density of 3 mA cm −2 , MnO 2 -TiN nanotube hybrid arrays using a PVA-KOH-KI-ethylene glycol (EG) gel electrolyte exhibited 0.7 mWh cm −3 and 115 mW cm −3 64 , comparable to our performance; and porous poly(3,4-ethylenedioxythiophene)(PEDOT) coated TiN nanotube array using a PVA-H 2 SO 4 -EG gel electrolyte exhibited 2.26 mWh cm −3 and 250 mW cm −3 , due to its high surface area of the porous material 65 . However, this PEDOT-based nanopore array was fabricated through HF corrosion of TiN 65 , suggestive of scalability problems for nearer industrial-scale processing. In contrast, the performance of electrode E3 was delivered through a process that can be easily scaled and operated for a wide range of materials.
When cycling at 100 mV s −1 , the electrode E3 maintained 90.2% capacitance after 10,000 cycles on the bench top (continuously exposed to ambient air and moisture with no packaging), offering encouraging potential in, for example, future wearable electronic applications.

Conclusions
A symmetric solid-state supercapacitor using three-layer electrodes was fabricated to exploit optimally the inherent advantages of the active materials. Porous TiO 2 placed at the interface between the electrode and the ion conducting membrane/separator contributed significant redox behaviour and charge storage behaviour, especially at slow scan rates. At the other side of electrode, at the interface with the current collector, exfoliated graphene reduced both contact and charge transfer resistance and increased EDL capacitance through exposed edge planes. The spray atomisation and deposition of the suspensions in layers onto the heated membrane facilitated rapid drying and the formation of a 3D inter-connected MWNT scaffold decorated with TiO 2 nanoparticles, both of which were coated with the ion-conducting ionomer. The resulting intimate interface between the TiO 2 , MWNTs and ion-conducting ionomer helped realise both redox-based and EDL contributions to capacitance. This rational design of a three-layer electrode was intended to maximise the contribution of each type of active material to improve overall performance, and to use hard-to-obtain active materials most effectively. By careful comparison of different electrode structures comprised of identical materials, the three-layer arrangement was shown to be the optimal arrangement of these particular materials. This paper has given a specific demonstration of a more general idea that placing different materials at different positions in structured electrodes, rather than using random mixtures as widely practised, can improve performance. This approach may thus find applications in other energy storage and conversion devices such as Li-ion batteries and fuel cells.

Methods
Before spray, the Nafion membrane was pre-treated by immersing in 1 M H 2 SO 4 at 60 °C for 30 min-a standard procedure to exchange perfluorosulfonate (SO 2 F) groups along the main polymer chain for SO 3 − H + groups 66 . Porous TiO 2 single crystals were synthesised by: TiF 4 was dissolved (20 mM to 400 mM) in water in a 125-ml-volume autoclave lined with Teflon to which 180 mM 1-methylimidazolium tetrafluoroborate was added. 650 mg silica template was added to 50 ml TiF 4 solution. A transparent mixture was formed and kept at 120 °C for 10 hr in an oven. After that, the precipitate (TiO 2 particles with the silica templates) was formed at the bottom of the Teflon reactor and was collected. The silica template was then selectively etched in 2 M aqueous NaOH at 80 °C for 60 min in a polypropylene beaker. The TiO 2 product was collected by centrifugation (3,000 rpm for 60 min) and washed several times in water and ethanol. Graphene was synthesised by: an aqueous suspension of 5 mg ml −1 graphite powder with 0.1 mg ml −1 NaC was prepared in a sonication bath (Ultrawave U1250D, 200 W, 30-40 Hz) for 51 hr. The resultant suspension was centrifuged at 3,000 rpm for 150 min. The supernatant containing graphene was collected 18 .
The weight of electrodes was measured by a microbalance (Sartorius) with 0.01 mg accuracy and electrode thickness using a Dektak 6M stylus profilometer (Veeco Instruments Inc). The conductivity of the electrodes was measured by using a standard four-point probe configuration and a Keithley 220 programmable current source meter on the electrodes deposited on Si wafers, with measurements repeated eight times on each electrode. Porous TiO 2 powders were characterised by XRD (Cu α radiation, λ = 1.5 A). Exfoliated graphene was investigated using Raman spectroscopy (Horiba LabRAM ARAMIS) with a 532 nm wavelength laser. The BET (Micromeritics Gemini V) specific surface areas were measured by using N 2 adsorption and desorption at 77 K. A full isotherm was recorded from relative pressures of 0.01 to 0.95 then back from 0.95 to 0.01. The BET surface area was calculated using the data between relative pressures of 0.05-0.3.
The surface chemistry was analysed by XPS in an ion pumped Thermo Scientific K-Alpha 128-channel detecting analyser equipped with an Al K X-ray source. The XPS analyser operated at a constant pass energy of 100 eV for wide scans and 20 eV for detailed scans. The etching of the samples for depth profile measurements was performed with Ar + sputtering at 1000 eV. The surface morphology of the electrodes was examined by SEM (JEOL 6500F at 10 kV), and high resolution TEM (JEOL 2100F at 200 kV). Electrochemical testing of solid-state supercapacitors was performed using a Reference 600/EIS300 Gamry potentiostat/galvanostat with a combination of CV, galvanostatic charge/discharge and EIS.
As a full cell can be treated as two capacitors in series, the capacitance of one cell C cell was calculated according to 34 : in which C 1 and C 2 are the capacitances of individual electrodes. In this type of symmetric solid-state supercapacitor, assuming two equal capacitors in series 67 , the capacitance of the electrode C electrode is related to the capacitance of the cell C cell by: Specific capacitance was calculated from both the CVs and galvanostatic charge/discharge curves. For the CVs, the specific gravimetric capacitance was estimated by integrating the area under the current-potential curve and then dividing by the sweep rate, the mass of film electrode and the potential window according to 68 . For the calculation of specific gravimetric capacitance, the total electrode mass including TiO 2 nanoparticles, MWNTs and ionomer gel was used, according to: where C electrode is the specific gravimetric capacitance (F g −1 ), m is the mass of one electrode (g), v is the scan rate (V s −1 ), V a − V c represents the potential window (V), and I is either the charging or discharging current (A). For specific areal capacitance, the area of one electrode was used for the estimation.
In the galvanostatic charge/discharge process, the specific gravimetric capacitance was estimated from the slope of the linear part of the discharge curve according to Eq. 5, where the discharge current I is normally used and t is the corresponding discharge time (s) from a voltage V 69 : For estimating specific areal capacitance from the galvanostatic charge/discharge process, the area of one electrode was used.
The volumetric and gravimetric energy densities E electrode and power densities P electrode per electrode were estimated from 69 : where x is the volume and mass of one electrode, and t is the discharge time. The volumetric energy density E cell and power density P cell per supercapacitor cell were also estimated from: where v cell is the volume of one supercapacitor cell including the current collectors, electrodes and Nafion membrane.