Facile synthesis and characterization of Bi13S18I2 films as a stable supercapacitor electrode material

Electrical double layer capacitors (EDLCs) featuring low-cost and solution-processable electrode materials have attracted signi ﬁ cant research interest for their green and economical applications in energy harvesting and storage devices. Here, we demonstrate a novel synthetic route for ﬁ lms of an underexplored 3-D hexagonal bismuth chalcohalide, Bi 13 S 18 I 2 , and investigate its potential as the active electrode material in EDLC-type supercapacitors. The synthetic procedure has been optimised and comprises the lowest annealing temperature (150 (cid:1) C) and the shortest processing time (1 h) currently reported. When integrated in a symmetrical EDLC with an aqueous NaClO 4 electrolyte, the Bi 13 S 18 I 2 -based device achieves a remarkable areal capacitance of 210.68 mF cm (cid:3) 2 with 99.7% capacitance retention after 5000 cycles. Both the Bi 13 S 18 I 2 powder and thin- ﬁ lm electrodes have been characterized through XRD, XPS, Raman spectroscopy, and SEM. The superior stability, low-cost, and facile synthesis of Bi 13 S 18 I 2 proves its promising potential for supercapacitor applications.


Introduction
Demand for efficient, economic, and eco-friendly energy storage systems has grown dramatically in recent years for applications in electronic appliances, electric vehicles, and a smart and sustainable energy grid. [1][2][3] Given the high production costs and environmental hazards of conventional energy storage technologies, researchers have focused on developing novel energy materials and optimising their performance in electronic devices. [4][5][6][7][8] In particular, interest has grown in high-Z ns 2 cations with an electronic conguration of [Xe]6S 2 5p 0 . 9,10 Among these main group cations, bismuth has attracted considerable attention because of its low-toxicity and structural-diversity. 11,12 For example, bismuth-based perovskite-like materials have been well-studied in applications such as lead-free photovoltaic devices, radiation detectors, and supercapacitor electrodes. [13][14][15][16][17][18][19][20] Among the inorganic bismuth-based materials, bismuth chalcogenides and chalcohalides have captured research interests for their manifold functionalities in photocatalysis, [21][22][23] photovoltaics, 24 and radiation detection. 25,26 Bi 2 O 3 and Bi 2 S 3 have already proven their promising potential in photocatalytic water purication, thermoelectric systems, supercapacitors, and electrochemical sensors. 21,27-29 Bi 2 S 3 has also displayed remarkably varied crystal morphologies, including nanorods, nanobelts, and microowers, which can be tuned for enhanced performance as the active electrode material in supercapacitors. 30,31 Among the common Bi-based chalcohalides, BiSI and BiSeI have been studied as potential photovoltaic materials due to their n-type semiconducting properties and high absorption coefficients, 32,33 while both theoretical and experimental studies have evaluated their promise in radiation detection. Still, despite BiSI's exhibition of strong photocurrent, 34 such Bi-chalcohalides have yet to yield satisfactory photovoltaic performance in solar cells. 32 Moreover, the fabrication of bismuth chalcohalides typically requires high annealing temperatures and laborious, multistep procedures, complicating research efforts and economic viability. To the best of our knowledge, no simple, one-step, solution-processing synthetic route has been reported.
Electrical double-layer capacitors (EDLCs) are best known for their fast charge-discharge cycles, high power densities, long life cycles, and greater energy densities compared to conventional capacitors. However, EDLCs typically suffer from poorer energy densities compared to so-called pseudo-capacitors or conventional batteries that employ strong faradaic reactions to chemically store charge. 35 Consequently, there has been a recent wave of research efforts directed toward advancing the energy densities of environmentally-benign, economical, and durable electrode materials for EDLC applications in order to enhance their commercial viability without sacricing long-life cycle stability. In this respect, bismuth chalcogenides including Bi 2 S 3 and Bi 2 O 3 have been widely studied and applied as highperforming active electrode materials intended as supercapacitors, albeit typically with substantial faradaic mechanism. 36 On the other hand, bismuth chalcohalides have oen been overlooked as potential energy storage materials. In this study, a relatively unexplored bismuth chalcohalide material, Bi 13 S 18 I 2 , has been synthesised from solution in a single and facile step. Traditionally, such materials are fabricated under high pressure and high temperature conditions requiring long reaction times. [37][38][39] Using our easily synthesised Bi 13 S 18 I 2 as the active material in EDLC-type supercapacitors, we have achieved a device areal capacitance of 210.68 mF cm À2 , a specic capacitance of 6.58 F g À1 , and superior cycle stability with 99.7% capacitance retention even aer 5000 cycles.

Material characterization
Synthesis. Bi 13 S 18 I 2 was synthesised via thermal decomposition of a precursor solution containing bismuth xanthate (Bi(xt) 3 ), a sulfur-containing organobismuth material, and BiI 3 . Bi(xt) 3 was synthesized from potassium xanthate and bismuth nitrate in a one-step reaction previously reported by Vigneshwaran and colleagues. 40 A wide range of Bi(xt) 3 : BiI 3 ratios from 2 : 1 to 26 : 1 were tested and spin-coated on quartz glass, forming a thin lm which was subsequently annealed at 150 C on a hot plate. The annealing temperature for the thin lms was set slightly higher than the decomposition temperature of xanthate to ensure reaction completion (Scheme 1).
Structural and electronic characterization. The diffraction patterns obtained for thin lms with 2 : 1, 10 : 1, and 18 : 1 ratios of the Bi(xt) 3 : BiI 3 precursor solution are recorded in Fig. 1. A crystallographic XRD pattern corresponding to Bi 13 S 18 I 2 was identied in each of the three spectra. The 2 : 1 ratio produces the pattern with the lowest peak intensity, attributed to the stoichiometric mismatch of the precursor solution. In contrast, using a 26 : 1 solution produced a thin lm with XRD ( Fig. S2 †) displaying both Bi 13 S 18 I 2 and Bi 2 S 3 crystal peaks, suggesting that the sulfur concentration of the precursor solution was too high. Under these considerations, a 18 : 1 Bi(xt) 3 : BiI 3 ratio was ultimately selected for further syntheses and characterization of the thin lms. We note that Bi 13 S 18 I 2 has been previously reported as [Bi(Bi 2 S 3 ) 9 I 3 ] 2/3 (ref. 38) and Bi 19 S 27 I 3 , 39 prompting Groom and colleagues to reconsider the crystal structure and propose the formula Bi 13 S 18 I 2 , 41 which we adopt in our study. Bi 13 S 18 I 2 contains [Bi 2 ] 4+ dimers, which justies the close proximity of the bismuth cations positioned at the origin of the unit cell (Fig. 2).
We also evaluated the effect of lm thickness on the crystallographic patterns. The XRD for thin lms comprised of 1, 3, and 5 layers are displayed in Fig. S3 † and the characteristic Bi 13 S 18 I 2 pattern is most clearly identied with the samples containing 3 and 5 layers. Fig. 3 displays the SEM images of the as-prepared Bi 13 S 18 I 2 . As the number of layers increases, the nano-particle crystals grow longer tubule-like nano-rod structures approaching 250 nm in length. This suggests that increased surface area may be achieved by multiple-layer deposition of Bi 13 S 18 I 2 during the thin-lm formation. We attempted to carry out BET isotherm determination of the surface area, however this is not possible on thin lms of such small mass.     Table 1 tabulates the observed experimental atomic percentages of specic bismuth, iodine, and sulfur in the lm. The values agree closely with the theoretical atomic composition of Bi 13 S 18 I 2. Diffuse reectance measurements were also conducted to obtain the optical band gap of Bi 13 S 18 I 2 in its solid state. An indirect energy gap (E g ind ) of 0.75 eV and a direct energy gap (E g dir ) of 0.91 eV were estimated, which compare well with the literature values in Table S1. † Furthermore, we performed cyclic voltammetry (CV) measurements on a thin lm sample spin-coated on FTO as the working electrode to evaluate the electrochemical behaviour of Bi 13 S 18 I 2 . The cyclic voltammograms in Fig. S4 † represent an average of 5 cycles at a sweep velocity of 0.1 V s À1 . The full CV scans indicate that Bi 13 S 18 I 2 undergoes irreversible redox processes, and the ionization potential (IP) can be estimated using the oxidation potential E ox against the Ag/AgCl electrode (4.4 AE 0.1 eV below vacuumlevel). 42 The electronic band gap of solid-state Bi 13 S 18 I 2 is estimated to be 0.99 V for the sample sweep at 0.1 V s À1 , calculated from the difference in E pa and E pc values and consistent with diffuse reectance (Fig. 5).
Supercapacitor studies Electrode preparation. The performance of Bi 13 S 18 I 2 as the active electrode material in an EDLC-type symmetric supercapacitor was evaluated using cyclic voltammetry (CV), galvanostatic charge-discharge, and electrochemical impedance spectroscopy (EIS). The Bi 13 S 18 I 2 -based electrodes were fabricated with a facile, single-step, and solution-processable procedure. An 18 : 1 mixture of Bi(xt) 3 and BiI 3 was combined with activated charcoal and polytetrauoroethylene (PTFE) powder and dispersed in DMF in sufficient relative amounts to form a 85 : 10 : 5 weight ratio of Bi 13 S 18 I 2 : charcoal : PTFE upon subsequent thermal decomposition of the Bi(xt) 3 . This heterogeneous solution was then drop-cast on 1 cm 2 sections of carbon cloth and heated at 150 C to achieve a high mass loading of approximately 8-10 mg cm À2 on each electrode. Although the Bi 13 S 18 I 2 constitutes the primary active material for the supercapacitor electrodes, the activated charcoal and   PTFE are added for enhanced electrical conductivity and mechanical stability respectively of the overall layer. The powder XRD pattern of as-prepared Bi 13 S 18 I 2 electrode is shown in Fig. 6, and the crystallite size of Bi 13 S 18 I 2 is calculated to be around 5 nm according to the Scherrer equation. The SEM images of the as-prepared electrodes show good coverage of the underlying carbon cloth with the activated Bi 13 S 18 I 2 material, with rod-shaped Bi 13 S 18 I 2 layering on top of the carbon cloth as well as binding directly to the carbon bres (Fig. 7b). A thin (25 mm) microporous membrane (Celgard 3501) previously soaked in a saturated NaClO 4 (aq.) electrolytic solution was sandwiched between two of these as-coated electrodes, and the entire set-up was assembled in a standard capacitor test cell (ECC-std, EL-CELL GmbH) for tight packing and ease of measurement (Fig. 7a).
Cyclic voltammetry and galvanostatic charge-discharge tests. The predominant charge storage mechanism of the Bi 13 S 18 I 2 -based supercapacitor was investigated through cyclic voltammetry (CV) scans and galvanostatic charge-discharge cycles. The CV tests were performed in a voltage window ranging from 0.0 V to 0.6 V, with scan rates varying from 500 mV s À1 to 10 mV s À1 (Fig. 7c). No oxidation or reduction peaks were observed in this window, with the lowest 10 mV s À1 scan rate displaying a rounded-rectangular shape indicative of a predominantly non-faradaic charge storage mechanism. We note that the potential values recorded for this two-electrode device are different from the three-electrode CV measurement above, where a separate reference electrode was used. For the two-electrode measurements we carried out on the supercapacitor device, we restricted the scans to a range where we saw no faradaic peaks to ensure we are only probing EDLC-type capacitance behaviour. The non-symmetry of the CV curves at the higher scan rates are attributed to leakage current during the ultra-fast charging process and should not be attributed to a faradaic process. Such non-faradaic behaviour is characteristic of electric double layer capacitors, in contrast with socalled pseudocapacitors or supercapacitors, which rely on strong faradaic reactions to store charge. 35 The areal and specic capacitances of the assembled supercapacitor device were calculated from the CV curves according to the following equations: 43 where DV is the voltage window of the negative (cathodic) current (V), s is the constant scan rate (V s À1 ), A is the average geometric area of the two electrodes (cm 2 ), and m is the combined mass of the active material on both electrodes (g). The device's areal and specic capacitances as calculated from  the CV curves spread from 53.65 mF cm À2 to 6.70 mF cm À2 and from 3.35 F g À1 to 0.42 F g À1 for scan rates increasing from 0.01 V s À1 to 0.5 V s À1 . The decrease in capacitance with increasing scan rate is a common feature of EDLCs and is caused by the different time regimes of charge transport and ion diffusion for the varying scan rates. At lower scan rates, electrolytic ions have sufficient time to diffuse into the pores of the Bi 13 S 18 I 2 active layer, increasing the charge accumulation and thus the capacitance. At higher scan rates, charge accumulation is conned to the surface of the electrodes, decreasing the electrodes' capacitances.
The galvanostatic charge-discharge measurements were similarly carried out over a 0.0-0.6 V window, with current densities varying from 2.0 mA cm À2 to 0.1 mA cm À2 (Fig. 7d). The charge-discharge curves are largely symmetric and the discharge curve displays remarkable linearity following an initial iR drop even at the lowest current densities, corroborating the non-faradaic, EDLC behaviour of the Bi 13 S 18 I 2 supercapacitor. The areal and specic capacitances were also calculated from the slope of the discharge curves in their linear regimes according to: where i is the constant current (A), dV dt is the slope of the discharge curve taken in the voltage range 0.0-0.1 V for consistency, and m and A retain their previously dened meanings. The device's areal and specic capacitances derived from the galvanostatic charge-discharge curves ranged from 105.34 mF cm À2 to 57.83 mF cm À2 and from 6.58 F g À1 to 3.61 F g À1 for current densities increasing from 0.1 mA cm À2 to 2 mA cm À2 . Pious and colleagues previously integrated (CH 3 NH 2 ) 3 -Bi 2 I 9 in a similar symmetric EDLC device and reported an electrode areal capacitance of 5.5 mF cm À2 . 20 Since our supercapacitor device employs two symmetrical electrodes in series, the electrode capacitance is equal to twice the device capacitance, or 210.68 mF cm À2 for our best trial at the lowest tested current density. The energy density and power density calculated from CV and galvanostatic charge-discharge measurements are summarized in Table S2, † and the energy density can be potentially increased by operating in a larger potential window. Control experiments were carried out showing that the capacitance arises from Bi 13 S 18 I 2 rather than the carbon cloth substrate or additives (Fig. S5, Table S3 †). In addition, an electrode prepared from Bi 13 S 18 I 2 powder suspension was also tested, showing only about 1/10 capacitance compared to solution-processed Bi 13 S 18 I 2 electrode (Fig. S6, Table S4 †). This is caused by inferior surface area of the active material Bi 13 S 18 I 2 in powder form as shown in the SEM images (Fig. S7 †).
The long-term cycle stability of the Bi 13 S 18 I 2 electrodes was evaluated by performing 5000 sequential galvanostatic chargedischarge cycles at the high current density of 2 mA cm À2 (Fig. 7e). Even aer 5000 cycles, the device retains 99.7% of its initial capacitance, demonstrating the electrodes' remarkable stability under intense electrical conditions. Moreover, SEM images of electrodes aer undergoing the 5000 cycles indicate that the surface coverage of the carbon bres with the activated Bi 13 S 18 I 2 remained largely the same aer cycling (Fig. S8 †). We note however that limited information can be gleaned from the SEM images, as the electrodes become heavily coated and obscured with crystallized NaClO 4 aer use in the supercapacitor device. Still, the FTIR and Raman spectra of the electrodes aer cycling display the same vibrational peaks as those of the non-tested electrodes with no signicant shis in the peak wavelengths or Raman shis (Fig. S9 †), indicating that the Bi 13 S 18 I 2 material underwent no systematic chemical degradation during the 5000 cycles. The drastically reduced intensity of the Raman spectrum for the cycled-electrodes is likely due to the thick NaClO 4 coating concealing the active layer. Finally, XRD was utilized to evaluate the crystal structural integrity of Bi 13 S 18 I 2 over the 5000 cycles. Fig. 6 compares the electrodes' XRD pattern both before and aer the 5000 cycles with the XRD peaks calculated for the pure Bi 13 S 18 I 2 powder.
Although the intensity of Bi 13 S 18 I 2 's XRD peaks also drop aer cycling, the persistent peaks near 24 , 28.5 , 30 , and 32 demonstrate that the Bi 13 S 18 I 2 largely retains its crystal morphology throughout the electrochemical cycling. The extra, sharp peaks are attributed to crystalline NaClO 4 and exposed carbon cloth bres.
The high capacitance performance is attributed to a combination of reasons. Firstly, the X-ray structure of Bi 13 S 18 I 2 indicates the presence of both Bi(III) and Bi(II) ions. This illustrates an ability of the Bi centre to accommodate varying charge which may explain the capacitance mechanism and would be of great interest to investigate further. The signicance of Bi 2+ is supported by the calculated band structure that was previously reported 41 and showed the lowest unoccupied states to be predominantly derived from the Bi 2+ centres along with contributions from the immediately-adjacent Bi 3+ and sulfur ions. As would be expected, the highest occupied states were predominantly iodide in character. Both the valence band and conduction band showed notable dispersion, at least along some directions, pointing towards an ability to delocalise and hence accommodate the added charge. Secondly, enhanced active material coverage and homogeneity have been achieved from a new solution-deposition method, which should facilitate the electrolyte diffusion into the pores. This process is also favourable for large scale solution-based manufacturing, and using exible carbon bre as substrate provides the possibility for roll-to-roll production methods. Finally, pH neutral NaClO 4 aqueous solution was used as the electrolyte, which has proven to be economic, eco-friendly and electrochemically superior with a wide potential window. 44 Electrochemical impedance spectroscopy. Electrochemical impedance spectroscopy (EIS) was performed to characterize the physical and electrochemical processes of the Bi 13 S 18 I 2 supercapacitor in response to AC current at varying frequencies.
All of the EIS measurements were conducted with a baseline potential of 0 V and an AC amplitude of 10 mV at frequencies ranging from 150 kHz to 0.1 Hz. Fig. 7f displays the imaginary (out-of-phase) impedance versus the real (in-phase) impedance of the Bi 13 S 18 I 2 supercapacitor in a Nyquist plot prior to the 5000 galvanostatic charge-discharge cycles. From the Nyquist plot and the Bode plot (Fig. S10 †), a modied Randles equivalent circuit was tted and shown as an inset in the Nyquist plot to analyse the resistive and capacitive elements in the supercapacitor.
The Nyquist plot shows typical EDLC-type behaviour with a semi-circle in the high-frequency regime and a linear branch in the low-frequency region. The intercept of the high-frequency end of the semi-circle with the real (Z 0 ) axis gives the value of R 1 in the equivalent circuit, representing the combined resistance due to the aqueous NaClO 4 electrolyte and the resistance in the current collectors, oen referred to as equivalent series resistance (ESR). Since all charge migration during the chargedischarge process occurs through the electrolyte, R 1 is tted in series with the other circuit elements and found to have a value of 0.89 U ( Table 2). The curvature and diameter of the semicircle are simulated with constant phase element Q 1 and resistor R 2 , which are tted in parallel to represent the two possible behaviours of the electrolyte during the charging and discharging of the supercapacitor. At higher frequencies or shorter charging times, the electrolytic ions have insufficient time to penetrate the surface of the Bi 13 S 18 I 2 electrodes, instead forming an electrostatic capacitive layer at the surface of the electrodes highlighted in Fig. 7a. This electrostatic capacitor is modelled with a constant phase element Q 1 with an ideality factor of 0.80, signifying an imperfect capacitor, and a relatively limited Y 0 of 22.8 m(F s À0.20 ) 1.25 . Y 0 is the numerical value of admittance at u ¼ 1 rad s À1 , 43 and Q À Y 0 is given by the Frequency Response Analyser EIS soware as: , u is the angular frequency, and n is the ideality factor. Q À Y 0 can be crudely approximated as the capacitance of the CPE with units of farads, but such false-equivalence has been strongly criticized. 45 Instead, we report Y 0 with its proper units of (F s nÀ1 ) 1/n to avoid confusion, but note that at high n (n z 1), the approximation Y 0 ¼ C becomes increasingly valid. The non-ideality of Q 1 is attributed to surface roughness and uneven coverage of the carbon cloth with the activated Bi 13 S 18 I 2 layer, which is apparent in the SEM images of the coated electrodes (Fig. 7b). At moderate frequencies, however, the ions penetrate the electrodes' active layer, introducing R 2 with a calculated value of 16.8 U, representing the charge transfer resistance and the bulk resistance of the active material and electrode pores. In series with R 2 is a Warburg impedance element W, a characteristic feature of a Randles circuit that simulates the charge transport and mass diffusion of the electrolyte into the activated electrodes at mid to low frequencies. The tted Warburg impedance gives a so-called Warburg coefficient with W À Y 0 giving the admittance of the impedance at an angular frequency of 1 rad s À1 . Our simulated W gives W À Y 0 ¼ 0.45 U À1 s 1/2 . A simple Randles circuit without the addition of Q 2 will produce linear behaviour with a 45 angle with respect to the Z 0 axis, characteristic of the Warburg diffusion element. The Nyquist plot for our experimental data, however, displays linear behaviour with a much steeper angle approaching 81.8 . To model this increased slope, we add a second constant phase element Q 2 in series with R 2 and W with an ideality factor n ¼ 0.95 and Y 0 ¼ 71 m(F s À0.05 ) 1.05 . The high n value suggests that Q 2 acts very nearly as an ideal capacitor, and the Q 2 À Y 0 closely matches the measured device capacitances derived from the CV and galvanostatic charge-discharge curves at lower scan rates or currents. We attribute this second, low-frequency regime capacitance to strong electrostatic charge storage once the electrolytic ions fully diffuse into the pores of the active Bi 13 S 18 I 2 layer. EIS was also used to characterize the stability of the Bi 13 S 18 I 2 supercapacitor aer the 5000 charge-discharge cycles. Fig. S11 † compares the experimental Nyquist plot for the same Bi 13 S 18 I 2 supercapacitor before and aer the cycle stability test was performed. While the low-frequency behaviour remained largely the same aer cycling, R 1 decreased slightly from 0.89 U to 0.81 U, Q 1 À Y 0 decreased from 22.8 m(F s À0.20 ) 1.25 to 13 m(F s À0.15 ) 1.18 , and R 2 increased from 16.8 U to 19.0 U. The slight decrease in the electrolyte resistance R 1 is likely due to surface lm and electrolyte conditioning aer multiple cycles, while the simultaneous decrease in Q 1 -Y 0 and increase in R 2 may be due to mechanical wear, decreased surface area of the active layer, and trapped ions inside the surface pores of the electrodes. The largely unchanged low-frequency behaviour, however, proves the long-term electrochemical stability of the Bi 13 S 18 I 2 active layer, as long-term pore penetration and ion diffusion does not signicantly affect the low-frequency diffusion and capacitive elements.

Conclusions
We have demonstrated the rst facile, low-temperature, and solution-processing synthesis of both powder and thin-lm samples of a relatively-unexplored bismuth chalcohalide material, Bi 13 S 18 I 2 . The optimal ratio for the Bi(xt) 3 and BiI 3 precursors has been studied and established, and the crystal structure of deposited thin lms match the theoretical pattern calculated by powder XRD measurements. The elemental composition of Bi 13 S 18 I 2 has been veried by XPS, and the direct and indirect band gap values (0.68 eV and 0.92 eV, respectively) are characterized via diffuse reectance measurements. The energy level diagram for Bi 13 S 18 I 2 has been estimated based on electrochemistry techniques, using solid state Bi 13 S 18 I 2 on FTO as the working electrode. In an extensive supercapacitor device study, Bi 13 S 18 I 2 has been employed as the active material in an EDLC device, and a simple and economical electrode fabrication process has been demonstrated. Optimization of the active material deposition via solution-processing methodology and choice of electrolyte solution have together enhanced the EDLC performance. Utilizing a high mass loading (7-8 mg cm À2 ) of active material and saturated aqueous sodium perchlorate as the electrolyte, we have achieved a non-faradaic EDLC with a superior areal capacitance of 210.68 mF cm À2 and an excellent 99.7% capacitance retention over 5000 charge-discharge cycles. The EDCL mechanism is supported by the lack of any redox peaks in the cyclic voltammogram and a largely linear galvanostatic discharge curve, suggesting the capacitance derives from the transfer of delocalised electrons, 46 a mechanism which is also consistent with the reported band structure. 41 For the

Material characterizations
Powder X-ray diffraction (PXRD) was performed on a Bruker (D8 Advance) diffractometer with monochromatic Cu-Ka radiation and wavelength of 1.5406Å. The measurements were performed at room temperature over a 2-theta scattering angle of 5-60 , with increments of 0.1 on the 2-theta scale. X-ray photoelectron spectra were acquired in a Thermo Scientic (VG Sigma Probe) XPS spectrometer using monochromatic Al-Ka as the source of X-rays. Scanning electron microscopy was performed by Gylen Odling using a Zeiss (SIGMA HD VP) Field Emission-SEM. Diffuse reectance measurements for powder and thin lm samples of Bi 13 S 18 I 2 were recorded on a Jasco (V-670) spectrophotometer with Spectra Manager soware. The powder sample was prepared by mixing thin lm material with a barium sulphite (BaSO 4 ) matrix in a 5 : 95 weight ratio. The Kubelka-Munk function was used to analyse the diffuse reectance data, and Tauc plots were constructed to obtain the optical E g for direct and indirect transitions. Cyclic voltammetry measurements were performed using a threeelectrode cell with the spin-coated Bi 13 S 18 I 2 on FTO-coated glass on top of a mesoporous TiO 2 layer as the working electrode. The sample was placed in dichloromethane, using tetrabutylammonium hexauorophosphate (0.3 M) as the supporting electrolyte, platinum as a counter electrode, and a Ag/AgCl reference electrode. The cyclic voltammetry was recorded in an Autolab (Type III) potentiostat at linear scan rates of 0.5 V s À1 , 0.2 V s À1 , and 0.1 V s À1 . The samples were cycled 5 times at each sweep speed and the average obtained, with degassing of the solvent under N 2 before each measurement. The redox process of the ferrocenium/ ferrocene couple was recorded to be used as the internal standard. The FT-IR spectra were recorded using Perkin Elmer FT-IR Spectrum 65 range from 400 to 4000 cm À1 , with resolution of 4 cm À1 . Raman spectra were acquired from a Renishaw InVia Raman Microscope with the excitation laser wavelength of 785 nm, with a spectral resolution of roughly 1 cm À1 .

EDLC studies
The precursor solution of Bi 13 S 18 I 2 was made by mixing BiI 3 and Bi(xt) 3 in a molar ratio of 1 : 18 together with activated charcoal powder and PTFE powder in 1 mL DMF as the solvent to make a nal Bi 13 S 18 I 2 : charcoal : PTFE mixture with a weight ratio of 85 : 10 : 5 (total mass: 565.7 mg). The resulting suspended solution was sonicated until homogenized, and around 200 mL was drop-coated onto 1 cm 2 of conductive carbon cloth (ELAT, NuVant Systems Inc.) substrate. Absorbing tissue was placed underneath the carbon cloth to absorb excess solution in order to maximize surface coverage and lm homogeneity of the carbon cloth while reducing the mass loading of the active Bi 13 S 18 I 2 layer. The coated electrodes were then annealed at 150 C for 1 h on a hotplate. The dimension of carbon cloth was measured under an optical microscope, and the mass of active material was determined by subtracting the mass of carbon cloth (density: 13 mg cm À2 , shown on the product manual) from the total mass of the dried electrode (estimated error: AE0.4 mg based on multiple repeat measurements). For the cyclic voltammetry and galvanostatic charge-discharge tests, a total of 16.0 mg of the activated Bi 13 S 18 I 2 layer was loaded onto two carbon cloths (dimensions: 10.2 mm Â 10.9 mm, 9.8 mm Â 10 mm). For the 5000 cycles and EIS testing, a total of 14.7 mg was loaded onto two carbon cloths (dimensions: 11.3 mm Â 11.0 mm, 10.7 mm Â 10.7 mm). When assembling the EDLC, two coated electrodes were separated by a thin polymer separator (Celgard 3501) pre-soaked in a saturated sodium perchlorate aqueous electrolyte solution and then sandwiched in a symmetrically assembled capacitor test cell (ECC-std, EL-CELL GmbH). Electrochemical measurements were carried out by connecting the test cell in a two-electrode conguration to an Autolab potentiostat with FRA2 module using General Purpose Electrochemical System (GPES) and Frequency Response Analyser (FRA) soware. The equivalent circuit modelling was carried out with the FRA soware. The control experiments with electrodes deposited from Bi 13 S 18 I 2 suspension were achieved by dispersing Bi 13 S 18 I 2 : charcoal : PTFE mixture in 1 mL ethanol, with same weight ratio and a total mass of 529.3 mg.

Conflicts of interest
The authors state there are no conicts to declare.
Edinburgh for nancial support. Open data: http://dx.doi.org/ 10.7488/ds/2479. We thank School of Physics and Astronomy in the University of Edinburgh for performing the SEM measurements, with help from Dr Andrew Schoeld. We thank Dr Andrey Gromov for the help of performing Raman measurements. We thank Dr Ronald Brown for help in performing XPS measurements. We thank Prof. Andrew Mount and Dr Dimitrios Kampouris for the fruitful discussion. KA thanks the University of Chicago/University of Edinburgh student exchange scheme for support.