Exploring nature-behaviour relationship of carbon black materials for potassium-ion battery electrodes

Materials: preparation and characterisation

Carbon-based electrodes were prepared following the same procedure for all three carbon black materials investigated in this work, namely C45 (Imerys, Timcal), C65 (Imerys, Timcal), and Super P (Alfa Aesar); their main physical properties have already been reported elsewhere^[45]. Each carbon black source was mixed with a binder, i.e., carboxymethylcellulose (CMC, Diacel) for C45, poly(vinylidene fluoride) (PVDF, Alfa Aesar) for C65 and Super P, at a weight ratio of 90:10; solvents (i.e., MilliQ water for C45 and N-methyl-2-pyrrolidone (Merck) for C65 and Super P) were used in excess in the slurry to obtain a homogeneous paste. This latter, prior to ball milling at 30 Hz for 15 min in a MM400 Retsch ball mill, was doctor bladed on a copper foil at a speed of 50 mm s^-1 by a film applicator (Sheen) at the wet thickness of 200 µm. The drying process was completed overnight in an aspirated fume hood. Later, 15 mm-diameter disks were cut by means of a manual hollow punch, and the obtained samples were dried under vacuum in a Büchi oven for 4 h at 120 °C. Afterwards, they were safely transferred into an argon-filled glove box (MBRAUN MB10 compact, O₂ < 0.5 ppm and H₂O < 0.5 ppm). The electrode mass loading values were, on average, 0.91 mg cm^-2 for C45, 1.01 mg cm^-2 for C65, and 1.11 mg cm^-2 for Super P, respectively.

Field emission scanning electron microscopy (FESEM) was carried out with an Auriga dual-beam FIB-SEM workstation (Zeiss) equipped with a 50 mm² X-Max Silicon Drift Detector (Oxford Instruments) for energy dispersive X-ray (EDX) spectroscopy. FESEM images were obtained both for pristine electrodes and electrodes that cycled for ten cycles at 0.05 A g^-1 and washed in ethylene carbonate:diethyl carbonate (EC:DEC, 50:50 v/v, Solvionic, battery grade) prepared in a glove box.

X-ray diffraction (XRD) patterns were collected at 2θ values ranging between 5° and 90°, with a counting time of 10 s per step and 0.03° step size, using a PANalytical X’Pert (Cu Kα radiation, λ = 0.154187 nm) diffractometer with a 2D solid state detector (PIXcel).

Specific surface areas of carbon black powder were determined through the Brunauer-Emmet-Teller (BET) method within the relative pressure range of 0.1-0.3 on an ASAP 2020 C Micromeritics Instrument. Nitrogen adsorption isotherms were recorded at -196 °C within the relative pressure range of 0-1. Prior to adsorption, samples were placed in the cell and evacuated at 150 °C for 2 h under high vacuum.

X-ray photoelectron spectroscopy (XPS) was carried out by means of a PHI 5000 Versaprobe spectrometer (Physical Electronics, Chanhassen, MN, USA) with a monochromated Al K-alpha line (1,486.6 eV). A double charge compensation system, made by an electron beam and Ar⁺ ions, was applied in order to neutralise non-conductive sample surfaces. Survey and high resolution (HR) scans were acquired by using a circular 100 µm-diameter spot. Multipak version 9.0 dedicated software was employed to perform HR spectra deconvolution procedures, using the Shirley function to subtract background signals and pseudo-Voigt curves to fit the photoelectron peaks. C1s peaks were used as a reference for the binding energy axis calibration by considering the C-C signal at 284.8 eV. Analysis was also conducted on C45, C65, and Super P substrates (with their relative binders, i.e., CMC and PVDF) before testing.

Cells: assembly and characterisation

Half-cells were assembled in coin architecture (LIR2032) inside an argon-filled glove box, with the carbon electrode as the cathode and a potassium metal foil as the anode; this latter was obtained from potassium chunks (Merck). Whatman glass fibre was used as a separator after being soaked with the liquid electrolyte. The latter was a KPF₆ (Merck) 0.80 M solution dissolved in EC:DEC, 50:50 v/v.

Electrochemical measurements were performed with a VSP-3e potentiostat provided by BioLogic. We carried out cyclic voltammetry (CV) at a constant scan rate of 0.1 mV s^-1, while variable-scan CV (VSCV) was performed at increasing scan rates (0.1, 0.2, 0.5, 1, 2, 5, and 10 mV s^-1) prior to 5 CV at 0.1 mV s^-1, at room temperature and between 0.01 and 3 V vs. K⁺/K. Galvanostatic charge and discharge measurements were performed on half-cells at a constant specific current of 0.05 A g^-1 for as many as 500 cycles to test the long cycling performance of the investigated carbon electrodes.

The same coin half-cell configuration specified above was exposed to a rate performance test at the specific current values of 0.05, 0.1, 0.2, 0.4, 0.8, 1, 2.5, and 5 A g^-1 active mass (ten cycles for each value), followed by 110 cycles carried out again at 0.1 A g^-1.

A galvanostatic intermittent titration technique (GITT) was carried out per each carbon-based electrode: prior to ten activation cycles at 0.05 A g^-1, several discharging pulses of 15 min each at 0.025 A g^-1 were followed by 45 min of rest, up to the cut-off voltage of 0.01 V. Once the discharge process was completed, the same protocol was applied for the charge phase, with a positive current up to 3 V. The SEI layer was electrochemically analysed through electrochemical impedance spectroscopy (EIS), recorded on each half-cell prior to galvanostatic cycling and after cycling at 0.05 A g^-1 for one, three, ten, and 20 times. EIS measurements were performed by the VSP-3e potentiostat in the frequency range from 100 kHz to 10 mHz, with a potential amplitude of 10 mV.

Materials characterisation

This work aims to better understand the K-ion storage mechanism into three commercial carbon black materials with different active surface areas, structure disorders, surface functionalities, and types of binder.

In this framework, the first step was intended to differentiate the three conductive carbon black materials in terms of specific surface areas and graphitisation degrees. FESEM micrographs (shown in Figure 1) show the regular and similar morphology of merged carbon nanospheres. More in detail, C45-based electrodes displayed a more homogeneous distribution of the binder between particles. Conversely, C65 and Super P nanoparticles exhibited some areas with undispersed PVDF droplets (highlighted by orange circles in Figure 1). The more uniform structure of C45 samples can be explained by the presence of negatively charged carboxyl groups owned by the active material^[45], which facilitate the binding process with the sodium CMC binder. Furthermore, during the slurry preparation and the casting process, the C45 mixture showed better dispersion in the solvents with respect to C65 and Super P. Indeed, the latter two were more difficult to manage, and this is probably due to the different electrical charge densities on the surfaces of the three carbon materials particles^[45]: electrostatic repulsion forces present on Super P and C65 surfaces resulted in a worse compaction behaviour and particle dispersion in the binder; thus, it caused the formation of agglomerates. The different surface functionalities of the carbon black and the consequent variations in particle compaction within the electrode can inevitably influence the intercalation of the alkaline ions inside the materials, along with the surface charge accumulation.

Figure 1. FESEM images of pristine electrode samples. PVDF droplets are highlighted with orange circles.

In order to better study the chemical-physical properties of the three different carbon black materials, N₂ adsorption/desorption and XRD analyses were performed and are reported in Figure 2. The XRD pattern comparison of different powders is shown in Figure 2A. For all three samples, two broad diffraction peaks are distinguishable and centred at 25.3° and 43.3°, respectively. The main peak centred at 25.3°, indexed to the (002) plane, is attributable to graphite-like crystallites, while the second peak (at 43.3°) reveals the presence of graphene layers (indexed to the (101) plane)^[46]. Nevertheless, due to the broadness of the peak, it can be assessed that all the samples exhibit a low graphitisation degree and a pseudo-amorphous structure. Not only does the poor stacking of graphene layers differ from the usual graphite-like structure, but it also results in a higher interlayer distance. Indeed, the displacement between reflection sites, denoted as d in Bragg’s law (see Supplementary Note 1), was found to be equal to 0.352 ± 0.003 nm for all three samples. The bigger distance with respect to the common graphite value (i.e., 0.335 nm) is beneficial for the reversible insertion of the wider K-ions, preventing electrode pulverisation^[47]. Figure 2 also shows N₂ adsorption/desorption isothermal curves (panel b) and the cumulative pore volume (panel c). All three samples show a similar pore size distribution, with a huge amount of mesopores and a negligible amount of micropores. Adopting the IUPAC classification, all three carbon materials show high mesopores density from 2 to 20 nm and mesopores from 30 to 50 nm. According to the pore size distribution, the resulting BET-specific surface area is comparable for C65 and Super P (61.61 and 60 m² g^-1, respectively, for C65 and Super P), while it is lower for C45 (45.99 m² g^-1).

Figure 2. (A) XRD patterns; (B) N₂ adsorption/desorption isothermal curves; and (C) cumulative pore volume curves of Super P (black line), C65 (red line), and C45 (blue line).

Potassium ions storage mechanism

Electrochemical measurements were performed to correlate the morphological features of the carbon black materials with their K-ion storage mechanism. In particular, CV curves allow us to better identify the voltage onsets for electrochemical reactions and the relative potential range for different storage mechanisms occurring in the cell. Namely, carbonaceous anodes are involved in three main electrochemical phenomena: (i) the SEI formation during the very initial cycles; (ii) the ion adsorption onto the electrode surface; and (iii) the insertion/extraction storage mechanism in the bulk phase. As clearly visible in Figure 3, during the first CV cycle, a broad reduction peak from 1.2/1.0 to 0.45 V indicates the first phase of SEI formation. This indexing is also confirmed by the work carried out by Wu et al. and Li et al. on potassium-based systems working with the same electrolyte^[48,49]. Furthermore, it is interesting to notice from Figure 3 that the area ascribed to the irreversible reactions is greater for Super P compared to C65, but what is even more unique is how negligible the SEI formation peak is in the C45-related current-potential graph. As a matter of fact, calculating and subtracting the areas under the CV curves of the 1st and 2nd cycle per each material results that Super P, C65, and C45 systems lost - in this potential range - 28%, 25%, and 6% of their initial capacity, respectively. Moving our focus on the previously mentioned two storage mechanisms, all three carbon materials exhibited, from the 2nd CV cycle, two broad cathodic peaks: the first one, from 0.01 to 0.45 V, mainly described the diffusion and insertion of K-ions, while the second one, comprised between 0.45 and 1.2 V, referred to all the surface-driven reactions. Since the redox potential of the K⁺/K couple is very close to that of Li⁺/Li (i.e., -2.93 V vs. SHE and -3.04 V vs. SHE, respectively), similar potential ranges are expected for the electrochemical reactions. As a matter of fact, in previous works, the Li-ion intercalation into the confined space of the pseudo-graphitic system was scanned between 0.01 and 0.3 V (vs. Li⁺/Li), and it contributed to almost all the specific capacities^[50]. In contrast, in the same potential interval, the bulk insertion mechanism for the present PIB carbon-based electrode led to a broader peak due to the slower and more difficult diffusion of bigger K-ions (i.e., 1.38 Å vs. 0.76 Å for Li⁺). Lastly, for the proposed potassium-based system, the area under the peak comprised between 0.45 and 1.2 V is notably higher than the pseudo-graphitic counterpart; this fact was not a surprise since the disordered structure and large defect area of carbons with a low graphitisation degree offer favourable kinetics for surface-dominated storage processes^[51].

Figure 3. CV traces at the constant scan rate of 0.1 mV s^-1 for six cycles for (A) Super P, (B) C65, and (C) C45 electrodes. Dashed green lines separate two voltage ranges (i.e., from 0.01 to 0.45 V and from 0.45 to 1.2 V), where SEI formation starts occurring and the ion insertion is more prominent.

Moving forward to the anodic scan, differences in the depotassiation mechanism arise. C65 [Figure 3B] worked mainly on a single oxidation peak centred at around 0.6 V, the intensity of which diminished through scanning cycles; a less visible peak in the lower potential region started increasing from the 2nd cycle. The same behaviour was observed for the C45 cell, excluding the peak around 0.25 V, which is more prominent. The two oxidative peaks approximately between 0 and 0.45 V and between 0.45 and 0.75 V may refer to the ion bulk deintercalation and the surface ion release, respectively. This assignment of the potential windows to the two depotassiation mechanisms will also be confirmed - in the forthcoming sections - through the results of Dunn’s method. The dynamic profile of C65 and C45 can be translated as the formation through the cycles of a more stable and conductive SEI layer, which created - at the same time - preferential pathways for the accommodation of K-ions in graphite interlayer space. This was witnessed by the arising of the related peak within 0-0.45 V, ascribed to the deintercalation. On the other hand, in the Super P profile, the oxidative pair became a single wide peak comprising both the depotassiation behaviours. Concerning the SEI layer, according to the work made by Xu et al., the binder choice determines the starting potential value of the electrolyte reductions and the products, which will constitute the SEI layer^[51]. The poor electrochemical stability of PVDF, together with its stronger interaction with the EC solvent, causes the triggering of the decomposition reactions at higher potential values, acting as a catalyst for the electrolyte decomposition and, thus, causing a higher irreversible capacity amount in the upper voltage range. As also suggested from the magnitude of the CV area ascribed to irreversible capacity, galvanostatic cycling results of the three carbon electrodes confirmed that the capacity lost in the upper voltage region (from 1.2 to 0.45 V) was notably higher in the PVDF-based systems. As a matter of fact, the discharge capacity loss during the galvanostatic cycling in the 1st cycle in the upper voltage region represented only 10% of the total C45 system loss, while 22% and 29% of the total irreversible capacity were calculated for C65 and Super P samples, respectively.

The voltage vs. specific discharge capacity profiles from the galvanostatic cycling performance at 0.05 A g^-1 in Figure 4 and Supplementary Figure 1 are in accordance with the concepts detailed above since - in the first cathodic polarisation - surface dominated reactions (e.g., SEI formation) started at 1.2 V for all three systems, while more extreme SEI formation was witnessed by a very large plateau starting around 0.8 V for Super P; a similar but shorter plateau was observed for C65, while no plateau at all occurred for C45. The latter observation confirmed that 10% of the total capacity was lost by the C45 sample during the first voltage region. A second plateau started at 0.3 V for all the voltage profiles, pointing at the K-ion intercalation between the graphene layers. The presence of a small slope for these curves (instead of a perfect plateau) was indicative of a very slow insertion process. From the second cycle, two different tilts were present: a near-slope from 1.2 to 0.45 V was ascribed to the presence also of surface-driven storage contribution, and a more horizontal trend below 0.45 V corroborated the dominance of K-ion insertion. The disappearance of the first plateau and its replacement by the near-slope (1.2-0.45 V) in the voltage profiles of C65 and Super P from the 2nd cycle validated the theory, according to which most of the surface reactions of the first cycle are ascribed to irreversible reactions. The differential capacity analysis, as shown in Figure 5, clearly confirmed the occurrence of an electrolyte reduction on the pristine electrode (starting from 1.2 V) only for the PVDF-based electrode. Indeed, in Figure 5C, most of the irreversible capacity is actually ascribed to the low potential region. As a matter of proof, the shaded area in Figure 4C is meant to visually confirm that the CMC-based system lost irreversible capacity mainly through thermodynamically accelerated side reactions in the low voltage region, while in Figure 4A and B, it emerges that irreversible reactions occurred throughout all the potential window for Super P and C65 samples.

Figure 4. Voltage profile of charge and discharge steps at the first and second cycle at 0.05 A g^-1 for (A) Super P, (B) C65, and (C) C45 electrodes. The filled areas are meant to better visualise the potential profile changes and represent the lost charge and discharge energies between the first and second cycles.

Figure 5. Differential capacity analysis of the potassiation and depotassiation of the first two cycles for (A) Super P, (B) C65, and (C) C45 electrodes.

Our evidence of the surface-driven storage mechanism has led us to stress that the spread of pseudo-amorphous and nanoparticulate electrodes calls to consider the concept of pseudocapacitive materials. Indeed, there are two different kinds of storage mechanisms that can occur in the electrode material: (i) Faradaic (electron-transfer) processes or (ii) ion accumulations at the electrical double layer. The two mechanisms can be differentiated by observing their cyclic voltammograms and galvanostatic voltage profiles. The capacitor response (accumulation of ion) will give linear potential dependency with time during galvanostatic charge/discharge, along with rectangular CV traces. Battery-like materials, instead, will display their Faradaic processes through plateaus and intense peaks in voltage profiles and voltammograms, respectively. The divergence from both abovementioned behaviours comes from a pseudocapacitive material response. In this latter, the charge storage is entirely Faradaic as in all battery-like materials, but they will mainly store charges through redox reactions on their surface instead of the inner regions, sharing similar electrochemical responses to capacitors. In this regard, CVs and potential profiles of the materials analysed in this work are the litmus test of pseudocapacitive behaviour^[52]. Indeed, given their pseudocapacitive responses (i.e., near-sloped potential profiles and symmetrical broad redox peaks in CV traces), further electrochemical analyses are worthy of being carried out.

A useful way to try getting insights into these “in-between” (promiscuous) materials is to analyse their current responses in CV at increasing scan rates (i.e., by VSCV) according to Dunn’s and Trasatti’s methods. Inspired by Cottrell’s equation^[52], Dunn and Trasatti stated that the current response at the voltage variation in CV is proportional to the square root of the voltage scan rate in diffusion-controlled processes, while the capacitive current response results proportional to the scan rate. For those materials with behaviour being indistinguishable among capacitor-like or battery-like, a current response that is the sum of the capacitive and the diffusion currents is considered:

Relevant consideration can already be drawn from the VSCV spectra shown in Figure 6: for all three carbon-based materials, the increase of the scan rate provokes the current response increase, but the presence of diffusion-controlled storage mechanism induces the escalating current peak shift, too. Based on the previous considerations, it can be inferred that both capacitive and diffusion-controlled capacities are involved. However, Trasatti’s and Dunn’s methods are indispensable for differentiating and quantifying the presence and entity of each of them. Trasatti’s method results suggested that Super P owned the highest amount of pseudocapacitive capacity (i.e., 65.5% of the total), followed by C45 (63%) and C65 (56%). Notably, in Figure 6, C65 shows the widest peak shift, while for Super P, it is the narrowest; therefore, it should not be surprising that they showed the highest and the lowest presence of diffusive currents, respectively. On the other hand, Dunn’s method - applied on the CV carried out under a 0.2 mV s^-1 scan rate - matches similar percentages [Figure 7], but it also identifies the voltage range in which the diffusion or charge accumulation is occurring. The capacitive contributions represented by the filled areas in Figure 7 confirm that the charge capacity values provided between 0.01 and 0.5 V are almost entirely ascribed to diffusion-controlled deintercalation. Meanwhile, from 0.5 to 3 V, only surface-driven reversible reactions contribute to the charge capacity. Regarding the discharging process, once the SEI layer has formed during the previous five activation CV cycles, almost exclusively K-ion insertion occurred from 1.5 to 0.9 V; from 0.9 V, the surface-driven reaction increased up to dominate at 0.5 V; instead, from 0.5 V down to 0.01 V, both of the two mechanisms occurred in equal amounts. It is also worth noticing that since diffusion-controlled storage is a kinetic-dependent process, as it is influenced by the applied scan rate, it is also determined by the applied current. As a matter of fact, Supplementary Figure 2 shows that the peak shift of the differential capacity curves at increasing specific currents of the rate performance confirms the potential range of the diffusion-driven storage again. Here, the arrows in the insets are meant to trace the intercalation peak centre shift from 1 to 0.5 V and the deintercalation peak from 0.25 to 0.6 V.

Figure 6. VSCV traces for (A) Super P, (B) C65, and (C) C45. The visible current shift in C65, due to diffusion-controlled storage processes, is encircled in blue, while the perfect rectangular CV shape related to the capacitive behaviour is framed in magenta.

Figure 7. Dunn’s method applied to the CV collected at 0.2 mV s^-1 for (A) Super P, (B) C65, and (C) C45. Separation of capacitive and diffusive contributions is highlighted by the filled areas.

To further validate the proposed voltage assignment to the diffusive and deposition mechanisms, the total current derived from Eq. 1 is hereby described by the following empirical equation (and by its logarithmic form):

Let v represent the voltage scan rate, with a and b as adjustable parameters. A capacitor will have the b parameter equal to 1, indicating the occurrence of surface-driven processes only, while in battery-like electrodes, diffusion-controlled currents will lead to a b value of 0.5. Electrode materials with b values comprised between 0.5 and 1 electrochemically behave neither totally capacitive nor battery-like. As shown in Figure 8, the b values, calculated for different voltages of the reduction branch, depict a behaviour that varies among the explored potential windows, describing the succession of reduction reactions occurring on the outer and the inner surfaces: the decrease of the b values starting from 1.2 V denotes the occurrence of diffusion-driven reductions, reaching the higher dominance at 0.8 V, where all these three carbon-based electrodes show b values of 0.5-0.55. Moving from this minimum value, the occurrence of surface-driven processes increases the b value up to its maximum at 0.3 V, where the capacitive current represents more than half of the total current. At 0.3 V, the C45 system reached the highest b value of 0.9, as confirmed by Dunn’s profile, where at around 0.3 V, the area under the cathodic curve is completely filled in blue [Figure 7C]. Moreover, the graph shows a further decrease during the final intercalation process through the inner region between 0.3 and 0.01 V. Instead, the b value for the C45 electrode at 0.01 V is still high, in accordance with Dunn’s results.

Figure 8. Cathodic b values obtained for the three carbon black materials.

The two diverse storage mechanisms can also be detected through their effect on the ion diffusion through the electrode thickness when the potential changes. The difficulty that an ion faces crossing interfaces and diffusing along intricate pathways can be quantified by the diffusion coefficient (DC), calculated as follows:

where τ is the time interval of the charge or discharge pulses, m_act is the net electrode active mass, S is the geometric area of the electrode, M_c is the molar mass of the carbon black, and V_c is its molar volume^[49,53]. Their ratio $$\frac{M_{C}}{V_{C}}$$is equal to the carbon black density (i.e., ρ), which is 2.2 g cm^-3 for Super P^[54] and 1.86 g cm^-3 for both C45 and C65 electrodes^[45]. The meaning of all the parameters is visually explained in Supplementary Figure 3. The coefficients are extracted from the electrochemical results of the cells that have already been activated by ten galvanostatic cycles at 0.05 A g^-1; thus, the SEI layer is supposed to have already been formed. Figure 9A shows the overall decreasing discharge diffusion coefficients with the decrease of the cell voltage. The initial DC values are high since, at the beginning of the discharge, ions are adsorbed on the carbon electrode surfaces, free of K-ions. As soon as the surface is covered with K-ions, the DC starts decreasing. When the intercalation process starts occurring for all three carbon electrodes, the DC keeps decreasing as the storage volume between the layer planes is filled by K-ions. However, it is interesting to notice a slight increase of the diffusion coefficient of C65 and, more distinctly, of C45 at 1.5/1.3 V, where the ion insertion starts occurring. Here, the better SEI layer allows a boosted diffusion of the ions through it. Opposed to C65 and C45 trends, Super P shows a continuously decreasing diffusion coefficient. The same understanding can be applied to the evolution of charge diffusion [Figure 9B], where the very first K-ions are easily removed from the carbon electrode outer surface with a very high DC. As the outer ions keep decreasing, the inner intercalated ones encounter more difficulty in leaving the interlayer space of the electrode: this is translated into a DC decrease up to 0.45 V. Moving from 0.45 to 3 V, desorption of ions from the electrode surface starts occurring. Given the high concentration of ions deposited onto the carbon surfaces, from 0.45 V, all three electrochemical systems experience a DC increase that persists as long as the ion concentration on their surface becomes too low and their pathways too intricate. The evolution pattern was the same for all three electrodes, but C45 showed every time a higher DC, as witnessing a more kinetically efficient process, such as the initial intercalation from 1.3 V during the discharge or the faster desorption from 0.45 V during the charge.

Figure 9. Diffusion coefficient at different voltage values in the (A) discharge and (B) charge steps.

These varied storage kinetics, also influenced by the electrolyte and binder choice, resulted in different electrochemical responses. Figure 10 shows the current rate (panel a) and long-cycling (panel b) performances of the three investigated systems. The irreversibility trends displayed by CV traces and differential capacity curves are confirmed by the capacity loss of the three carbon electrodes during the first cycles of their rate performance profile [Figure 10D]. As a matter of fact, Super P lost - during the first discharge of the rate performance - 806 mAh g^-1, which is almost twice the values of C65 and C45 (i.e., 460 and 300 mAh g^-1, respectively). Sharing a similar mass loading among all three electrodes, both rate performance and the long-cycling profiles overall show higher specific capacity values for the C45 system, together with a better capacity retention ability. Indeed, after 500 cycles of the long cycling performances, the C45 system retained 64% of the discharge capacity provided at the 10th cycle, compared with 57% and 56% of C65 and Super P, respectively. Furthermore, it is also vital to consider how much of the K-ions adsorbed and intercalated during the discharge were reversibly recovered during the charge: this is expressed by the Coulombic efficiency (CE) values shown in Figure 10C, where the better reversibility of the ion shuttling process in C45 is easily noticeable. The graph suggests that, during the very first cycles, the CMC-based system reached better reversibility of the potassiation/depotassiation mechanism; yet, after 70 cycles, the CE for C65-based cells exceeded the C45 ones, remaining always higher than the latter. Indeed, it seems that C45 is able to form a better SEI layer during the initial cycles, but something keeps decomposing onto the C45 SEI since its CE never reaches 99%. Keeping in mind that the SEI layer can enhance the ions conduction and block the electron transfer, it can allow the system to address more challenging (stressful) applied currents. As a matter of fact, from Figure 10A, the undeniable superiority of C45 and C65 in meeting increasing current demand provides satisfying energy densities. On the other side, for the highest two imposed current rates, Super P-based cells delivered almost null specific capacity. With the purpose of better visualising this divergence, the charge and discharge potential curves of the second-to-last cycle of the tens of each current rate are displayed, together with the related capacity retention percentage calculated with respect to the 9th cycle of the first current rate (i.e., 0.05 A g^-1), as shown in Supplementary Figure 4.

Figure 10. (A) Rate performances, (B) long cycling performances, and (C) Coulombic efficiency values during the long cycling test for half-cells assembled with the three different carbon-based materials. (D) Cumulated irreversible capacity (in mAh g^-1) lost by the three systems during the first five cycles of the rate performance test. Each bar (with its reported value) represents the capacity lost between one cycle and the subsequent. The entire stack with the value in green displays the total capacity lost after five cycles.

As a concluding remark of this electrochemical section, it must be stated that - besides the different chemical nature of the three investigated carbon-based electrodes - the binder also plays a reasonable role in affecting the performance and stability of the cells. In this article, CMC has been used with C45, while PVDF has been chosen for C65 and Super P systems; this choice is motivated by state-of-art protocols adopted by the scientific community, where the best electrode fabrication processes have already been optimised considering the best compatibility among active material, binder, and solvents. Therefore, we have not explored C45:PVDF, C65:CMC, and Super P:CMC systems since they would not have resulted in uniform electrodes, leading to unreliable and non-reproducible electrochemical performance.

SEI layer characterisation

Reversible redox reactions are not the only processes occurring within the anode and cathode potential window. As known, indeed, irreversible side reactions lead to the formation of the so-called SEI layer, a passivation film made of decomposition reaction precipitates, which is responsible for the irreversible capacity, decrease of mass transport of the electrolyte liquid phase, and increase of the electrolyte resistance detected during the initial cycles. Nevertheless, this passivation layer has the potential to even boost the electrochemical performances of the storage system, acting as a protective blanket on the electrode that, thanks to its mechanical flexibility, prevents further electrolyte decomposition and possible active material exfoliation and pulverisation. The SEI layer is made of products coming from unavoidable reduction reactions of the solvent and anions constituting the electrolyte; thus, its morphology, composition, and electrochemical features are determined by the chosen electrolyte solution. Low molecular weight organic solvents are the most used; in particular, ethylene carbonate (EC) provides high polarity and high dielectric constant, promoting salt dissolution; however, given its high viscosity and high melting point, it is thinned with a low viscosity solvent, which promotes the ion mobility, such as diethylene carbonate (DEC). As regards the salt, KPF₆ is probably the most widely used compound for PIB electrolytes, featuring a well-balanced mix of electrochemical responses. A homogeneous, stable, and flexible SEI layer with negligible electronic conduction and low diffusion of the liquid phase, but high selectivity and permeability of K-ions can actually improve the cell performances and lifespan. The achievement of this result depends on the SEI layer composition, which is made by organic and inorganic products.

From the literature on LIBs^[55-59], the formation of a double-layered SEI structure has been widely proved. This structure consists of an inner dense layer comprising insoluble and stable inorganic compounds and an outer part, which is mainly organic, soluble, and metastable. Presumably, the inorganic products of the system investigated in this study will be K₂CO₃, KF, K₂O, and phosphorous oxides (-PO_x); meanwhile, a bigger range of organic compounds are possibly found, such as oligomers [e.g., poly(ethylene oxide) (PEO)-like] and potassium alkyl alkoxides and carbonates (ROK and ROCO₂K, where R is a low molecular weight alkyl group). XPS techniques were adopted to study the three electrodes before and after ten galvanostatic cycles at 0.05 A g^-1 to shed light on the chemical composition of the SEI layers and relative differences among them, also correlating half-cells electrochemical performances to SEI layer structures and elemental composition (see Supplementary Note 2). Starting from the analysis performed on the bare carbon electrodes before being tested, we can infer the differences in the C1s peak (see Figure 11A) among them, especially related to the region between 286 and 290 eV, where most of the carbon- and oxygen-related bonds are located. When we deconvoluted the three curves, we obtained the data shown in Table 1 and Figure 11A. It can be noted that Super P was the one with the lowest amount of C-OH bonds, while the C45 sample was that with the highest C=O bond content^[60]. The more ordered carbonaceous structure belonged to the C65 electrode, which showed a sharper peak in the C-C component (at 284.8 eV), with a low full width at a half maximum (FWHM) value (equal to 0.92 eV). The increase in the FWHM value (equal to 1.2 eV for C45) is attributed to an increase in the disorder of the carbonaceous material arising from the peripheral bonds between carbon and oxygen. These data are in accordance with the results shown by Spahr et al., obtained through time-of-flight secondary ion mass spectrometry measurements^[45].

Figure 11. XPS C1s and K2p HR regions for as-prepared electrodes (A) and tested electrodes (B).

Starting from C1s and K2p HR regions of tested electrodes [Figure 11B], which overlap each other, we can see that there are differences both in carbon- and potassium-related peaks. If we want to delve into more details and correlate our deconvoluted data with SEI layer composition, we have to check for “fingerprint” chemical shifts that can be assigned to specific molecules or compounds. In particular, if we want to consider the inorganic phases, we have to look at the K2p, P2p, and F1s regions (see Figures 11 and 12). According to Caracciolo et al., we fitted the K2p doublet (K2p_3/2 and K2p_1/2) with three couple of curves: one related to K₂CO₃, one to KF, and the latter to the remnant due to KPF₆, at higher binding energy^[60]. Notably, insoluble inorganic compounds, such as K₂CO₃ and KF, are widely known to effectively passivate and protect the active surface from contact with fresh electrolytes, avoiding further decomposition^[48,51,61]. Probably, this is due to their insolubility in the solvent, which ensures they remain on the electrode, protecting its surface. Furthermore, they are supposed to provide good ion conductivity and effectively block electron transfer, thereby improving the electrochemical response at higher current rates^[62].

Figure 12. XPS HR spectra for tested electrodes, highlighting (A) F1s and (B) P2p regions. Intensity has been normalised in each graph to better compare signals.

For the purpose of pushing the investigation of the SEI layer structure, it is necessary to understand the reaction chain occurring within the electrochemical system. The EC is the first component that starts decomposing, namely at 1.2 V in PVDF-based systems or at 1 V in CMC-based ones (see CV traces in Figure 3), having the lowest unoccupied molecular orbital higher than that of KPF₆ and precipitating together with K-ions on the electrode surface as K₂CO₃, constituting the inner inorganic layer^[51,59,61]. The higher reduction potential for the EC decomposition with respect to the theoretical one (i.e., 0.9 V) is ascribable to the pseudo-amorphous and disordered structure of the carbonaceous electrode, which offers different electric fields and, thus, covers a broader range of decomposition potentials. Additionally, the aforementioned PVDF-K⁺-EC chemical affinity increases the EC reduction activity in PVDF-based systems up to 1.2 V. The K⁺-coordinated EC molecules can undergo a two-electron reduction reaction that provides intermediate radical [EC]^- anions, which are further decomposed to precipitate mainly as K₂CO₃ and potassium ethylene dicarbonate [PEDC, (CH₂OCO₂K)₂] but also as (CH₂CH₂OCO₂K)₂ or KO(CH₂)₂CO₂(CH₂)₂OCO₂K. The latter are very soluble, and they can also diffuse back from the SEI layer to the electrolyte phase. Simultaneously, coordinated EC molecules also start undergoing one-electron reduction processes to produce more soluble potassium alkyl carbonates such as (CH₂CH₂OCO₂K)₂, PEDC, KO(CH₂)₂CO₂(CH₂)₂OCO₂K, or CH₃OK and KOH. It can be hypothesised that the peaks from 0.9 to 0.45 V present in C65 and Super P dQ/dV curves [Figure 5] and CV traces [Figure 3] are ascribable to EC decomposition into these organic, metastable, and mainly soluble potassium alkyl carbonates, which precipitate upon the previously formed inorganic layer constituting the outer organic SEI layer. However, being metastable and soluble, they will decompose through subsequent cycles, being responsible for the capacity decay and low CE values of the first cycles. Moving toward lower potentials, even DEC starts decomposing, leading to the formation of potassium propionate (CH₂CH₃COOK) and diethyl dioxane dicarboxylate (C₉H₁₅O₄COOK). From ≈ 0.45 V, the most stable and insoluble inorganic species is produced, i.e., KF. Indeed, it is difficult to believe that the SEI layer is made only by one inorganic inner layer covered by an outer organic blanket; it is, instead, more accurate to say that the overall structure will be mosaic-like since the organic, metastable, and soluble species can easily be attacked by more reactive products, resulting in a heterogeneous layer of organic and inorganic products. For example, the reactive and unstable KPF₆ is relevantly decomposed in the most stable KF and PF₅. Being the latter a strong Lewis acid, it will electrophilically react with the solvent carbonyl groups to produce insoluble ether-containing molecules, such as (CH₂CH₂O)₂ and (CH₂OCOOCH₃)₂, or even longer chains (e.g., PEO). However, at the same time, PF₅ can react electrophilically with water traces present in the electrolyte or adsorbed onto the electrodes to produce the highly reactive HF. The latter attacks the organic part of the SEI layer, cleaving the abovementioned oligoether to return smaller soluble organic molecules. Nonetheless, it can also cause beneficial secondary reactions: by decomposing organic potassium alkyl carbonates, additional KF is produced. Finally, PEDC can be interconverted to K₂CO₃ from its reaction with water traces^{[51,59,61,63]}. As reported in Table 1, Super P electrodes are the ones showing the lowest amount of KF and the highest amount of KPF₆ and K₂CO₃. On the other hand, C45 is the one with the highest amount of KF and the lowest amount of remnant KPF₆. When examining P2p spectra, it is interesting to highlight that, apart from the P-F component, which is slightly constant for all three samples (with a small increase for sample C45), there is another component related to -PO_x species at lower binding energy^[64], which is the highest for the C65 sample and the lowest for Super P. The P2p signal is quite noisy due to the low signal intensity since the relative atomic concentration of phosphorous in all the samples is around 2 at%. Among all the organic phases, we can distinguish PEO from the C-OH component, which is the highest in C45. We can also attribute the C=O component to PEDC, and this is higher in Super P and C45 samples.

To draw some final insights from the XPS analysis, it can be said that the very first cathodic polarisation of C65 and Super P samples experienced a relevant number of irreversible reactions, as inferred from dQ/dV curves [Figure 5], which - according to our suppositions - are mainly related to the one and two-electron reduction processes of the [EC]^- anions, which produce a high number of potassium alkyl carbonates that can diffuse back in the electrolyte or decompose again as K₂CO₃. As a matter of fact, PEDC presence is higher in Super P and C45 samples, but, at the same time, there is a higher amount of produced K₂CO₃ in Super P (70.1%) with respect to C65 (55.4%) and C45 (50.0%) (see Table 1). Thus, it can be speculated that C65 had produced a relevant amount of PEDC during the first cathodic reduction through the voltage window from 0.9 to 0.5 V, but at lower potentials most of this potassium alkyl carbonate has been decomposed into K₂CO₃. Instead, for the Super P system, the yield of organic compounds is so high that, even after secondary decompositions, this system holds the highest quantity of both PEDC and K₂CO₃. As regards the C45 sample, given its different chemical nature, potassium alkyl carbonates have been produced mainly at potentials lower than 0.5 V and lately decomposed into K₂CO₃ up to a quantity similar to that observed for C65. Furthermore, as suggested by the absence of residual KPF₆ salt on the C45 sample, more KF and PF₅ are formed, leading to further KF and relevant presence of PEO. Finally, considering the beneficial presence of the ion-conducting and electron-blocking K₂CO₃ for both C65 and C45 samples, as well as the increasing production of the effectively passivating KF from C65 (42.9%) to C45 (49.6%), the superior electrochemical rate performance of C65 and C45 is well substantiated. Instead, the presence of PEO, which is ion-conducting and can act as a binder for the heterogeneous organic and inorganic blended SEI layer, ensures both a higher diffusion coefficient and a flexible and elastic protective SEI layer for the C45 system. The decrease of the interfacial resistance ascribed to the conductive SEI layer is also witnessed by the decrease of the semicircle diameter upon charge/discharge cycles; this occurs for all three electrodes, the Nyquist plots of which are shown in Supplementary Figure 5.

Relative atomic concentrations of C, O, F, and K elements are given in Supplementary Table 1.

FESEM images [Figure 13] of the surfaces of the three samples (relative to samples which underwent ten galvanostatic cycles at 0.05 A g^-1) further corroborate the XPS results. C45 and C65 maintained a disordered and flexible structure, similar to the pristine one (see Figure 1), with a uniform passivation layer. Instead, Super P resulted in a very different structure, more embedded, and with most of the empty volumes filled by decomposition products. This thick and non-homogeneous SEI layer justified the severe initial capacity loss, as well as poor diffusion coefficient and rate performance, but also confirmed the significant presence of the conductive K₂CO₃, which allows a proper specific capacity, at least for lower current rates.

Figure 13. FESEM images of the three carbon materials taken after ten galvanostatic cycles at 0.05 A g^-1.