Synthesis of layered silicon-graphene hetero-structures by wet jet milling for high capacity anodes in Li-ion batteries

While silicon-based negative electrode materials have been extensively studied, to develop high capacity lithium-ion batteries (LIBs), implementing a large-scale production method that can be easily transferred to industry, has been a crucial challenge. Here, a scalable wet-jet milling method was developed to prepare a silicon-graphene hybrid material to be used as negative electrode in LIBs. This synthesized composite, when used as an anode in lithium cells, demonstrated high Li ion storage capacity, long cycling stability and high-rate capability. In particular, the electrode exhibited a reversible discharge capacity exceeding 1763 mAh g−1 after 450 cycles with a capacity retention of 98% and a coulombic efficiency of 99.85% (with a current density of 358 mA g−1). This significantly supersedes the performance of a Si-dominant electrode structures. The capacity fade rate after 450 cycles was only 0.005% per cycle in the 0.05–1 V range. This superior electrochemical performance is ascribed to the highly layered, silicon-graphene porous structure, as investigated via focused ion beam in conjunction with scanning electron microscopy tomography. The hybrid electrode could retain 89% of its porosity (under a current density of 358 mA g−1) after 200 cycles compared with only 35% in a Si-dominant electrode. Moreover, this morphology can not only accommodate the large volume strains from active silicon particles, but also maintains robust electrical connectivity. This confers faster transportation of electrons and ions with significant permeation of electrolyte within the electrode. Physicochemical characterisations were performed to further correlate the electrochemical performance with the microstructural dynamics. The excellent performance of the hybrid material along with the scalability of the synthesizing process is a step forward to realize high capacity/energy density LIBs for multiple device applications.


Introduction
Lithium-ion batteries (LIBs) remain the energy storage technology of choice for next-generation automotive and grid storage applications, to economically increase their energy/power densities and range longevity [1]. However, the dominant anode material in most commercially LIBs is still graphite, having a theoretical capacity limited to 372 mAh g −1 [2]. Silicon (Si) is foreseen as the contender for next generation anode materials owing to its high specific capacity (3579 mAh g −1 ), availability, cost, and environmental benignity [3]. Recently, considerable studies have shown that crystalline silicon is a promising negative electrode candidate [4][5][6]. However, there is a need to re-design and modify Si-based anodes to overcome the physicochemical degradation of the electrode, resulting from the electrochemically-driven volume expansion (>300%) from the lithium alloying reactions [4,7]. Upon cycling, the continuous lithiation/delithiation of silicon can cause electrode pulverization, conductivity loss, continuous formation of unstable solid electrolyte interphase (SEI) film, determining capacity fade [8,9]. Additionally, Si is a semiconductor, which prevents redox reactions and electron transport [10]. This electrochemical and mechanical instability of Si-based electrodes is currently precluding their commercial viability but remains a compelling research challenge [11].
To address the above issues, numerous approaches have been explored over the years to realise the potential of silicon anodes to achieve prolonged cycling. This includes reducing the particle size to nanoscale [12,13], modifying the morphology of nanomaterials [14], distributing active silicon particles onto active/inactive host matrix [15], carbon lamination [16], binders systems [17,18] and also by incorporating diverse additives-based electrolyte system [19,20]. Researchers have found that one effective way to achieve prolonged cycling is to fabricate uniform distributions of active silicon nanoparticles (SiNPs) onto a flexible and conductive matrix, either to accommodate the volume changes in Si or to enhance electrical conductivity [21]. Recently, graphene-based materials have been exploited as a conductive additive in silicon-based anodes to substantially ameliorate concerns relating to Si as the dominant anode mass in LIBs [22][23][24]. Pristine graphene exhibits excellent electrical conductivity, mechanical strength and has a large surface area, providing substantial sites for Li + ion interaction-this makes it a preferential choice among other additives for silicon-based anodes [22,[25][26][27][28]. Furthermore, graphene's unique hierarchical structure and turbostratic distribution of the flakes when deposited as a coating, generates porous layers which enable the unrestricted flow of electrolyte throughout the electrode's architecture and operational lifetime [29]. The incorporation of graphene has been postulated to restrict electrochemical fusion of Si particles resulting from volume expansion upon lithium incorporation, whilst enhancing and maintaining the electrical contact between Si particles [30]. It maintains the electrode architecture against the huge volume changes from silicon and also improves the kinetics of the electrochemical reactions [30,31]. A homogeneous distribution of active silicon particles within the matrix is crucial to achieve high performance LIBs. This is because SiNPs tend to agglomerate, which adversely affects the electrochemical performance [32]. Previous attempts to manufacture hybrid Si-graphene anode materials have been carried out by exploiting chemical vapour deposition [33,34], atomic layer deposition [35], electrospinning [36], vacuum filtration [37][38][39], dip coating [40] and electrostatic spray deposition [41,42]. However, these processes are complex and expensive. Also, in most cases, only thin coatings can be achieved, which are difficult to pair with high-energy density cathodes in order to manufacture full cells. In contrast to the aforementioned synthesis routes, here we demonstrate a high-pressure wet-jet milling method, previously developed to efficiently exfoliate graphite and produce few-layers pristine graphene flakes (FLG WJM ) [43]. Here, the graphite and Si particles are processed simultaneously leading to a Si-FLG WJM composite that can be produced both in powder form or in solution and in large quantities (currently up to 1 kg per day in a lab-scale pilot line). This method does not require any pyrolysis steps and it is compatible with battery scale-up manufacturing lines, providing a facile and versatile approach to synthesize defect-free and high-quality multifaceted dispersion materials.
This novel strategy uses a liquid-phase mechanochemical approach in a continuous liquid flow, yielding a layer-by-layer structure. We found that the SiNPs are sandwiched in between the graphene flakes and interconnected in multiple directions with branch-like carbon black 'chains' . This interconnected porous framework is suggested to stably accommodate the volume expansion strains from SiNPs to facilitate and maintain transportation of electron and ions, in a mechanically robust microstructure. The evidence for this is based on studies of electrochemical impedance spectroscopy (EIS) combined with physical characterisation techniques able to image and quantify, over multiplescales, the evolution of internal morphology and the build-up of resistance [44].
To the best of our knowledge, the method here proposed yields the best capacity retention of any anodic materials proposed so far with such a large Si wt% content. To support this statement, table 1 outlines a range of other studies, which used a variety of methods to generate silicon-carbon composite materials. Some of these approaches used a considerably lower Si content than our study (50%-70%) and this is an important consideration when comparing electrochemical performance. We demonstrate a very large reversible gravimetric discharge capacity exceeding 1700 mAh g −1 (98% retention after 450 cycles with a coulombic efficiency of 99.85% with a current density of 356 mA g −1 ). This combined with the industrial compatibility of the composite production process, make these findings of significant interest for further testing within an industrial environment.

Results and discussion
The Silicon-graphene hybrid material directly obtained through a wet jet milling process  (Si-FLG WJM ) was used to prepare electrodes for electrochemical characterization by the addition of a conductive carbon(Super C65, TIMCAL C-NERGY) and a polymeric binder, i.e. PAA-Na, with a weight ratio of active material: polymeric binder: conductive carbon of 70:14:16. Figure 1(a) shows a scheme of the sample formulations reported and tested in this study. Formulation A indicates the electrodes based on Si-FLG WJM material. Furthermore, to compare the influence of Si particles dimension, graphene type and hybridisation approach, five additional electrode formulations were used. Basically, Si-FLG WJM performance in lithium cells have been compared to: (i) a Si-dominant electrode prepared with micrometric silicon (µmSi) particles and labelled as Formulation B; (ii) a composite prepared using µmSi particles and commercial graphene nanoplatelets (GNP, consisting of flakes with lateral dimensions of a few microns and thickness of 6-8 nm) and labelled as Formulation C; (iii) a composite obtained by the mixing of SiNPs and FLG WJM (Formulation D); (iv) FLG WJM based electrode as Formulation E and, finally, (v) GNP based electrode labelled as Formulation F. For all the Si-graphene composites, the weight ratio has been kept 1:1.

Structure and morphology
Ample sharp folds or defects on the FLG WJM flakes may act as anchoring region for SiNPs. Figures 1(b)-(g) shows the SEM images of the pristine powders (on the top) and the electrodes cross section (below). The FLG WJM powder synthesized via mechanochemical wet-jet milling exhibits a parallel planar morphology of large FLG WJM flakes ( figure 1(b)). The detailed characterization of FLG WJM is reported by Del Rio Castillo et al [43]. In the Si-FLG WJM composite (figures 1(c) and (d)) we can observe that SiNPs, having a dimension of c.a. 100 nm in diameter, are homogeneously distributed on the surface of the FLG WJM flakes and anchored on and between the FLG WJM flakes stacks.
Looking at the electrode cross section (figure 1(e)), in the electrode with Si-FLG WJM (Formulation A), it is interesting to note that FLG WJM encompasses the SiNPs forming a sandwich-like 3D structure. Such structural arrangement is not visible in the other two formulations, in which the two components were simply co-dispersed and not processed by wet jet milling (figures 1(f), (g) and S1(a) (available online at stacks.iop.org/2DM/8/015012/mmedia)).
Furthermore, it is evident that the retention of SiNPs onto FLG WJM 's surface is maintained, despite undergoing high shear mixing during electrode dispersion preparation. This demonstrates a good degree of Si-FLG WJM powder's structural integrity and robustness. It can be seen in the cross-sectional image of figure 1(e), that the FLG WJM flakes form a continuous conductive network throughout the Si particles, and to the current collector, resulting in a mechanically robust sandwich-like architecture. This inter-connected layered arrangement may promote a higher rate performance and prolonged cycle life [40], further supported later in this report by an impedance study and supported by previous literature [27,28,[56][57][58]. From the above microstructural analysis, it can be concluded that the Si-FLG WJM composite (figure 1(e)) via wet-jet milling approach containing SiNPs is well distributed throughout the FLG WJM matrix. The homogenous spread of porosity is visible in the layered composite. We conclude by pointing out that the exploited synthesis route is economical and scalable compared to the majority of the chemical methods.

Electrochemical characterisation
The electrochemical performance of the Formulation A electrode was verified in lithium cells by galvanostatic cycling in the voltage range of 0.05-1 V at 0.2 C (1 C = 3579 mA g −1 ) and compared it with the other five formulations described in figure 1(a). When cycling at the full capacity of Si (corresponding to a current density of 716 mA g −1 at 0.2 C) we expected to verify the contribution of graphene on the structural integrity and capacity retention, given its favourable tensile properties. Furthermore, to minimize the volume expansion and thus avoid the likelihood of pulverization, the cells were cycled also using a limited charge (lithiation) capacity of Si, i.e. 1800 mAh g −1 [59]. However, Formulation A (green curve in figure 2(a)) demonstrates to maintain higher capacity values during prolonged cycles and, after 50 cycles, the capacity seems to stabilize at around 2300 mA g −1 . Looking at the other formulations, particularly evident are the differences between the Formulations A and D (both prepared starting from nm Si and FLG WJM ) and Formulations B and C (prepared using µm Si and/or GNPs). Formulation D, similarly to Formulation A, exhibits a more stable cycling profile. However, the discharged specific capacity achieved after 50 cycles is lower in comparison to Formulation A (2000 mA g −1 , 13% lower than Formulation A). This observation further confirms the effectiveness of using wet-jet milling approach for simultaneous exfoliation of graphite and uniform dispersion of nmSi in between the FLG WJM layers. The shear force applied during the wet-jet milling process, results in better distribution of nm Si in the matrix. During the subsequent charge/discharge cycles, there is less possibility for the re-agglomeration of nm Si and consequently, a lesser degree capacity fading is observed in the initial cycles. It is evident that in both Formulations C and B present a discharge capacity that decreases more rapidly. In particular, Formulation B, consisting of µm Si without graphene, achieves half of the initial specific capacity only after 100 cycles.
Compared with Formulation B, the addition of GNPs (Formulation C) is able to improve the performance of this type of silicon. In summary, Formulation A, the composite prepared by wet jet milling, shows higher discharged capacity values compared with all the other formulations. It is worth mentioning that electrodes with more Si-FLG WJM powder and less additive mass (Si-FLG WJM /carbon/binder: 90/5/5) were also prepared. The cyclic performance is compared with the electrodes with more additives in figure S2. Whilst the trend is similar for both formulations after 80 cycles, the specific capacity is lower for the electrode with reduced additive mass, which can be attributed to in part by the lower conductivity of the electrode. Additionally, a lower amount of polymer binder will influence the cohesive and adhesive connections to the current collector, which will relate to resistance changes.
The electrochemical behaviour observed for Formulations B and C is likely to be related to the larger Si-particle dimensions, resulting in pulverisation and higher tortuosity for Li-ion transport. As a consequence, an increase in the cell resistance is expected, while it seems that this phenomenon is alleviated in the case of nm Si particles, as the ones used in Formulations A and D.
The electrochemical performance of the cells showed in figure S3 has been reported calculating the specific capacity related to the whole mass of electrode (active material, conductive carbon and binder) and related to the mass of active materials, including as active material also the graphene for the composite electrodes. The results are reported in figure S3. As observed in figure S3(a), Formulation A at the end of the first cycle achieves a specific capacity of almost 1500 mAh g −1 (referred to the mass of Si + graphene) that decreases to 1200 mAh g −1 after 50 cycles (at 0.2 C). Despite the capacity values becoming reduced compared with the data reported in figure 2(a), this electrode is still able to exibit high reversible capacity values. Also in this case, Formulation A was compared with the other Formulation B-D, as showed in figure S3(b). This comparison further confirms that Si-FLG WJM electrode prepared by the wet jet milling procedure offers better electrochemical performance in terms of reversible specific capacity and cyclability. To better appreciate the results obtained, we also calculated the "theoretical" specific capacity related to the total mass of Si-FLG WJM electrode based on the performance in lithium cell of graphene dominant electrodes (Formulations E-F) as reported in figure  S4. For Formulation A, the maximum capacity value obtained during cycling, the specific capacity associated with the total mass of silicon and graphene is 54.6% and 45.4% respectively. This is according to TGA measurements, figure S10. Considering Formulation A, based on the specific capacity obtained during the first lithiation at C/20, i.e. 3372 mAh g −1 for Si and 300.7 mAh g −1 for FLG WJM , the "theoretical" specific capacity can be calculated as 1977 mAh g −1 .
The analysis confirms that Formulation A shows a very promising electrochemical performance, reaching a value very similar to the considered "theoretical" capacity. However, from figure S4, we have to note that the graphene dominant electrodes present a large irreversible capacity with a large first cycle loss of 80.21 mAh g −1 and 251.77 mAh g −1 , for FLG WJM and GNPs, respectively. In fact, when cycled at 0.2 C after the first formation cycle, the FLG WJM can contribute only around 143.27 mAh g −1, whilst GNPs contribute to 127.40 mAh g −1 reversibly after 200 cycles. The reversible capacity obtained from the graphene-dominant electrodes was lower than reported in a previous study [31] and this is due to the higher applied current density (i.e. 715.8 mA g −1 in comparison to the literature's 357.9 mA g −1 ) in this study.
Further improvements in the electrochemical performance are obtained by cycling the cell at limited capacities (50% of capacity, i.e. 1800 mAh g −1 ) as shown in figure 2(b). It should be noted that all the graphene-containing formulations cycled up to 200 cycles with a current density of 358 mA g −1 .
Formulation A has shown prolonged cycle life with almost no capacity degradation even after 450 cycles (figure 2(b)), compared with other formulations comprising graphene, i.e. GNPs, (Formulation C and D). The main reason for capacity loss in Formulation C and D is most likely attributed to the collapse of the electrode's architecture (microstructure) with continuous growth of SEI (common with Si materials). This further results in the pore clogging and increased resistance towards Li + ion movement [44,60,61], as discussed in the following sections (where we present the tomography and impedance studies). From the cycling profiles, it is evident that the incorporation of graphene in the Si-electrode has a significant and positive effect on the cyclic stability, and that the Si-FLG WJM composite produced by the mechano-chemical method via continuous liquid exfoliation provides the superior performing materials.
The lithium intercalation behaviours of commercial GNPs and FLG WJM are similar to that of microcrystalline graphite [30]. However, studies have shown that with decreasing flake thickness additional strain is exerted upon the graphene flakes during ion insertion [62]. The voltage profiles (voltage vs. capacity) for the first cycle of the hybrid Si-FLG WJM powder composite and Si-dominant electrode is shown in figure 2(c). With the incorporation of graphene into the system, the lithiation capacity below the characteristic 0.1 V had decreased by 27% in comparison to the Si-dominant electrode. This can be attributed to the electrochemically activity of FLG WJM which participates in the lithiation process between 0.9-0.1 V, but to a much lower lithiation capacity.
During the first lithiation step, below 0.1 V, the two samples show a similar behaviour: a long flat plateau characteristic of silicon anodes (vs Li) and corresponding to the formation of amorphous metastable lithium silicide phase (a-Li x Si) via solid-state amorphisation [63,64]. Conversely, during the delithiation step, the two voltage profiles are quite different. The addition of graphene in the composite leads to a reduction of discharge/charge hysteresis loop. Furthermore, at lower potential (around 0.1 V) it is possible to detect two little plateaus, related to the contribution of graphene in the redox process.
We further explored the rate performance of all the formulations at different current densities including, 0.2 C, 0.5 C, and 1 C (figure 2(d)). As can be seen from figure 2(d), the first cycle at C/20 shows some differences with the cycling behaviour reported in figure 2(a). Nevertheless, the capacity values can be considered comparable, since for all cells they remain higher than 2800 mA g −1 .
Focusing on the electrochemical performance, the Si-FLG WJM composite electrode possesses 81.77% of its initial capacity at 0.2 C. By increasing the current density, the reversible capacity drops to 1789.08 mAh g −1 at 0.5 C (61.13% retention) and subsequently to 1333.09 mAh g −1 at 1 C (45.55% retention). Upon lowering the current density to 0.2 C, the reversible capacity of Si-FLG WJM composite increases to 2371.27 mAh g −1 at 0.2 C (81.03% retention). Thus, Si-FLG WJM composite electrode is presumed to undergo a minimal degradation under higher current density. This is due to the conductive network and buffering of the volumetric changes exerted from SiNPs.
In order to augment our understanding of the hybrid composites, the redox reactions and the transformation steps of Si-FLG WJM composite vs Li/Li + during lithiation/de-lithiation processes were analysed. The differential capacity plot (dQ/dV vs. V) in figure 3(a) shows more details on the electrochemical reactions taking place during lithiation and delithiation vs. Li/Li + . Upon lithiation, the Si-dominant electrode shows two redox peaks corresponding to the formation of the amorphous Li 3.5 Si [30,[64][65][66][67]. In contrast, in the Si-FLG WJM composite electrode we can observe the appearance of three additional peaks corresponding to the contribution of FLG WJM in the redox process. Herein, during lithiation, the fourth cathodic peak encompasses two reactions corresponding to the formation of LiC 12 . Furthermore, the cathodic peak at 0.06 V is attributed to the formation of LiC 6 [64]. Similar to the lithiation process, five stages can be detected during the de-lithiation process in the Si-FLG WJM composite electrode, corresponding to the reformation of a-Si and LiC 72 [30,66].
It can also be seen that with increasing cycle number there is a decrease in peak intensity and also a shift in the redox peak voltages (see figures 3(b) and (c)). The above can be attributed to the increase overpotential within the electrode i.e. impedance increases due to a continuum of SEI growth and architectural breakdown. In particular, after 200 cycles, the areas beneath the lithiation peaks have decreased by a significant amount for the Si-dominant electrode compared with the Si-FLG WJM composite electrode. In conclusion, as expected, the differential capacity plot of the Si-FLG WJM composite indicates that both Si and FLG WJM are electrochemically active within the operating potential range.  The beneficial effect of the commercial GNPs for the electrochemical performance of silicon can be noticed also in Formulation C composite that exhibits enhanced cycling performance compared with the Si-dominant formulation (Formulation B), which exhibits 87% less specific capacity. However, when compared to the other formulations (Formulation B, C and D), the Si-FLG WJM composite by wet jet milling (Formulation A) displays improved capacity retention. The coulombic efficiency for all the formulations are shown in figure S5.
In summary, the reversible capacity retention of Si-FLG hybrid systems are consistently higher than the Si-dominant electrode for all the different current densities tested. Both the cycling stability and rate capability confirm that the Si-FLG WJM composite (Formulation A) is more resilient to capacity fade from volume expansion strain in comparison to other graphene incorporated Si anodes. Alongside the unique morphology of Formulation A, the 3D porous layered structure of the electrode attributes to the superior cyclability. Herein, the poor capacity retention in Si-dominant electrodes is attributed to the higher resistance from continuous growth and breakdown of SEI along with a diminishing pore network. Therefore, the Si-FLG WJM composite electrode displayed superior cyclic performance, which could be ascribed to the unique porous multilayered morphology of the electrode. The FLG WJM flakes provide structural rigidity and hinders Si particles from detaching due to pulverization, from the FLG WJM matrix. Finally, it is conceivable that the stability of Formulation C underwent structural degradation under higher current density due to larger Si particle size in comparison to the ones used in Formulation A.

Morphology evolution
In order to directly correlate the electrode microstructure to the batteries performance, we exploited focused ion beam in conjunction with scanning electron microscopy (FIB-SEM) tomography to study 3D microstructure of battery materials (see figures 4(a) and (b)) [71][72][73][74].
FIB-SEM tomography provides high resolution, down to tens of nanometers, which is otherwise below the capability limits of the conventional laboratory source X-ray technique [75]. FIB-SEM tomography not only enables visualisation but also quantitative evaluation of the complex geometrical parameters like phase volume fractions, phase surface areas, electrode thickness, phase size distribution, and porosity [76][77][78][79][80][81][82][83]. The high resolution allows for a detailed study of finer pore sizes and different phases involved. This provides insights in clarifying structure-property-performance relationships.
Porosity plays a key role in cell performance through the electrochemical reaction kinetics, as well as the Li + ion's transport properties including tortuosity [44,84]. The morphological evolution of the Si-FLG WJM composite (Formulation A) and Si-dominant (Formulation B) for pristine electrodes and cycled electrodes (200 cycles) were analysed using (FIB-SEM), which allows for the visualization and quantification at the three-dimensional (3D) scale [74,82,83,85,86]. The morphological evolution for remainder formulations are shown in figures S6, S7 and S8. These studies are precious to offer more evidence of the benefits on the electrode structural changes brought by the incorporation of graphene in silicon-based anodes. Results are given in figures 4(c) and (d) that reports the comparison of Si-FLG WJM composite electrode (Formulation A) and Si-dominant (Formulation B) before and after cycling. The region of interest for our samples ranged between 25 × 10 µm 2 to 25 × 33 µm 2 , since the cycled electrodes had a variable change in thickness. From this, we obtained a 3D reconstructed volume of 20 × 10 × 15 µm 3 across all the samples. The porosity can be calculated from the processed images. Figures 4(c) and (d) highlight a segregation of pore groups upon cyclic aging. In Si-dominant electrodes (Formulation B), merging among the particles in the cycled electrodes can also be seen. This is most likely to result from the huge volume expansion and electrochemical fusion of particles-similar to observations made in previous studies [32,44,87]. Despite the coalescence of SiNPs after cycling, cycled Si-FLG WJM electrodes were found to retain their porous 3D morphology ( figure 4(c)). The 3D conductive network from FLG WJM , along with the flexibility from the nano-flakes, plays favourable towards the electrode's stability during lithiation/de-lithiation. This accommodates the huge volume changes of SiNPs while maintaining electrical connectivity of all throughout the electrode.
The fact that Si-FLG WJM hybrid electrodes shows superior additive properties in comparison to Si-dominant electrode could be attributed to an increased innate structural flexibility of the coating [35]. As shown in figure 5(a), the porosity retention (after 200 cycles) is higher in all Si-graphene-based hybrid anodes (Formulation A, C and D) in comparison to the Si-dominant electrode (Formulation B). When SiNPs are used, it can be seen that the porosity of the cycled electrodes at full capacity decreases by only 15% and 36% for the Formulation A and D, respectively (see figure 5(a)). This implies again the benefit of using wet-jet milling process to uniformly disperse nmSi in the graphene matrix compared to a simple dispersing method in Formulation D.
In particular, with cells cycled at half capacity, it can be seen that the Si-FLG WJM composite (Formulation A) shows little change in porosity of~11% compared to the pristine electrode. In this sense, these pores also help to maintain sufficient void structure around the active particles proving 'ion-conducting channels' to the electrode architecture [44]. Reduced porosity or increased tortuosity may adversely affect the lithium permeability and diffusion into the active material, resulting in capacity loss [44]. The observation of cracks on the electrode surface is consistent with previously reported studies [32,44,88,89]. This observation can also be supported by the tortuosity factor and impedance studies for the electrodes, which will be extensively discussed in the following sections.
The presence of graphene in the electrode believed to buffer the stresses in the structure during lithiation of Si, and this influence is very much evident in the electrode's thickness post-cycling. The initial thicknesses of the pristine electrodes were recorded as well as the post-cycling thicknesses (see SEM and FIB-SEM characterisation methods for details) and they are shown in figure 5(b). In the case of electrode with µm silicon particles (Formulation B) there is an increase in the thickness of electrodes after cycling at full capacity by >204%. Whilst the thickness value post-cycling decreases to 160%, 176% and 187% in the Si-FLG WJM composite (Formulation A), nm Si with FLG WJM (Formulation D) and µm Si with the commercial GNPs (Formulation C), respectively (see figure 5(b)).
Another observation is that electrodes containing graphene that have been cycled at half capacity, have a lower increment in the electrode thickness with respect to Si-dominant electrodes cycled at full capacity of silicon. Comparing the surface areas in contact with the electrolyte for a pristine electrode and a cycled electrode, as reported in figure 5(c), it can be seen that there is a large decrease in the specific surface area for all the formulation. Specifically, in the Si-dominant electrode (Formulation B) the surface area is reduced of >61%. This value slightly decreases when the commercial GNPs is added (Formulation C). In contrast, for the Si-FLG WJM composite (Formulation A) and nm Si with FLG WJM (Formulation D) is ca. 43% and 49%, at full capacity, respectively (see figure 5(c)). All formulations are seen to have reduced surface area and porosity upon full lithiation, with a parallel increase in electrode thickness.
We also calculated the porous volumes of the electrodes after cycling. As observed in figures 5(d) and (e), there is a significant change in the pore volume after 200 cycles for each of the electrode formulations. Upon aging, there is a shift towards the lower end of the pore volume spectrum, which may be an indication that the pore volume becomes reduced due the electrochemical fusion of the expanded Si particles in Si-dominant anodes (Formulation B). However, upon careful observation in figures 5(d) and (e) it can be seen that there are many pores with large volume, which can be explained by the large volume expansion and pulverisation that active silicon particles undergo [44]. This is believed to correlate with a decrease in charge transfer impedance upon full capacity cycling [44], and will be discussed later on in section 2.5.
The FLG WJM component in Formulation A electrodes facilitates a flexible-porous network for the active silicon particle undergoing volume changes upon lithiation and de-lithiation and maintaining 'ion-conducting channels' for the electrolyte. Overall, results in figure 5 offer direct evidence of Si-FLG WJM composite as the superior structure bestowing larger pore channels, longer and flexible conductive network, to tolerate the continuous volumetric strains from silicon particles upon cycling. Therefore, it can be concluded that by accommodating volume expansion, the electrode structure is more robust and experience less electrochemical fusion. For cells cycled under full capacity, this particle fusion (incorporating SEI) is more significant with continual SEI growth and particle cracking dominating cell degradation. Further, the thickness and surface area change for graphene dominant formulations are presented in Table S1 and S2. The graphene-dominant electrode's analysis shows that the FLG WJM is much more electrochemically stable and robust compared to the commercial GNPs.

Tortuosity factor calculation
The tortuosity factor (τ ) is an important parameter which is directly proportional to the ionic transport property of ions through pores [44,90]. τ is calculated by simulating the steady-state diffusion flow through the 3D pore network [44,91]. In our work, tortuosity is calculated using the reconstituted structures from FIB-SEM characterisation, and figure 6 presents a series of measurements comparing pristine electrodes with those of cycled electrodes at full and half capacities. Formulations A to D were analysed in three directions x,y,z in the electrode areas sampled. Formulation A (figure 6(a)) has clearly retained the lowest τ values in all directions in both the pristine and post 200 cycled samples. This is in alignment with the resulting superior electrochemical performance of the Si-FLG WJM electrodes in terms of a retained higher level of cycling efficiency and longevity. Formulation D (figure 6(d)) also maintains a comparable level of x-y-z tortuosity after 200 cycles; these were based on a composite electrode of nano Si powder and FLG WJM . Formulation B ( figure 6(b)) demonstrated the highest level of tortuosity (especially in the x direction). These electrodes, based on predominantly micron-Si, experience the highest degree of volume expansion and particle fusion. This will dominate the structural transformation and it is consistent with the porosity reductions outlined in figure 5. Tortuosity values calculated for Formulation C (figure 6(c)) were between those of A and D and of B; these electrodes were based on a composite of FLG WJM and micron-Si. The FLG WJM enables porosity retention, mechanical resilience and also reduces the degree of electrochemical fusion of micron-Si particles. Long-range electrical conductivity also improves charge transfer levels-this is discussed in more depth in the next section on impedance characterisation. From figure 6 generally, it can be seen that τ values for each of the electrodes decrease upon cycling at full capacity, except for the formulation Si-dominant (Formulation B). The explanation for this is that the Formulation B does not have any flexibility effect, which can help to sustain the volume changes, thereby the mesopores are continually filled by fragmented SEI layers.
For electrodes cycled at full capacity, τ decreases, as the volume changes from silicon particles have resulted in macro-voids and these cracks act as carrier path for the electrolyte [44]. The gradual decline of pore volume with broken SEI impedes the flow of Li + ions through the porous network [44]. The architectural fragmentation is minimal in the Si-graphene hybrid anodes as inferred from impedance measurements reported in the following section and also reflected inτ .
Formulations with Si-FLG WJM (Formulations A and D), when cycled at full capacity, have a lower τ when compared with cells cycled at half capacity [44]. At full capacity, the electrode undergoes macro cracking with large volume changes exerted from the silicon fraction. It is assumed that Si's electrochemical fusion has been prevented by incorporation of graphene.

Electrochemical impedance spectroscopy studies
The significance of incorporating graphene to improve the electrodes structural stability is further demonstrated by the sizeable improvement in the impedance relative to that of the Si-dominant anode. Measurements were taken every 10 cycles. The equivalent circuit used for the calculations is shown in figure 7(a).
In Nyquist plots showed in figures 7(b)-(d) and (h)-(j), the intersection point of the X-axis at 100 kHz refers to the series resistance (R series ). The first semicircle due to the SEI (R SEI ) can be identified between the frequency range of 100 kHz to 10 kHz, while the second semicircle of the charge transfer (R CT ) can be identified between the frequency range of 10 kHz to 10 Hz. The impedance response at lower frequencies between 10 Hz to 10 mHz seen as an angular line is generally attributed to diffusion processes [92]. Furthermore, the CPE 1 and CPE 2 reflect the double layer capacitance of the surface of SEI and active particles. Figures 7(b)-(d) shows the Nyquist plots for full capacity cycling (3579 mAh g −1 ) after 20, 50 and 100 cycles respectively, while figures 7(h)-(j) are the Nyquist plots for half capacity cycling (1800 mAh g −1 ) at the same cycle numbers. It is clear that in both cycling conditions, the high frequency and medium-high frequency (100-10 kHz) responses of Formulation B increase the most significantly, being much larger than other formulations after 100 cycles. In accordance with cycling performance in 2 a and 2b, this indicates that the main failure mechanisms of the Si-dominant electrodes are caused by significant increases in both series resistance and SEI resistance.
With reference to figures 7(e)-(g) and (k)-(m), the fitted impedance data of longer term cycling at full and half capacity respectively, there is a clear alignment with features of the corresponding Nyquist plots in figures 7(b)-(d) and (h)-(j). Generally, the series resistance (R series ) gradually increases upon cycling (as shown in figures 7(e) and (k)), which can be caused by structural degradation within electrodes, during the volume changes as a function of cycle number. It can also be observed that for electrodes containing mainly Si (Formulation B) cycled at full capacity (figure 7(f)), there is significant growth in series resistance after 30 cycles. This is likely to be caused by severe electrode pulverisation, which consumes a large quantity of Li ions when forming new SEI. This is a continual process which reduces the Li-ion concentration in the electrolyte, thus decreasing the ionic conductivity. Similar behaviour has also been noticed for the Si-dominant electrode (Formulation B) cycled under half capacity (figure 7(k)).
As observed in figures 7(f) and (l), the R SEI generally increases with the cycle number, which indicates that the SEI layer is continuously growing caused by electrode pulverisation under large volume changes. The Si only electrodes (Formulation B) exhibit a steep rise in the R SEI after 50 cycles at full capacity cycles (figure 7(f)) due to the onset of severe electrode pulverisation, whereas the same situation is delayed up to 160 cycles for half capacity cycling (figure 7(l)). Comparing with that, Si-graphene electrodes and particularly the Si-FLG WJM demonstrate much more stable R SEI over cycling. It is possibly because the FLG WJM flakes assist to maintain a stable structure under the large volume changes of SiNPs and postpone the significant electrode pulverisation, thereby, reduce the Figure 7. (a) The electrochemical equivalent circuit used for fitting the Nyquist plots; Nyquist plots during the charging process for all formulations cycled with full capacity of silicon (3579 mAh g −1 ) after (b) 10 cycles, (c) 50 cycles and (d) 100 cycles and its fitting result for the full cycle range of (e) series resistance (f) SEI resistance (g) charge transfer resistance; with half capacity of silicon (1800 mAh g −1 ) after (e) 10 cycles, (f) 50 cycles and (g) 100 cycles and the its impedance fitting results at full cycle range of (h) series resistance (i) SEI resistance (j) charge transfer resistance. The legends for (b) also applies to (c)-(m). opportunity to form new SEI on the freshly exposed Si surface.
It can be observed from figure 7(m) that for all electrodes cycled under half capacity, the R CT is mostly maintained at a relatively stable level.
The initial decrease for the first 10 cycles can be explained as a result of denser packing of active particles with conductive graphene networks, under internal pressure from Si's large volume expansion [92]. While cycled under full capacity (figure 7(g)), for electrodes with Formulation B, C & D, the R CT increased after a few cycles and exponentially declines thereafter. To explain the R CT result, it is better to firstly understand that the CT process normally involves two steps, namely, de-solvation of the solvated Li + ions in the electrolyte and transportation of these Li ions through the particle surface until it accepts an electron at the electrode's surface [44,93]. Therefore, it is possible that when the internal pores being clogged up with SEI species [32,61], there are less available Li ions to be de-solvated and transported to particle surface, which in turn increase the R CT .
This situation will be changed when there are large cracks appearing within electrode. The large voids allow comparatively easy transportation of electrolytes within the electrode and thereby facilitate the availability of Li ions to be transferred onto the surface of active particles, thus reduced the R CT [44].
However, for electrodes with Formulation A, it can be noticed that the R CT remains at a stable level even cycled under the full capacity. It indicates that there is neither significant electrochemical agglomeration nor severe electrode pulverisation occurs during first 200 cycles. This is agreement with our tortuosity findings in figure 6, which also supports our hypothesis that the synergistic effect of Si and FLG in the electrode composite enhances the cycling longevity, as FLG WJM flakes impart long-range charge transfer networks and mechanical rigidity. Additionally, any pulverized Si particles could precipitate onto FLG WJM flake surfaces and thereby still be electrically connected to contribute to capacity.

Conclusions
This study demonstrates that a facile one-step wet-jet milling applied to a dispersion of silicon and graphite, leads to a Si-FLG WJM composite that constitutes a high-performance anode. This exhibits enhanced capacity retention and the ability to function at a higher current density of 716 mA g −1 . In this composite, the SiNPs remain sandwiched between fewlayer graphene flakes, the latter obtained by the direct liquid-phase exfoliation of graphite. Silicon and FLG WJM form a layered porous conductive network. The uniform distribution of SiNPs on the FLG WJM flakes surface forms an optimised ion-conducting channel structure, allowing the effective transport of electrons and Li-ions with unconstrained electrolyte movement. This Si-FLG WJM composite powder when used as anodic material yields highly reversible electrochemical performance (reversible discharge capacity >1763 mAh g −1 after 450 cycles) with exceptional capacity retention and higher rate-capacity compared to the Si-dominant electrode or to electrodes obtained by simply mixing Si and FLG WJM or commercial graphene nano-platelets. The superior stable cycling and rate performance of the Si-FLG WJM composite electrode can be attributed to several factors: (i) the high surface area of FLG WJM flakes, with enhanced electrical conductivity (providing increased electron transport pathways between the silicon particles); (ii) the highly porous network of Si-FLG WJM composite allowing facile pathways for electrons and lithium ions, enhancing the electrical conductivity and lithium diffusion rate; (iii) the voids surrounding the SiNPs can accommodate the volume changes from silicon particles (essential for superior performance); (iv) the FLG WJM flakes restricts the silicon particle from electrochemical fusion (coalescence) resulting in more stable electrochemical performance; and (v) the planar FLG WJM flakes-like structure imparts a degree of flexibility to the electrodes and thereby endure the volume strains during silicon lithiation.
Due to the 3D carbonaceous framework, the Si-FLG WJM composite electrode demonstrates excellent cycling stability even under silicon's full capacity cycling. This excellent performance with further optimisation of the electrode formulation is propounded to improve high-capacity cycling stability and also scalability of the product makes it a prospectus anode material for LIBs.

Synthesis of graphene powder (FLG WJM )
Graphene was prepared starting from natural graphite (+100 mesh, Sigma Aldrich) following the method developed by Del Rio-Castillo et al [43]. Graphite is dispersed in N-methyl-2-pyrrolidinone (NMP) solvent (>97%, Sigma Aldrich) in order to obtain a dispersion of 100 g of powder in 10 l of solvent. Then, the dispersion was subjected to a wet-jet milling procedure [43] to allow the exfoliation of graphite in fewlayer graphene flakes. In this process, the exfoliation of graphite is promoted by the shearing forces due to the turbulent flow of the solvent forced to pass through the nozzles. Figure S11 reports representative Raman spectra for FLG WJM and GNPs. The Raman analysis reveals the higher quality in terms of defects and number of layers of the FLG WJM material.

Synthesis of silicon-graphene (Si-FLG WJM ) powder
Si-FLG WJM composite powder was prepared following a modified wet-jet milling procedure. In this case, a mixture of graphite particles (+100 mesh, Sigma Aldrich) and silicon powder (<100 nm, Sigma Aldrich) in equal amount were dispersed in NMP solvent (>97%, Sigma Aldrich) in order to obtain a dispersion of 100 g of powder in 10 l of solvent. The dispersion was mechanically mixed on a stirrer to homogenize the dispersed flakes/powders. Following this, the mixture passed through a series of nozzles of wet-jet mill apparatus following the procedure details in Del Rio Castillo et al [43]. The wet-jet mill apparatus allows for the simultaneous exfoliation of graphite and the disaggregation of silicon particles, with their distribution onto the graphene flakes. This occurs by creating a turbulent flow when the solvent is forced to pass through the nozzles. By using the wet-jet milling technique, disaggregation of silicon particle is promoted by the collision of the particles in a highly pressurized stream of liquid dispersion. The simultaneous exfoliation of graphite is promoted by shearing forces, due to the turbulent flow of solvent forced to pass through the nozzles. To obtain the final assynthesized powder, the solvent was evaporated by a rota-vapour and the powder was dried overnight under dynamic vacuum at 70 • C.

Electrode materials
Four different formulations were generated to fabricate the electrode dispersions starting from Si-FLG WJM composite powder, µm sized silicon powder with average particle size of 2.2 µm (Elkem Silgrain e-Si), nm Si powder with average particle size of <100 nm (Sigma Aldrich), as-synthesised graphene powder and commercial GNPs (thickness of 6-8 nm, XG Sciences). For the conductive additive, carbon black (Super C65, TIMCAL C-NERGY) was used. The aqueous-soluble binder used was partially neutralized polyacrylic acid (PAA), made from PAA powder (average Mw~450 000, Sigma-Aldrich) and sodium hydroxide (according to the procedure proposed by Huang et al [31]). Figure 1(a) outlines all the formulation information for this study.

Electrodes preparation
For a typical Si-graphene composite anode, the initial electrode dispersion consisted of silicon graphene powder, conductive carbon black and water as the solvent (for every 10 g of silicon 20 g of deionised water was added). The components were dispersed in a high-speed Homodisperser (Model 2.5, PRIMIX) for 30 min at 1000 rpm. The resulting mixture was ultrasonically processed (UP400S, SciMED) for two periods at 60% amplitude for 7.5 min each. Following this, the Na-PAA binder was added to the dispersion prior to high-speed homodispersion for 30 min at 1000 rpm. Finally, the dispersion was transferred to a Filmix disperser (Model 40-40, PRIMIX, Japan) for homogenisation and breakdown of any secondary agglomeration. This was a two-step process, first at 10 ms -1 for 30 s and then at 25 ms -1 for 30 s. The coating process involved spreading the asprepared dispersion onto a current collector of copper foil (10 µm, Oak Mitsui, electrodeposited) using a draw-down coater with a spreading blade (K control coater Model 101, RK Print, UK). The blade gap between the blade tip and the copper foil can be adjusted between the range 0-100 mm, for this study, a blade gap of 100 µm was used. Following deposition onto copper foil, the coating was dried on a preheated hot plate, set at a temperature of 50 • C. The coating was subsequently vacuum dried at 50 • C for an overnight period. Finally, electrodes were prepared from the coatings in a dry room (with a dew point of −45 • C) to ensure there was no exposure to moisture during the cell assembly process [44]. The mass loading of the dry film was 1.14 ± 0.18 mg cm −2 in the cut out electrode. More details on the specifications of each set of electrodes are presented in Table S3.
Electrochemical performance was characterised using a multi-channel potentiostat, VMP3 (Bio-Logic) at room temperature. For the first cycle a Crate of C/20 was used to initiate SEI formation and from the second cycle, a rate of C/5 was applied where C denotes capacity of the cell (1 C = 3579).
Further galvanostatic tests were carried out limiting the cell capacity at 1800 mAh g −1 , stopping the discharge/recharge at 2.5 h (50% state of charge) to minimise any pulverisation from large volume expansion upon lithiation [59] for improved cycle life.
Electrochemical Impedance spectra (EIS) were recorded at the end of the first cycle and repeated every 10 cycles.

Electrodes flexibility characterisation
The nano-indentation test was conducted using a Berkovich nano-indenter (NanoTest Extreme, Micro Materials Ltd, UK) and is reported in detail in our previous work Malik et al [44]. Figure S12 summarizes the results of these tests and highlights the effective better mechanical flexibility to accommodate volume expansion stresses induced by graphene.

SEM and FIB-SEM characterisation
After completion of cycling, the cell was disassembled in a glove box under an argon atmosphere. SEM characterisation was performed with field-emission SEM microscope (Sigma, Carl Zeiss, Germany). Cycled electrodes were then transferred to the focused ion beam in conjunction with scanning electron microscopy (FIB-SEM, Scios, FEI) instrument for crosssectional analysis and tomography, to understand the microstructural evolution of the Si-graphene electrodes. From the cross-sectional images, the thickness of electrodes excluding the copper foil was measured. The 3D tomography dataset were collected automatically using 'slice and view' software (FEI) and is reported in detail in our previous work Malik et al [44] (see also figure S13 and related discussion).

Image processing
The detailed image processing steps of SEM images from FIB-SEM is reported in our previous work Malik et al [44].

Acknowledgments
This project has received funding from the European Union's Horizon 2020 research and innovation program under the grant agreement No. 696656-GrapheneCore2 and the grant agreement No. 881603-GrapheneCore3. The authors also acknowledge the ISCF Faraday Challenge project "Towards a Comprehensive Understanding of Degradation Processes in EV Batteries" for funding under EP/S003053/1.