Link between anisotropic electrochemistry and surface transformations at single-crystal silicon electrodes: Implications for lithium-ion batteries

Silicon is a promising negative electrode material for high-energy-density Li-ion batteries (LiBs) but suffers from significant degradation due to the mechanical stress induced by lithiation. Volume expansion and lithiation in Si are strongly anisotropic but associated early interfacial transformations linked to these phenomena and their implications for electrode performance remain poorly understood. Here we develop a novel correlative electrochemical multi-microscopy approach to study local interfacial degradation at the early stages for three different surface orientations of Si single crystals: Si(100), Si(110) and Si(311), after Li-ion electrochemical cycling. The experimental strategy combines scanning electrochemical cell microscopy (SECCM) measurements with subsequently recorded scanning transmission electron microscopy images of high-quality cross sections of Si electrodes, extracted at selected SECCM regions, using a novel Xe + plasma-focused ion beam procedure. These studies reveal significant surface orientation–dependent nanoscale degradation mechanisms that strongly control electrode performance. Si(100) was immune to interfacial degradation showing the best lithiation reversibility, whereas local nanoscale delamination was observed in Si(110) leading to a lower Coulombic efficiency. Continuous electrochemical deactivation of Si(311) was associated with delamination across the whole interface, Li trapping and formation of thick (ca. 60 nm) SiO 2 structures. These results demonstrate surface crystallography to be a critical factor when designing Si-based


INTRODUCTION
Li-ion batteries (LiBs) are essential for the decarbonisation of the economy, particularly through the electrification of the transportation system. However, the high demand for electric vehicles with longer ranges requires the use of new chemistries for LiBs with higher specific capacities than prevailing materials. Silicon (Si) provides about 10 times the theoretical specific capacity of graphite and is, therefore, promising as the negative electrode in nonaqueous LiBs. [1][2][3] However, the use of Si introduces some challenges, such as a large volume expansion during lithiation that leads to mechanical instability, 4,5 a high reactivity of the solid-electrolyte interphase (SEI) that cannot fully passivate the surface to prevent side reactions causing continuous loss of Li + and rapid capacity fading, [6][7][8][9] and diffusion-controlled lithium trapping. 10,11 Several strategies have been pursued to circumvent these issues, such as reducing the size of Si particles down to 100 nm to avoid mechanical failure, 12,13 use of nanowires, 14 amorphous Si 15 or heteroatom doping. 11 These studies highlight the importance of understanding how the physicochemical properties of Si materials affect their performance and structural transformations, which is essential to identify design rules for enhanced materials and better batteries.
For instance, it is well known that crystalline Si follows a two-phase lithiation reaction with transformation to an amorphous phase during delithiation. 16,17 As a consequence, crystalline Si materials show anisotropic properties in LiBs. 18 Si undergoes anisotropic lithiation and volume expansion, with preferential expansion occurring perpendicular to {1 1 0} planes. 4,[19][20][21] This is believed to occur due to variations in lithium interfacial mobility at different crystallographic planes leading to specific interface-limited reaction rates. 22 Mechanical stresses generated by lithiation and expansion of Si also depend on particle shape and crystallinity 12,13 and thereby are connected to structural instability ultimately leading to crack initiation. 5,13 Structural degradation of Si electrodes after lithiation/delithiation cycling has been well studied by scanning electron microscopy (SEM), which has allowed visualisation of the appearance of microscale surface cracks 5,23 or the fracture of Si structures. 13 Transmission electron microscopy (TEM) is more suitable for imaging the onset of structural degradation, but the challenges to locate those inherently small defects in extended electrode surfaces have led to TEM studies usually targeting Si electrodes with massive degradation after long-term cycling. 24,25 In situ TEM has enabled the visualisation of crack propagation and fracture in Si nanoparticles with high resolution. 12,26 However, precise control of electrochemistry requires a liquid cell 27,28 ; thereby, in situ TEM studies of Si battery materials have mostly focused on the structural evolution without any correlation to changes in electrochemical performance caused by degradation. 4,29,30 Therefore, although cracking and fracture in Si materials is widely assumed to deteriorate battery performance, 31 revealing how degradation at early stages directly affects the electrochemical behaviour has been elusive.
Here, we introduce a novel correlative electrochemical multimicroscopy approach to establish a direct link between the electrochemical response of crystalline Si electrodes for LiBs and transformations of interfacial structure. In recent work, we used scanning electrochemical cell microscopy (SECCM) [32][33][34] in tandem with shellisolated nanoparticles for enhanced Raman spectroscopy (SHINERS) to investigate SEI formation at Si(1 1 1) electrodes. 35 Here, we use SECCM to perform local charge/discharge cycling of Si in an Li-ion electrolyte and correlate the responses with changes in interfacial structure visualised by scanning TEM (STEM) at commensurate regions after optimal preparation of high-quality Si cross sections by plasma focused ion beam (p-FIB). We study three distinct monocrystalline electrode orientations -Si(1 0 0), Si(1 1 0) and Si(3 1 1) -which show different electrochemical characteristics and associated degradation pathways that we are able to rationalise. This new way of examining local interfacial processes enabled the identification of nanoscale surface degradation events such as pore formation and delamination at the very early stages, which are connected to poor cycling performance and specific Si crystal orientations.

RESULTS AND DISCUSSION
The correlative multi-microscopy approach is illustrated in Figure 1 Li/Li + ) to study lithiation/delithiation processes and the initial stages of the SEI formation ( Figure 1b). This approach provides significant statistics as many repetitions (n = 49, vide infra) are recorded and allows the identification of any spatial heterogeneity in electrochemical response. SECCM leaves a footprint corresponding to the meniscus (droplet) contact at each analysis point which can be readily identified in SEM images ( Figure 1c). As such, selected spots may be extracted as part of a lamella (thin, electron-transparent cross section) using FIB

Preparation of Si cross sections by p-FIB and STEM imaging of pristine Si samples
We developed a novel Xe + p-FIB approach to prepare high-quality Si cross sections for high-resolution STEM imaging as traditional Ga + FIB techniques were found to induce various forms of damage into the lamella, including the redeposition of the milled material, implantation of Ga + ions and amorphization (Figure 2a,b). The amorphized material is typically formed as a layer across the top surface of the lamella, which then impacts upon and compromises the analysis of the SEI and/or Si oxide layers. Use of Xe + ions is known to substantially reduce the issues associated with Ga + FIB preparation. 36,37 However, in many cases, a thin (∼10 nm) amorphized layer may remain (Figure 2c), and surface protection by a Pt layer could lead to Pt redeposition on Si ( Figure 2d). To address these issues, we modified a known procedure (details in Section S1.3) that entailed initially protecting the Si surface by electron-beam deposition of two thin layers of C and Pt/C before any use of the ion beam. The lamellae were all prepared using an FEI Helios 5 Laser p-FIB system, which is capable of C and mixed-  Figure S6). Although we applied a voltammetric scan rate that was several orders of magnitude larger than typically used to study electrochemistry at Si electrodes under LiB conditions, the voltammetric responses, particularly for Si(1 0 0) and Si(1 1 0), were similar to those obtained routinely with more conventional electrochemical experiments at slower scan rates, 5,24 but with some new phenomena observed. These studies add to a growing literature that demonstrates the benefits of employing SECCM at relatively fast scan rates for the investigation of battery electrode materials, 34,35,38,39 facilitated by the relative immunity of SECCM to ohmic effects compared to equivalent macroscopic measurements, 40,41 in part as a consequence of the conical tip geometry and the low overall current magnitudes.
The voltammetric response was rather homogeneous across the 49 points on each of the Si surfaces at this scale (∼few µm diameter spots).
For Si(1 0 0), two reduction processes in the first charge/discharge cycle are tentatively assigned to electrolyte reduction (R 1 ) and lithiation (R 2 ), with a new reduction process (R 3 ) appearing upon cycling attributed to a second lithiation process. This is the typical response for two-phase lithiation, 18,42,43 entailing lithiation of both the crystalline Si originally present and amorphous Si generated after delithiation of the crystalline phase. 44,45 The increased current density upon cycling, which is seen here, is usually associated with an increased amount of active material after amorphization of Si. 46 Lithiation of the native oxide layer might also occur, 47 particularly in the first cycle, but the contribution of this process to the overall electrochemistry in further cycles should be minimal due to the low thickness of this layer ( Figure 3).
On the reverse sweep, two oxidation processes (A 1 and A 2 ) with increasing current density upon cycling are attributed to the delithiation of two different phases of Li x Si y . 18 In general, the peak potential (E pa ) was relatively stable for A 1 , with A 2 slightly shifting to more positive values upon cycling ( Figure S7). Further evidence of the spatially homogeneous response across the Si(1 0 0) surface is provided by representing SECCM data as lithiation (R 2 : j lithiation at +0.05 V, Figure 4d) and delithiation (A 2 : j delithiation at +0.60 V, Figure 4e) activity maps on Si(1 0 0) with the corresponding SECCM footprint in Figure 4f. Both j lithiation and j delithiation in the first cycle were uniform with a tight distribution of values across the surface ( Figure S8). Small differences (ca. 3.0% and 3.4% relative standard deviation for j lithiation and j delithiation , respectively) can be ascribed to experimental variability. This spatial homogeneity was largely retained after 10 charge/discharge cycles ( Figure S9) with the aforementioned increase in current density.  Figure S10). For instance, Si(1 1 0) shows only one main delithiation process consistently upon cycling (E pa evolution in Figure S7), which might be a consequence of more uniform delithiation kinetics for different Li x Si y phases than on Si(1 0 0), leading to voltammetric overlapping or the loss of lithiated silicon (vide infra). In contrast, a completely different cycling behaviour was observed for Si(3 1 1), where reduction currents decreased upon cycling without any significant increase in delithiation current density. Indeed, only a delithiation process could be detected on Si(3 1 1) with relatively low current density that also decreased upon cycling ( Figure S11). Interestingly, an additional oxidation process at a peak potential ca. +1.16 V (labelled as A 3 in Figure   S11) is tentatively assigned to an Si(3 1 1) surface reaction as the potential is too positive for a delithiation process, and the voltammetric feature also disappeared after the second cycle.  (1 1 0)), which can be associated with the initial SEI formation providing partial surface passivation that minimises electrolyte reduction in successive cycles.
The Q a /Q c ratio was significantly higher for Si(1 0 0) compared with Si(1 1 0), which indicates either more reversible lithiation/delithiation via the Si(1 0 0) surface, or lower stability of the initial SEI on Si(1 1 0) leading to a more reactive surface for side reactions. reversibility. Transient increases in Coulombic efficiency during these initial charge/discharge cycles are common with Si electrodes due to the well-known instability and breathing effect of the SEI layer. 7,48 This leads to its continuous formation, which has been identified as the main mechanism for lithium loss in Si-based LiBs. 49 In stark contrast to the above, the magnitude (in absolute value) of both Q c and Q a decrease markedly upon cycling for Si(3 1 1). This suggests the loss of electrochemical activity of the Si surface: Q c from −1.20 ± 0.08 mC cm −2 (1st cycle) to −0.16 ± 0.01 mC cm −2 (10th cycle) and Q a from 0.100 ± 0.007 mC cm −2 (1st cycle) to 0.038 ± 0.002 mC cm −2 (10th cycle). This particular behaviour led to a very low Q a /Q c ratio during the whole experiment (0.237 ± 0.006 at 10th cycle), which reaffirms the relatively poor performance of Si (3 1 1) as an LiB electrode compared to Si(1 0 0) and Si (1 1 0

Correlative STEM imaging of cycled Si interfaces
Correlative STEM was conducted after SECCM to study the morphol-  Figure S13. Both Li and F accumulated near the SEI/Si interface in contrast to C that was more concentrated away from this interface. This observation is consistent with the bilayer SEI model, 6,51 where the inner layer is enriched in inorganic species such as LiF, whereas organic species mostly form the top layers of the SEI.
Some small defects such as pores and cavities were also observed in the SEI layer, highlighted by the green arrows in Figure 5a. These nanoscale defects, exposing the Si electrode to the electrolyte, are likely to be sites for further electrolyte reactivity and thus associated with a Coulombic efficiency lower than 1, as measured at 10 charge/discharge cycles (Q a /Q c of 0.87). Figure S14  Higher resolution STEM imaging in Figure 5b shows the formation of pores at a location where delamination also occurred, with a relatively large pore size of around 20-30 nm. Some of the Si lattices are still observed as the pore size was smaller than the thickness of the lamella.
EELS mapping in Figure S15a  Si(3 1 1) showed even greater interfacial degradation after SECCM cycling (Figure 5c). The SEI/Si interface was bulged and formed a wavy and zigzag-like structure, associated with the large volume change and mechanical stresses along this surface orientation upon cycling.
Such morphological transformations in Si surfaces upon electrochemical cycling have been previously reported. 24,53 Multiple cracking and delamination on the Si(3 1 1) interface led to a fast decay of electrochemical performance with low lithiation reversibility during SECCM (vide supra). The poor lithiation reversibility suggests that degradation takes place on Li x Si, which would be irreversibly lost and unavailable for delithiation. Additionally, EELS mapping ( Figure S16a) shows the formation of a relatively thick (ca. 60 nm) SiO 2 layer at the SEI/Si interface, which may contribute to blocking Li transport pathways and poor electrical conductivity during electrochemical cycling. SiO 2 generation could be related to the specific anodic process detected on Si(3 1 1) at the first cycle shown in Figure S11, which was absent on Si(1 0 0) and Si(1 1 0). In turn, this could then be responsible for the subsequently different electrochemistry after the first cycle for Si (3 1 1).
It is, however, difficult to unambiguously determine the specific process leading to the formation of this SiO 2 layer. High-resolution STEM images in Figure S16b show the atomic scale Si interfacial structure with various orientations, which was caused by volume and morphological changes induced by cycling. The diffraction patterns showed the presence of a mixture of Si and Li x Si y that might indicate Li trapping on the Si(3 1 1) surface as an additional contribution to the low electrochemical reversibility observed.

CONCLUSIONS
In summary, a novel correlative electrochemical multi-microscopy approach has been developed to study interfacial degradation at early stages in Si single-crystal electrodes, under conditions relevant to Li-ion cells. Local electrochemical cycling and co-located interfacial imaging at the precise locations where electrochemistry is measured of three independent Si surface orientations, namely Si(1 0 0), Si (1 1 0) and Si(3 1 1), revealed a strong link between anisotropic cycling performance and nanoscale degradation phenomena. Si(1 0 0) showed overall the best lithiation reversibility (i.e. Coulombic efficiency) and a stable, homogeneous and degradation-free interface after cycling.
The Coulombic efficiency was significantly lower for Si(1 1 0), which is ascribed to the localised formation of nanoscale pores by delamination of lithiated material (Li x Si y ). Si(3 1 1) suffered from general interfacial deterioration, Li trapping and formation of thick SiO 2 structures, all of which contribute to a very low reversibility and continuous decrease of electrochemical activity upon cycling. Based on these observations that unambiguously link the early stages of interfacial degradation and cycling performance, we identify surface crystallography as a critical factor to consider when designing more stable Si-based LiBs. Indeed, our results suggest that the Si(1 0 0) orientation should be promoted in materials used for this kind of battery to increase cycle life and performance. Further, our novel correlative imaging approach opens future avenues for elucidating structure-activity-degradation relationships at the very early stages across a variety of battery materials beyond silicon and lithium ion. Our approach is compelling not only because it provides a means to link structure-electrochemical activity changes directly in the same microscopic locations, but also because multiple such measurements can be made across a surface to create large, statistically robust datasets. Nigel D. Browning https://orcid.org/0000-0003-0491-251X Patrick R. Unwin https://orcid.org/0000-0003-3106-2178