Silicon-carbon composites for lithium-ion batteries: A comparative study of different carbon deposition approaches Free

Silicon-based anodes for lithium-ion batteries have been the subject of extensive research efforts due to the fact that their theoretical gravimetric capacity surpasses that of graphite by ten times.^1–5 However, the considerable volume change upon lithiation and delithiation introduces significant constraints on the materials design. It is well-known that silicon needs to be used in a nanostructure form to avoid pulverization upon cycling and rapid capacity fading.⁶ This introduces additional complications. The high specific surface area of nanoparticle-based coatings leads to significant solid electrolyte interface (SEI) formation, resulting in low first cycle coulombic efficiency (CE),⁷ although the use of a modified electrolyte formulation [for instance, by adding fluoroethylene carbonate (FEC)] can alleviate this issue.^8–11 In addition, silicon has poor electrical conductivity and the use of active layers which are composed of an assembly of nanoparticles further exacerbates this problem. Silicon-carbon nanostructures have been proposed as a viable solution to some of the problems mentioned earlier.¹² A carbon shell can provide additional protection to the active material by buffering the mechanical stresses resulting from the lithiation/delithiation process.¹³ The higher conductivity of carbon can enhance charge transport from the active material to the metal collector. Several groups are seeking to disperse silicon particles in a carbon matrix with the dual goal of reducing the specific surface area and increasing the conductivity of the assembly.^14,15 Xu et al. dispersed nanosized silicon in a polyvinylidene fluoride solution and performed a 700 °C pyrolysis step to obtain a silicon-amorphous carbon core–shell structure.¹⁶ Zhou et al. synthesized hollow silicon-carbon heterostructures by pyrolyzing a mixture of sucrose powder.¹⁷ Yin et al. used an electrospray synthesis of silicon microspheres with sodium alginate as a carbon source followed by an additional pyrolysis step to coat the particles using polyvinyl chloride as a secondary carbon source.¹⁸ Ren et al. reacted silicon with methyl chloride gas at 400 °C to subsequently grow silicon-carbon nanobranches after a low-temperature nitric acid post-treatment.¹⁹ Greco et al. proved that the use of few-layered graphene flakes outperform graphene oxide and amorphous carbon additives due to a superior electrochemical performance of the graphene flakes.²⁰ Our own group has also previously used carbon nanotubes and polyvinyl pyrrolidone (PVP) to make a silicon quantum dot-polymer matrix anode.¹⁵ Carbon coatings have also been achieved by flowing hydrocarbon compounds into high-temperature reactors. Su et al. decomposed benzene precursor at 1000 °C to deposit a carbon layer around spherical silica spheres by a chemical vapor deposition (CVD) method.²¹ Evanoff et al. used chemical vapor deposition of ethylene at 700 °C to apply an outer carbon coating to core–shell carbon nanotube-silicon structures.²² Sourice et al. developed an elegant laser pyrolysis process for the synthesis of silicon particles using silane as precursor and their direct, in-flight coating with a carbon shell using ethylene as precursor.^23,24

This brief literature overview suggests that there are a large number of options when it comes to realizing a core-shell silicon-carbon structure. This raises the question of how different carbon sources and different coating methodologies can influence the performance of a finished battery. To the best of our knowledge, and despite the copious silicon-carbon studies for lithium-ion batteries, there is no comprehensive work that explores this issue. The present manuscript addresses this question by analyzing the composition and morphology of silicon-carbon heterostructures produced either by high-temperature carbonization of a polymer precursor or by chemical vapor deposition of a gaseous precursor. The results of the materials characterization (transmission electron microscopy or TEM, energy dispersive X-ray spectroscopy or EDS, Raman and Fourier transform infrared spectroscopy or FTIR are correlated with the electrochemical performance of anodes based on these structures. These samples, produced using different approaches to the synthesis of the carbon shell, contain silicon particles produced with the same protocol, thus ensuring that the results are directly comparable. Our main finding is that a CVD coating is preferable under many points of view. This gas-phase approach allows the deposition of a highly conformal coating of the silicon nanoparticles and an easy control of the silicon-to-carbon ratio. The polymer carbonization route on the other hand inevitably leads to an increase in the oxidation level of the particles and results in a lower quality carbon coating. As a result, the anodes based on the CVD-coated particles show the best stability upon electrochemical cycling.

This manuscript is organized as follows: the sample preparations techniques are described first. The results of the materials characterization with respect of composition, morphology, and structure of the carbon coatings are then described in detail. The electrochemical performance of the structures is then discussed.

The in-house synthesis of silicon nanoparticles was performed by a nonthermal plasma process that has been thoroughly studied and described elsewhere.^25,26 Silane (SiH₄) gas was continuously fed to a 50-mm diameter, 375-mm long quartz plasma reactor operated at a pressure of 3.5 Torr. A 13.56 MHz (RF) power supply was used to initiate the discharge. The high energy and reactivity of the plasma reactor ionizes the gas and leads to the nucleation and growth of ∼10 nm nanoparticles. The aerosol was transported into a second in-flight quartz reactor heated by a 50-mm long tube furnace to a temperature of 1100 °C. During the high-temperature stage, the nanoparticles coalesce and grow into branched structures composed of particles with a primary size around 50 nm. A schematic of the system is shown in Fig. 1. We have found that the in-flight annealing step is highly beneficial to the anode performance, likely because of a reduction in specific surface area (from ∼250 m²/g for the nonannealed nanoparticle to 35 m²/g for the in-flight annealed particles). Surface area measurements were obtained by a Brunauer–Emmet–Teller theory Quantachrome analyzer. The details of the in-flight nanoparticle growth process go beyond the scope of this manuscript, which focuses on how the carbon shell affects the device performance, and will be summarized in another report. In addition, we should stress that the same batch of powder was used to fabricate all the batteries reported in this manuscript.

One hundred milligram of Si powder were introduced in a borosilicate glass test tube and placed into the tube furnace. C₂H₂ gas was transported into a 25-mm quartz reactor heated at 650 °C by a tube furnace. A needle valve was installed at the furnace exhaust to control the reactor pressure and tune the growth rate. The pressure was set constant at 380 Torr. These process parameters were selected, after careful trials, to achieve a conformal coating of carbon onto the silicon particles. Temperature higher than 650 °C leads to nonuniform carbon coatings and to the undesired nucleation and growth of carbon nanoparticles. The coating thickness was adjusted by varying the coating time (5, 15, and 30 min).

One hundred milligram of Si particles and 250 mg of PVP (Sigma Aldrich) were diluted in 15 ml of ethanol. The solution was poured into a test tube and vacuum-dried at 50 °C to absorb the polymer onto the nanoparticles surface. The test tube was then placed inside a 25-mm tube furnace system. The Si particles-PVP powder was heated up under a 70 sccm Ar flow at 700 °C for 1 h. The pressure was kept at 0.23 Torr during the carbonization process. The same process was repeated for a second sample with 500 mg of PVP in order to investigate variations in carbon composition.

Similar to PVP, 100 mg of Si particles and 250 mg of sucrose were diluted in 15 ml of water. The remaining steps were identical to those described above for PVP. A second sample with 500 mg of sucrose was also made.

The carbonized powder was mixed with carboxymethyl cellulose (CMC, Sigma Aldrich) diluted in water (1% solution, by weight), and with carbon black Super P (Alfa Aesar). The slurry composition was such that after drying, the weight percentage of the coating was 0.7:0.15:0.15 Si-C powder:carbon black:CMC. The slurry was applied on top of a 50 × 50 mm copper foil and spread evenly with a Meyer rod. The anode was dried under vacuum for 12 h at 90 °C. The weight loading for the samples described in this contribution is between 0.5 and 1 mg/cm². Individual 12.5-mm anodes for coin cell batteries were punched from the 2 × 2 coatings and subsequently used for battery fabrication. Polymer separators were used for the battery assembly. Lithium foil was used as counter electrode. A solution of lithium hexafluorophosphate (LiPF₆) in 1:1 v/v ethylene carbonate/diethyl carbonate was used as electrolyte. FEC was added to the electrolyte at 10% volume fraction. Coin cells were cycled between 0.01 and 1.5 V with an Arbin Instruments, Co., multichannel potentiostat. Electrochemical impedance spectroscopy (EIS) is performed on a Gamry potentiostat with the frequency range from 10 kHz to 1 mHz.

TEM analysis was performed on a Tecnai T12 transmission electron microscope. STEM imaging was performed on a FEI Titan Themis 300 instrument with EDS capability. TEM samples were prepared by drop casting the Si-C powder onto lacey carbon grids. A Horiba LabRam HR instrument equipped with a 532 nm laser was used to characterize the carbon coating in every sample. The spectra were recorded with a grating of 1800 lines/mm and a laser power of 0.151 mW. EDS characterization was obtained with a FEI Nova NanoSEM450 system equipped with an Oxford Instruments Aztec Synergy software and a X-max 50 mm² detector with resolution of 127 nm at Mn Kα. The accelerating voltage was kept at 5 kV with a working distance of 5 mm. FTIR was performed using a modular interferometer from Newport Scientific. A deuterated triglycine sulfate (DTGS) detector was used for the measurements discussed in this manuscript.

The as-produced particles without a secondary in-flight annealing step are around 10 nm in size. However, upon heating at 1100 °C, particle morphology changes considerably. Figure 2(a) shows branchlike structures formed by 50–100 nm particles clustered together. The plasma environment within the first reactor leads to the negative charging of nanoparticles,²⁷ and as a result, it is difficult to increase the particle size above 10–15 nm using this process. On the other hand, nonthermal plasma processes provide a highly reactive environment that leads to the rapid and efficient conversion of the precursor into nanoparticles.²⁶ We have found that the addition of an in-flight thermal annealing step is a convenient way to increase the particle size while still taking advantage of the high reactivity of the plasma for the nanoparticle nucleation. During the in-flight high temperature annealing stage, the particles agglomerate and sinter to form dendritic nanostructures. Figure 2(b) shows the TEM micrograph for the silicon-carbon particles produced via carbonization of PVP. The carbon appears as a thin and nonuniform coating surrounding the silicon cores. TEM micrographs of sucrose samples, not shown here for brevity, show similar features. Figures 2(c) and 2(d) show high magnification micrographs of the particles after the 30 min C₂H₂ CVD process, with the difference that now a 4 nm layer of carbon is surrounding the particles. There is a uniform coating that provides evidence of the high conformality intrinsic to this method. Lattice fringes from the 111 planes of silicon are distinguishable in Fig. 2(d), suggesting that there is a sharp interface between the silicon core and the carbon shell. The quality of the C₂H₂-based coating is confirmed with STEM mapping presented in Figs. 3(a)–3(c). The distribution of carbon [Figs. 3(b), 3(e), and 3(h)] is mainly limited to the surface of the silicon nanostructures [Figs. 3(a), 3(d), and 3(g)]. For comparison, the same analysis was performed for PVP and sucrose-treated samples [Figs. 3(d)–3(f) and Figs. 3(g)–3(i), respectively]. The superimposed map images [Figs. 3(c), 3(f), and 3(i)] unequivocally show that a more uniform coating is achieved with the CVD method. High uniformity is desirable for a few reasons. It limits the volume change associated with the lithiation-delithiation process (clamping effect), thus improving cycling stability, and it is expected to be highly beneficial with respect of charge transport through the film.

The composition of all the samples was measured by EDS. Samples were prepared by diluting the particles in chloroform and drop casting them on a copper foil. Special care was taken on ensuring the coatings were thick enough to avoid the detection of the signal from the substrate. Table I summarizes the results. The C₂H₂ CVD process appears to provide good control on carbon deposition since there is a linear increment in atomic percentage as the deposition time is increased from 5 to 30 min. The oxygen content in these samples is low (∼5%). We attribute the presence of oxygen to the growth of a native oxide layer at the particle surface. No special step was taken to prevent air exposure after the silicon nanoparticle synthesis and before the carbon coating process. The scans corresponding to both PVP and sucrose samples do not show a strong correlation between atomic percentage and the amount of carbon precursor added to the silicon particles, i.e., it is not possible to control the composition when using such polymeric precursors. Doubling the mass of either PVP or sucrose does not change the amount of carbon-to-silicon ratio significantly. On the other hand, the higher oxide level in the PVP and sucrose samples strongly suggests that the oxygen from both precursors is contributing to further oxidation of the active material. This is generally considered detrimental with respect to both the specific capacity of the anode and with respect to first cycle coulombic efficiency. It is well known that silicon oxide can be reduced to silicon during the first lithiation cycle at the cost of the irreversible formation of a lithium oxide phase.²⁸ It is important to highlight that PVP and sucrose have previously been used as carbon sources on the synthesis of Si-C nanostructures for lithium-ion batteries.^29–31 

Figure 4 shows the Raman spectra of samples synthesized by CVD of C₂H₂. The well-known silicon peak appears at ∼520 cm⁻¹ in all the samples, indicating the presence of Si nanocrystals. The interpretation of the carbon component of the spectra requires a deeper analysis. The great flexibility natural to the C–C bond leads to a variety of vibrational modes due to changes in bond orientation.^32,33 Moreover, defects and bond disorder in the material also contribute to the Raman signal.^34,35 It is therefore not trivial to identify the nature of the material, given all its potential orientations and structures. Ferrari and Robertson propose a three stage model for a better understanding of disordered and amorphous carbon.³³ The G peak position and the I_d/I_g band ratio are indicators of the degree of disorder in the material.^33,36,37 More importantly, second order vibrations around ∼2400 cm⁻¹ are also associated with the presence of graphitic carbon.³⁸ Nanocrystalline graphite has a G peak position located at ∼1600 cm⁻¹ and an I_d/I_g ratio close to 2.0, whereas those values are ∼1510 cm⁻¹ and 0.20 for amorphous carbon, respectively. In the case of the C₂H₂-CVD particles, the G peak appears around ∼1590 cm⁻¹, pointing to the presence of nanocrystalline graphite in the sample. The I_d/I_g ratio of the three samples is close to 1.0. Particles synthesized after 15 and 30 min of CVD clearly show a shoulder at ∼2400 cm⁻¹, suggesting that the carbon coating is at least partially graphitic. The G peak positions and I_d/I_g ratios of all samples are summarized in Table II.

Figure 5 shows the Raman spectra of the PVP and sucrose samples. There are no remarkable variations on The G peak position or the I_d/I_g ratio with respect to the C₂H₂ samples. However, the carbon signal appears to be weaker judging from the higher intensity of the silicon peak relative to the carbon contribution. In fact, the 250 mg PVP sample does not show any evident carbon signal. This correlates well with the low carbon content obtained in the EDS analysis. There were no second order vibrations in any of these samples.

The EDS data above indicate that the oxygen content of the PVP and sucrose samples is considerably higher than that of the C₂H₂ CVD powder. FTIR provides additional information of the influence of process parameters on the powders' chemical configuration. The FTIR measurements were performed on a zinc selenide ATR crystal. The two carbonized samples and the CVD-coated particles were first sonicated in chloroform to give dispersion with the same weight loading. The samples were then applied via drop-casting onto the ATR crystal to achieve equally uniform coatings of particles. The measurement was taken after few minutes to ensure that the solvent is fully evaporated. Chloroform was chosen for sample preparation because of its volatility. The signal strength from the IR source was also monitored to ensure that the total integrated absorbance is as close as possible between the three samples, to ensure that the measurements are directly comparable. The three absorbance spectra are shown in Fig. 6. For the case of the C₂H₂ sample, the absorbance is broadband and featureless. The samples obtained by carbonizing sucrose and PVP show much stronger absorbance in the 700–1300 cm⁻¹ range, suggesting that a more complex chemical configuration is present for these samples. The peaks around ∼1050 cm⁻¹ and ∼1250 cm⁻¹ are assigned to asymmetric stretching of SiO₂ and O–Si–C bonds.^39–41 These peaks are present on the PVP and sucrose-based powder, suggesting that the oxygen present in the precursor effectively oxidizes the silicon particles. Moreover, a peak at ∼850 cm⁻¹ is clearly distinguishable for the PVP case. This is consistent with Si–N bonding,^42,43 suggesting that even some of the nitrogen present in the polymer precursor reacts with silicon during the thermal annealing step.

The cycling performance of the batteries with the C₂H₂ CVD anodes is shown in Fig. 7. It should be stressed that these anodes have been cycled at a 0.1 C rate, and that a slow cycling rate is actually more demanding with respect to stability compared with faster cycling (such as 1 C) since the material goes through the full charge-discharge cycle, i.e., it experiences the maximum volume change. The three samples have a first-cycle coulombic efficiency (CE) of ∼82%. This initial value is generally below 100% because of the irreversible decomposition of the electrolyte during SEI formation and because of the occurrence of irreversible reactions that take place during the first lithiation cycle, such as the electrochemical reduction of the silicon oxide layer.²⁸ Irreversible binding of lithium to defects and residual hydrogen within the carbon matrix has also been proposed a mechanism that may be detrimental to first cycle performance.^44–46 The CE value should be as high as possible, as this is particularly critical with respect to integration in a full cell, i.e., a cell that does not use lithium foil as a counter electrode (i.e., a large reservoir of lithium ions). Graphite-based anodes typically have a first cycle CE of 93%. The 82% value reported here, while comparing favorably with many other reports in the literature, clearly needs further improvement. Figure 7(a) indicates that the CE continuously increases upon further cycling, reaching 99% after roughly 80 cycles. Figure 7(b) shows the variation in CE over the first ten cycles. The sample with the thinner coating (5 min CVD treatment) requires a higher number of cycles to stabilize. On the other hand, the 15 and 30 min samples reach a 97.5% CE within the first 10 cycles. Moreover, the sample with the thicker carbon coating clearly shows both the highest capacity and capacity retention over almost 100 cycles [Fig. 7(a)]. The capacity values reported in this contribution are calculated over the total mass of the material; therefore, the sample with the thicker carbon coating (smaller silicon weight fraction) shows even higher charge-discharge capacity compared to the samples with the lower carbon content. By taking into account the weight fraction of the active material, we calculate the specific capacities of the 5, 15, and 30 min samples to be ∼1500, ∼1800, and ∼2400 mA h g⁻¹ after 80 cycles, respectively.

The samples treated using PVP and sucrose as carbon sources show markedly lower performance. With respect to the first cycle CE [Fig. 8(b)], only one sample has a value approaching 82%, while the others are well below 80%. A variety of factors contribute to this, such as an increase in the degree of oxidation and a lower quality of the carbon coating, as shown by the Raman data, especially for the PVP samples. Moreover, all these samples show rapid capacity fading, with the capacity dropping below 1000 mA h g⁻¹ within few cycles [Fig. 8(a)]. We should also note that the samples produced via carbonization of PVP or sucrose are somewhat prone to sudden failure [see Fig. 8(a), in which two of the samples die after roughly 20 and 30 cycles], while on the other hand, the acetylene-treated samples show great reproducibility.

Based on the analysis presented earlier, we reach the conclusions that the quality of the carbon layer, both in terms of uniformity and degree of ordering in the carbon structure, is essential to achieve a good performance in terms of both first cycle coulombic efficiency and overall cycling stability. This is clearly shown by the reasonably good coulombic efficiency (∼82%) of the CVD C₂H₂ samples over that of the PVP and sucrose-based anodes (∼72%), which did not have a uniform carbon coating. Despite this, we reproducibly observe a reasonably good first cycle coulombic efficiency for the sample produced by carbonizing sucrose with a 2.5/1 sucrose-to-silicon ratio (the 250 mg sample; see Fig. 8). This sample shows a relatively high value of both carbon-to-silicon and oxygen-to-silicon ratio after carbonization at 700 °C (33% and 23%, respectively, as shown in Table I). This result confirms that while the presence of silicon oxide contributes to irreversible capacity loss in the first cycle, other factors also affect the first cycle performance. The interfacial chemistry between the active material and the SEI formation is known to play a crucial role in the first cycle performance.⁴⁷ 

EIS measurements were performed to better understand the differences in performance between the samples under consideration. The measurements were performed in the lithiated states of the first, second, and fifth cycle. The results are shown in Fig. 9 for the samples produced by CVD of acetylene (30 min coating) and by carbonizing PVP (500 mg for 100 mg of silicon) and sucrose (250 mg for 100 mg of silicon). The Nyquist plot is usually interpreted in terms of contributions from a series of impedances representing different charge transport mechanisms.^48,49 Following this common approach, a good fit of the experimental data was obtained by modeling the network as a series of two resistors (R₁ and R₂) each in parallel with constant-phase elements, followed by an additional Warburg diffusion element to account for ion transport through the active layer. The value of the two resistors is adjusted to fit the shape of the two high-frequency semicircles, while the Warburg element fits the low-frequency linear part of the plot. The fitting results are summarized in Table III.

We find that both samples processed by carbonizing a polymer precursor show inferior electrical stability compared to the acetylene-treated sample. While its total impedance is actually relatively high, it changes only slightly between cycles 1 and 5 (the sum of R₁ and R₂ is 222.2 Ω at cycle 1 and 234.5 Ω at cycle 5). The samples processed via carbonization of PVP and sucrose show a significantly larger increase after 5 cycles. It is difficult to provide a conclusive interpretation of what is causing the increase. The resistance R₁, associated with the high-frequency semicircle in the Nyquist plot, is usually attributed to charge transfer impedance across the SEI layer, while R₂ is attributed to the network impedance.⁴⁸ The PVP sample has a relatively stable value of R₁ over the first 5 cycles, while R₂ increases dramatically (from 24.9 to 82.4 Ω). The sample produced via sucrose carbonization shows a different behavior, with the values of both R₁ and R₂ increasing dramatically (from 76.2 to 191.6 Ω for R₁ and from 48.3 to 129.6 Ω for R₂). This suggests that PVP actually performs better at electrically stabilizing the structure, in particular, with respect to improving the stability of the SEI layer. There are small but important differences in the chemical configuration of the PVP- and sucrose-treated samples (see FTIR in Fig. 6), and we hypothesize that these may affect the interface between the particle and the SEI layer. Nevertheless, the impedance R₂ for these two samples increases significantly during the first few cycle, and this correlates with a much faster capacity fade compared to the acetylene-coated sample. Following the example of Guo et al.,⁴⁸ we attribute the increase in R₂ to a decay in the conductivity of the mesoporous layer, i.e., to the degrading of the particle-to-particle contact. The highly conformal and uniform coating achieved by the acetylene CVD process is effective at maintaining the overall conductivity of the active layer. Further investigation is needed to elucidate how the poor conductivity of the network may lead to the nonuniform lithiation in the active layer, and how the consequent nonuniform swelling may in turn further degrade the electrical contact between the particles.

We have used a two-steps plasma-hot wall reactor process to synthesize silicon nanoparticles in the 50–100 nm size range. The resulting powder was coated with carbon using two categories of techniques: a chemical vapor deposition process using acetylene as a precursor, and the carbonization of common polymers such as PVP and sucrose which have been absorbed onto the particle surface. We have performed extensive materials characterization and we have fabricated anodes for lithium ion batteries using these materials. We have found the CVD process is highly preferable since it allows the realization of a high-quality conformal carbon coating, it enables the direct control of the coating thickness, and it avoids further oxidation of the particles. On the other hand, the carbon coatings obtained by carbonization of the polymer are highly nonuniform, generally yield lower quality carbon, and the process is accompanied by uncontrolled growth of an oxide layer. These differences translate in a clearly superior performance for batteries realized using the CVD-coated silicon nanoparticles, with respect to both cycle stability and first cycle coulombic efficiency. Impedance spectroscopy confirms that the CVD coating is conducive of a structure with higher stability with respect to electrical conductivity as well. Future studies will focus on how to further improve the quality of the carbon shell, with the goal of achieving better charge transport kinetics and improved electrochemical performance.

This work was primarily supported by the National Science Foundation under Award No. 1351386 (CAREER). A.A.B. acknowledges the support of the “Consejo Nacional de Ciencia y Tecnologiá” (CONACYT, Mexico) and the University of California Institute for Mexico and the United States (UC MEXUS). TEM measurements were performed at the Center for Electron Microscopy and Microanalysis (CFAMM) at UC Riverside. The authors acknowledge the assistance of Krassimir Bozhilov for the STEM-EDS measurements and of Juchen Guo for assistance with the EIS measurements.