Scaling up "Nano" Li4Ti5O12 for High-Power Lithium-Ion Anodes Using Large Scale Flame Spray Pyrolysis (2024)

A direct correlation is noted for differing solvent and precursor solutions (given in Table I). In general, for the samples employing MeOH as processing solvent, a higher surface area is observed in all cases. While it is very similar for these samples - i.e., 131.1, 133.4, and 130.1 m² g⁻¹ for sample A, B, and D, respectively - the surface area of sample C, synthesized using a solvent mixture of xylene, acetonitrile, acetic acid, and EtOH is significantly lower (85.1 m² g⁻¹). This can be directly correlated to the low combustion enthalpy of MeOH compared to that of xylene and acetonitrile. Lower temperatures in the FSP process lead to a lower level of sintering of the synthesized particles as they move away from the flame and hence a higher surface area is achieved. Additionally, the lithium precursor utilized with MeOH as a solvent, has a much lower melting point than that used in the mixed solvent regime, which will also have an effect on resulting surface area.

XRD analysis of all samples reveals that in addition to the targeted spinel LTO phase also traces of anatase TiO₂ (#), rutile TiO₂ (*), as well as a second lithium titanate phase (Li_0.57Ti_0.86O₂, §) are present in all samples (Figure2), which is in agreement with our previous study.²¹ The amount of these additional phases, however, varies depending on both the ratio of the Li:Ti precursors and the utilized processing solvent (mixture). The comparison of the two samples obtained by using an equilibrium ratio of Li and Ti precursors but different processing solvents (sampleB and C) shows that the system employing a mixture of xylene, acetonitrile, acetic acid, and EtOH (sampleC) comprises a substantial amount of rutile TiO₂ (Figure2 and Table II), which commonly leads to rather high initial irreversibility^29–31 and strongly surface area-dependent lithium storage capability.^31,32 Sample B, synthesized by using MeOH as processing solvent, contains a significantly decreased percentage of rutile TiO₂ and the second lithium titanate phase (Li_0.57Ti_0.86O₂), but instead a substantial amount of Li₂CO₃ (11.24%, see Table II and Figure2). Increasing the initial amount of lithium precursor (sampleD, Li:Ti - 4:4), indeed, results in a further increase of Li₂CO₃ (Figure2). In contrast, lowering the relative amount of lithium precursor (sampleA, Li:Ti- 4:6) allows the synthesis of a Li₂CO₃-free product (Figure2 and Table II). Moreover, sample A comprises the highest amount of Li₄Ti₅O₁₂ (>86%) and the lowest amount of rutile TiO₂ (Table II). Thus, we focused for the further characterization on sample A only.

In a next step, SEM analysis was performed, revealing almost spherical-shaped nanoparticles with a very homogeneous size distribution and an average diameter of less than 20 nm (Figure3). This outstanding morphology was achieved thanks to the FSP synthesis method, allowing the avoidance of long heating/sintering time. These results were confirmed by carrying out high-resolution transmission electron microscopy (HRTEM, Figure4). In addition, the analysis of the interplanar distances allowed the identification of purely spinel LTO, anatase TiO₂, and rutile TiO₂ structured nanoparticles (Figure4a to 4d), while for the very few, relatively larger particles (Figure4e) obviously the spinel and anatase phases co-exist in the same particle (Figure4f), comparable to the results very recently reported by Jiang et al.³³ Indeed, it appears that such particles consist of a LTO core, surrounded by a thin layer of anatase TiO₂. This is in agreement with a recent study of He et al.³⁴ who observed a thin layer of rutile TiO₂ at the surface of the as-synthesized LTO particles, suggesting moreover that the rutile TiO₂ phase may be present at the particle surface as well.

The presence of the negligible anatase phase within the powder was also confirmed by Raman microscopy, which is highly sensitive to the detection of anatase TiO₂. A peak at 144 cm⁻¹ associated with the E_g mode of TiO₂ anatase was observed (Figure5).^35–37 Characteristic vibrational modes for the spinel Li₄Ti₅O₁₂ structure assigned to A_1g + E_g + 3F_2g were also observed at 232, 444, and 676 cm⁻¹, respectively.^35,37,38 The evolution of these bands during lithium insertion was observed via in situ Raman microscopy with spectra collected at different potentials during the discharge process at a C rate of C/15 (Figure5). At OCP (2.76 V vs. Li/Li⁺), in addition to the corresponding peaks for the anatase and Li₄Ti₅O₁₂, a peak at 519 cm⁻¹ assigned to the δ(CH₃) modes of DMC from the electrolyte was also observed.^38,39 During the discharge process the intensity of the Li₄Ti₅O₁₂ peaks at 232, 444, and 676 cm⁻¹ decreases as the lithium insertion proceeds and finally vanishes below 2.0 V. During this period the intense anatase peak at 144cm⁻¹ also weakens. At 1.83 V anatase TiO₂ begins to undergo a phase transition from tetragonal to orthorhombic Li_xTiO₂ that leads to notable changes in the Raman spectra. The E_g peak at 144cm⁻¹ splits into a doublet peak at 163 and 177 cm⁻¹ (B_2g and B_3g), and additional peaks arise at 228, 330, 530, and 625 cm⁻¹, in agreement with previous literature.³⁶ All Raman peaks weaken into the background noise at potentials lower than 1.57 V. During the subsequent charge process, at 1.73 V, the doublet peak associated with B_2g and B_3g modes of anatase TiO₂ in addition to the peaks at 230, 330, 530, and 625 cm⁻¹ reappears. The anatase and LTO peaks present in the spectrum prior to the initial discharge are observed at the end of the charge step (2.5 V), albeit with weaker intensities. Thereby, the in situ Raman measurements show that lithium insertion and removal proceeds within both the minor anatase and the majority LTO phase in the sample.

For the initial evaluation of the electrochemical performance of nanoparticulate LTO as anode material for lithium-ion batteries, electrodes were prepared using CMC as binder, which, compared to PVdF, revealed superior electrochemical performance with ZnFe₂O_4,⁴⁰ anatase TiO_2,^41,42 and LTO.⁴³ Following our previous investigation,²¹ the electrode slurries were cast on copper foil. In fact, this combination (LTO-CMC-Cu) is expected to enable the optimum electrochemical performance of nanoparticulate LTO^43,44 – apart from a potential further improvement by optimizing the relative electrode composition.

The results, obtained by galvanostatic cycling of such electrodes, are presented in Figure6. The first cycle potential profile (Figure6a) shows the characteristic potential profile for nanoparticulate LTO, revealing a constant voltage plateau at about 1.55 V upon discharge, associated with the two-phase reaction from spinel Li₄Ti₅O₁₂ to rock-salt-structured Li₇Ti₅O₁₂ upon lithiation (i.e., the insertion of 3 Li⁺ and 3 e^-),^16,17 and a rather sloped shape before and after the plateau.^17,21 The first cycle coulombic efficiency is around 85%, indicating the occurrence of electrolyte decomposition and other irreversible reactions at the electrode/electrolyte interface.^34,45–47 Also, the presence of rutile TiO₂ presumably contributes to the initial irreversibility as mentioned previously. Comparing the potential profile recorded for the first and second cycle shows that these irreversible processes, indeed, occur along the whole operational voltage range upon discharge, while the charge, i.e., the delithiation process, is, indeed, highly reversible (Figure6a). In fact, the coulombic efficiency increases continuously upon cycling, finally exceeding 99.9% and the electrode presents a highly stable performance and excellent rate capability, providing specific capacities of 133, 131,129, 127, 124, and 115 mAh g⁻¹ when applying C rates of 1C, 2C, 5C, 10C, 20C, and 50C, respectively (Figures6b and 6c). This outstanding high rate performance is assigned to the high surface area, allowing an increased electrode/electrolyte contact area, the short lithium ion and electron transport pathways in such small particles, as well as a faster conversion of the insulating spinel phase into the highly conductive rock-salt phase Li₇Ti₅O₁₂ for sufficiently small particles.^17,48

As a first, preliminary conclusion it may be stated that these results clearly reflect the improved synthesis conditions for nanoparticulate LTO compared to our previous study.²¹ The first cycle coulombic efficiency was increased from about 77% to around 85% and the initial reversible capacity at C/10 was enhanced from 131 mAh g^-1 to 141mAhg⁻¹, while at the same time the active material mass loading was increased from 70 to 80 wt% and the specific surface area was increased from about 92 to 131 m² g⁻¹. In combination with the reduced amount of phase impurities the latter certainly contributes to the improved rate capability (now 115 mAh g⁻¹ at 50C compared to about 100 mAh g⁻¹ in the previous study²¹). This result is particularly remarkable as the amount of conductive carbon was reduced in the present study (only 10 wt% rather than 15 wt%). The optimized LTO nanopowder not only provides enhanced energy but, moreover, improved power density.

Targeting the application of the optimized LTO nanopowder in commercial devices, we optimized in a second step the electrode composition concerning their overall cost, replacing copper with aluminum, by far cheaper and lighter, as current collector. Moreover, we studied the influence of the binder on the electrochemical performance and the interfacial resistance. For this comparison, we prepared electrodes using CMC, PVdF (the state-of-the-art binder for lithium-ion batteries), and PAA. In fact, these three materials are at present the most commonly studied and utilized binding agents.⁴⁹ The results of the comparative electrochemical characterization by means of galvanostatic cycling are presented in Figure7. First, the different electrodes were subjected to constant current cycling at 1C (Figure7a). All electrodes reveal a rather low first cycle coulombic efficiency of 80%, 68%, and 61% for the LTO-based electrodes comprising PAA, CMC, and PVdF, respectively. As apparent from the corresponding potential profiles (Figure7b), this low efficiency is basically related to the occurrence of irreversible reactions during the constant voltage (CV) step applied after discharge to 1.0 V, presumably related to side reactions at the electrode/electrolyte interface, i.e., electrolyte decomposition.^34,45–47 Indeed, the irreversible capacity recorded during the CV step accounts for about 10%, 18%, and 20% for PAA-, CMC-, and PVdF-based electrodes, respectively. The large differences (stressed by the application of a CV step at 1.0 V) highlight the substantial impact of the utilized binder on the electrode/electrolyte interface stability and suggest that it is decreasing in the order PAA > CMC > PVdF. Besides, it appears interesting to note that the differential capacity plots (dQ/dE vs. E) of the first cycle (Figure7c) reveal a marginally larger gap between the reduction and oxidation peak for PVdF compared to PAA and CMC (PVdF > PAA > CMC), indicating a slightly enhanced reversibility of the two-phase-equilibrium redox reaction (i.e., the Ti⁺³/Ti⁺⁴ redox reaction) for PAA- and CMC-based electrodes.

Despite the very promising results observed in the first cycle for PAA, however, these electrodes showed the poorest cycling stability and a continuous capacity fading upon subsequent constant current cycling at 1C (Figure7a), indicating that CMC and PVdF are finally more suitable as binding agents for nanoparticulate LTO-based electrodes. As the capacity fading is obviously not related to a degradation of the active material itself, it is assumed that the binder degrades, resulting in a loss of contact between the LTO particles and/or the electrode coating and the current collector.

When evaluating the rate capability of the different electrodes (Figure7d), the achievable capacity is initially the same for CMC- and PAA-based electrodes at C rates up to 5C, while it is slightly lower for the PVdF-based ones. Nonetheless, for further increased C rates (10C and 20C), the capacity decreases rather rapidly for PAA and much less for CMC and PVdF. At 20C the specific capacity is 138 mAh g⁻¹ for both CMC and PVdF and only 90 mAh g⁻¹ for those electrodes containing PAA. Again, since the comprised active material is the same, the power capability of the studied electrodes is dependent solely on the utilized binder, underlining the earlier finding that PVdF and CMC are obviously more suitable for FSP-synthesized, nanoparticulate LTO.

To further investigate the influence of the utilized binder on the electrochemical performance, we performed also a comparative electrochemical impedance spectroscopy (EIS) analysis (Figure8). It may be noted that there is, indeed, so far no general consensus on the interpretation of the impedance response of LTO-based electrodes in scientific literature. Several, partially conflicting models for fitting the experimentally obtained data have been reported,^34,50–54 which we will refer to in the following discussion of our results.

Generally, for all the electrodes, the Nyquist plots reveal two semi-circles at high frequencies (largely overlapping), one semi-circle at medium frequencies, and a sloped almost linear increase at low frequencies (Figures8a to 8c). Each semi-circle, describing a specific electrochemical process in the cell, can be expressed as a resistor R in parallel to a capacitor C. For the description of non-ideal systems, however, C is commonly replaced by a constant phase element Q.⁵⁵ Accordingly, the recorded data was fitted using the equivalent circuit R₁ + Q₂/R₂ + Q₃/R₃ + Q₄/(R₄+Q₅) utilizing the BioLogic software.

R₁ describes the resistance of the electrolyte (R_el)^54,55 and shows, as expected, roughly the same value for all electrodes, as it is independent from the utilized binder, while being constant throughout cycling (Figures8d to 8f).

The two semi-circles at high frequencies R₂/Q₂ and R₃/Q₃ occur in very similar frequency ranges. Thus, especially in case of PVdF, it is rather difficult to calculate the precise values for these two resistances separately. Also, as initially mentioned, controversial interpretations are reported in literature for these high frequency phenomena. According to Wu et al.⁵² the semi-circle at high frequencies, observed in their study, represents the electronic resistance of the material. Kitaura et al.⁵⁴ attributed it to the grain boundary resistance in the electrode, while Schweikert et al.⁵¹ claimed that it is related to the SEI formed on the Li metal electrode. Considering the frequency range, it appears, indeed, reasonable that the two semi-circles Q₂/R₂ and Q₃/R₃, observed herein, are related to electrochemical processes occurring at the electrode/electrolyte interface - including also the Li metal/electrolyte interface.⁵⁶ Therefore and in order to avoid a misinterpretation of the results, we will refer to R_int as a sum of R₂ and R₃ in the following discussion, thus taking into account all contributions to the interfacial resistance, instead of discussing R₂ and R₃ separately. Nevertheless, electron conduction between the active material particles as well as between the LTO particles and the aluminum current collector certainly represent a major contribution to these interfacial phenomena.⁵⁷ The latter is frequently affected by the presence of a native oxide layer on the aluminum current collector surface,⁵⁸ potentially resulting in a reduced adhesion of the electrode coating. Basically, R_int is very similar for all the three electrodes after ten cycles (Figures8d to 8f), though slightly higher for PVdF-based ones (Figure8d). The major difference, however, is the development of R_int upon cycling; while it is rather stable for CMC (Figure8e) and PAA (Figure8f) it is steadily increasing for PVdF (Figure8d). This increase may be assigned to a progressive loss of contact between the active material and the current collector, suggesting a stabilized contact between the LTO particles and the Al foil in case of CMC and PAA compared to PVdF, related to the presence of hydroxyl (CMC) and carboxymethyl (PAA) functional groups, which strongly interact with the native alumina surface layer.^59,60

The semi-circle observed at medium frequencies (described by Q₄/R₄ in the equivalent circuit and R_ct in Figures8d to 8f) is related to the double layer formation at the electrode/electrolyte interface and the concomitant charge-transfer processes, i.e., the energy barrier for Li insertion into the LTO particles and the simultaneous reduction of the titanium cations. Accordingly, it is affected by the Li⁺ diffusion in the electrolyte as well as the ionic and electronic conductivity of the active material.⁴¹ Since the electrolyte and active material are the same for all studied electrodes, any difference between the three must be related to a difference in charge-transfer at the electrode/electrolyte interface caused by the chemical composition and/or the distribution of the binder, covering the LTO particle surface.^60,61 For electrodes comprising PVdF and CMC, R_ct is very similar and slightly decreasing upon cycling (Figures8d and 8e), indicating a stable/slightly facilitated charge transfer upon continuous lithium ion (de-)insertion. For PAA-based electrodes, it is initially lower (Figure8f) but increases upon cycling, indicating a deteriorating charge transfer, presumably related to a partial loss of contact, i.e., an inferior binding capability of PAA compared to PVdF and CMC. As a matter of fact, this finding is in very good agreement with the previous observation of an advantageous first cycle performance of PAA-based electrodes but subsequent steadily decreasing capacity upon constant current cycling (Figure7a). To provide an explanation for these findings and their interrelation, we may assume the following: The initially very low charge transfer resistance (Figures8c and 8f) and the enhanced reversibility of the Ti³⁺/Ti⁴⁺ redox reaction (Figure7c) in case of PAA may be explained by the strong interaction of the carboxyl functional groups and titanium cations present at the LTO particle surface.⁶² In fact, considering these results, one may expect an enhanced cycling stability and rate capability for PAA-based LTO electrodes. However, it is known that LTO electrodes in contact with common lithium-ion electrolytes suffer the so-called "gassing", resulting from structural changes at the particle surface.^63,64 As the interaction between PAA and the LTO particle surface is thus much more pronounced than in case of CMC or PVdF, we may assume that consequently these structural changes have a larger impact on PAA-based electrodes compared to the other two binders, potentially leading to a steadily decreasing adhesive strength between the LTO particles, i.e., a loss of inter-particle contact, as indicated by the continuous increase of the charge-transfer resistance for such electrodes (Figures8c and 8f), and eventually resulting in the observed capacity fading upon cycling (Figures7a and 7d).

Finally, the sloped line at low frequencies can be assigned to the Li diffusion within the active material particles. This process is described by the constant phase element Q₅ rather than the commonly used Warburg-like diffusion element W due to the roughness of the electrode which causes a deviation of the real system from the ideal description.⁵⁵ Reasonably, Q₅ is similar for each binder then (Figures8a to 8c).

Considering now the sum of these results (coulombic efficiency, reversibility, cycling stability, rate capability, impedance analysis, but also cost and environmental benignity), CMC appears as the most suitable candidate for commercial electrodes. Thus, we evaluated finally also the long-term cycling stability at elevated C rates (10C) to study the power performance of such electrodes (Figure9). Following the outcome of the comparative electrochemical characterization, we did not apply a CV step in this experiment and narrowed the operational potential range to 1.2 and 2.2 V as cutoff potentials for the discharge and charge, respectively. Indeed, CMC-based electrodes show a substantially enhanced first cycle coulombic efficiency of 90.8% and remarkable cycling stability, providing still 77% of the initial specific capacity after 1000 cycles, and a coulombic efficiency of almost 100%. This excellent performance renders these electrodes highly promising for large-scale, high-power applications.