Adverse Effects of Trace Non-polar Binder on Ion Transport in Free-standing Sulfide Solid Electrolyte Separators

PIB (850 kg mol⁻¹; Scientific Polymer Products, Inc., New York, U.S.) was dried at 100 °C under vacuum overnight and dissolved in anhydrous toluene (dried using 4 Å molecular sieves for a minimum of two weeks). Li₆PS₅Cl (3–5 μm; NEI Corp., New Jersey, U.S.) was used as received.

PIB was dissolved in toluene (7.4 wt% PIB), and subsequently, LPSCl was added to make 0, 1, 2, or 5 wt% PIB content mixtures with a SE mass ratio (dry SE:polymer solution) of 0.6. The slurries were mixed via ball milling with 5 mm ZrO₂ balls for a minimum of 18 h and subsequently tape cast in an argon atmosphere onto a silicone-coated mylar substrate (The Tape Casting Warehouse) using a drawdown bar set to an 0.008-inch gap. After drying for 20 min, the top sides of the films were covered with an identical silicone-coated mylar substrate, and the entire sandwich assembly was sealed in a pouch for calendering (MTI cold roller press, MSK-HRP-MR100DC) at room temperature. The films were then removed from the pouch in an Ar-filled glovebox and dried under vacuum for two days. An illustrated process schematic is shown in Fig. 1.

Zoom In Zoom Out Reset image size

All pressure-dependent ionic conductivity and Li stripping/plating data were collected using poly(ether ether ketone) (PEEK) cells. PEEK cell assemblies were fabricated in-house and consisted of a ½'' PEEK mold fixed to a steel baseplate, a stainless-steel spacer, electrodes on either side of the sample, another stainless-steel spacer, and a stainless-steel plunger. ¹⁸

All temperature-dependent ionic conductivity measurements were performed in CR2032-type coin cells. Samples were weighed and cold pressed with carbon-coated aluminum (C@Al) current collectors in a ½'' stainless steel split die at 600 MPa for 5 min. Formed C@Al|LPSCl|C@Al symmetric cells were ejected from the die and hermetically sealed in a coin cell with a spring and 500 μm-thick stainless-steel spacer. The symmetric cells tested in this study were cold-pressed to form good contact with the electrode surface. A pair of Nyquist plots demonstrate the effects of the pressing step on the interfacial resistance in Fig. S1.

Toluene-treated LPSCl or PIB-LPSCl composite samples were prepared as described above for temperature-dependent conductivity measurements. The cells were secured in coin cell holders and placed in a programmable climate control chamber. Potentiostatic electrochemical impedance spectroscopy (PEIS) was conducted at temperatures between 80 and 20 °C in 10 °C increments for 4 cycles using a VMP3 potentiostat (Biologic). Samples were held at each temperature for 80 min to ensure thermal equilibrium before each measurement. Measurements were performed between 1 MHz and 100 mHz with an excitation amplitude of 10 mV. The ionic conductivity, σ, was calculated using the following Eq. 1, where R is the resistance at the high-frequency intercept of the Nyquist plot, A is the cross-sectional area of the cell, and L is the separator thickness.

$$\begin{eqnarray}&&\sigma ={\left(\displaystyle \frac{R\,\cdot \,A}{L}\right)}^{-1}\end{eqnarray} \tag{ 1 }$$

Pressure-dependent ionic conductivity measurements were taken in PEEK cells at room temperature. Approximately 50 mg toluene-treated LPSCl (denoted as 0 wt% hereafter) or PIB-LPSCl was measured out into a PEEK cell with two ½'' C@Al current collectors. 0, 1, 2, and 5 wt% PIB-LPSCl were pressed at 20 MPa for 5 min. At the end of 5 min, conductivity was recorded, and the pressure was released. After an additional 5 min, 10 MPa stack pressure was applied, and conductivity was again recorded. This process was repeated with a stack pressure of 5 MPa. Finally, the conductivity was recorded at approximately 0.5 MPa. The approximate time between measurements was 10 min. Final pellet thicknesses ranged from 323 to 364 μm post-fabrication.

Li stripping and plating tests were performed in PEEK cells at room temperature. Cells were cold pressed at 600 MPa for 5 min with Li-coated copper electrodes (Li@Cu; 40 μm Li layer; MSE) and cycled with a stack pressure of ∼1 MPa. Cells were cycled at ± 40 μA/cm² for 100 charge/discharge cycles, with 1 hour for striping followed by 1-hour plating. To assess critical current density, the Li|LPSCl|Li symmetric cells were respectively cycled for 6 charge/discharge cycles at 80 μA cm⁻², 118 μA cm⁻², and 236 μA cm⁻².

The morphologies of the SE samples were evaluated using scanning electron microscopy (SEM) (Zeiss MERLIN Field Emission SEM) equipped with an In-lens detector. An accelerating voltage of 1 kV was used.

ImageJ and MATLAB's Image Processing Toolbox were used to process micrographs. Collected tiff files were first cropped and converted to grayscale images. Using a reference image and MATLAB's imhisteq function, grayscale images underwent histogram matching to equalize brightness and contrast. Images were then converted back to tiff files for further processing. In ImageJ, the image type was converted to 8-bit, and pixel value thresholding was performed to isolate visible pores or dips on the sample surface as white spots on a black background. Apparent surface porosity values were taken as the percentage of white pixels. For each sample, results from four images at different magnifications were processed: 500×, 1 k×, 5 k×, and 10 k×.

All Raman maps were collected on a confocal Raman spectrometer (WITec, GmbH 532 nm, objective = 20×, a grating with 600 grooves/mm, numerical aperture (N.A.) = 0.42, local power <500 μW). The laser spot diameter was estimated to be 1 μm. The scan region was set to 50 × 50 μm², with a step size of 1 μm²/pixel. The integration time was set to 3 s for each point. All Raman mappings were analyzed using Witec ProjectPlus software and the K-means clustering algorithm integrated into the Scikit-learn platform with a similar method previously reported. ^19,20

K-means clustering analysis was performed to distill the information from the 2,500 spectra in each mapping frame (50 × 50 μm²). Our team has demonstrated in several previous studies that the K-means clustering analysis on the Raman mapping of the object can unveil structural heterogeneity.

Briefly, this method is to group the total number (denoted as n) of Raman spectra (x₁, x₂, x₃...x_n) within the Raman mapping into K sets (K ≤ n), to minimize the sum of squares within-cluster, defined by the objective function, J as

$$\begin{eqnarray}&&\begin{array}{c}J=arg\,{\rm{\min }}\,\displaystyle \sum _{i=1}^{k}\displaystyle \displaystyle \sum _{x\in {S}_{i}}{\parallel x-{c}_{i}\parallel }^{2}\end{array}\end{eqnarray} \tag{ 2 }$$

in which c_i is the mean of points (or centroid). ²¹ It serves as the cluster spectrum in the K set, S_i. The value of the centroid is updated iteratively, based on

$$\begin{eqnarray}&&\begin{array}{c}{c}_{i}^{new}=\displaystyle \frac{1}{\left|{S}_{i}\right|}\displaystyle \displaystyle \sum _{{x}_{j}\in {S}_{i}}{x}_{j}\end{array}\end{eqnarray} \tag{ 3 }$$

Consequently, the total number of n spectra can be categorized into several clusters with similarities, with each cluster color coded in the label map.

The addition of PIB allowed for the formation of thin sulfide SE separators. Figure S2a demonstrates the dimensional integrity of the SE thin films as a function of the PIB loading. Without any binder, the material would not form a film. The LPSCl powder formed brittle and plate-like chunks when slurry-processed with toluene and tape cast. 1 wt% PIB-LPSCl films were also very brittle and broke into small pieces when handled (Fig. S2a). For this reason, all further characterization performed on 1 wt% PIB-LPSCl in electrochemical cells was done using many flakes stacked together in a multi-layer, mosaic-type fashion. On the other hand, robust free-standing films were achieved with a minimum of 2 wt% PIB.

The as-cast PIB-LPSCl film thickness ranged from 120 to 250 μm. Film quality visually appeared to increase with increasing binder content (Fig. S2b). Upon calendaring, the thickness was reduced by 45 ± 8% for 2 wt% PIB-LPSCl and 52 ± 11% for 5 wt% PIB-LPSCl. The thickness of 1 wt% PIB-LPSCl did not decrease significantly upon calendaring (12 ± 12%). Samples that were cold pressed after calendaring realized additional thickness decreases of 13 ± 8%, 25 ± 16%, and 21 ± 17% for 1, 2, and 5 wt% PIB-LPSCl, respectively.

To assess morphological changes that occur as a result of binder loading, SEM of the cross-sections and surface planes of calendered, pressed pellets was performed. Representative SEM images of 0, 1, 2, and 5 wt% PIB-LPSCl samples are shown in Fig. 2. From the cross-sectional views, particle boundaries become more pronounced at 5 wt% binder loading, revealing a population of small (≲1 μm) particles and one large (>15 μm) particle aggregate structure. At lower binder loadings, particle size homogeneity is not as easily assessed due to interparticle caking. Attempts to map binder distribution within the films using EDX were unsuccessful due to carbon contamination within the sample vacuum chamber (Fig. S3). From the planar views, surface porosity appears to increase with increasing binder content. Using image processing techniques, the apparent porosity (expressed as % image area) was estimated. Example processed images are shown in the right-hand column in Fig. 2. To validate these results, bulk porosity was also tabulated.

Zoom In Zoom Out Reset image size

The porosity, φ of the LPSCl samples was calculated based on the following equations, ^22,23

$$\begin{eqnarray}&&\begin{array}{c}{\rho }_{c,theoretical}={\left[\left(\displaystyle \frac{{w}_{PIB}}{{\rho }_{PIB}}\right)+\left(\displaystyle \frac{{w}_{LPSCl}}{{\rho }_{LPSCl}}\right)\right]}^{-1}\end{array}\end{eqnarray} \tag{ 4 }$$

where w_PIB is the weight fraction of PIB, w_LPSCl is the weight fraction of LPSCl, ρ_PIB is the mass density of PIB (0.92 g cm⁻³), ρ_LPSCl is the mass density of LPSCl (1.64 g cm⁻³) and ρ_{c,theoretical} is the theoretical mass density of the composite.

$$\begin{eqnarray}&&\begin{array}{c}{\rho }_{c,experimental}=\displaystyle \frac{{m}_{c}}{\pi {r}^{2}h}\end{array}\end{eqnarray} \tag{ 5 }$$

The experimental composite mass density, ρ_{c,experimental}, was calculated using the mass of the PIB-LPSCl composite, m_c, radius of the die, r = 0.635 cm, and thickness of the composite, h.

$$\begin{eqnarray}&&\begin{array}{c}\varphi =\left(\displaystyle \frac{{\rho }_{c,theoretical}-{\rho }_{c,experimental}}{{\rho }_{c,theoretical}}\right)\times 100 \% \end{array}\end{eqnarray} \tag{ 6 }$$

Bulk porosity and apparent porosity values are plotted in Fig. 3. Bulk porosity of 2 wt% PIB-LPSCl is in good agreement with other reported values. ¹² Also shown in Fig. 3, bulk porosity is 3%–4% greater than the apparent porosity, which in combination with SEM observations, suggests internal porosity of the composites is more prominent than their surface porosity.

Zoom In Zoom Out Reset image size

The surface structure and chemistry of LPSCl thin films were characterized using Raman mapping and unsupervised learning methods. A typical Raman mapping based on single peak intensity (e.g., 429 cm⁻¹ for P-S stretch mode in PS₄ ³⁻ representing LPSCl) is shown in Figs. 4a–4b. The Raman mapping based on single peak intensity revealed that both samples had LPSCl deficient locations. However, while Raman mapping based on single peak intensity can show the chemical distribution of a given species, it fails to reflect the relative distribution of multiple components and may miss important spectral features of the object. ²⁴

Zoom In Zoom Out Reset image size

Therefore, we employed K-means clustering analysis to distill information from the 2500 spectra in each mapping frame (50 × 50 μm²). We clustered the 2500 spectra of each Raman mapping into five groups for each sample, with each cluster representing a characteristic structure or chemistry and coded by a specific color (see "Experimental" for details) as shown in Figs. 4c–4d. The corresponding similarity loading maps are given in Fig. S4. The centroid spectrum of each cluster represented the characteristic features of all spectra in that category, as depicted in Figs. 4e–4f. The centroid spectra of the 0 wt% PIB-LPSCl exhibited featured Raman bands for the LPSCl, with a distinguished peak centered at 429 cm⁻¹, overlapping with that of the pristine LPSCl powder. This indicates that LPSCl processed using toluene retains its structure. The P-S stretch mode is centered at 429 cm⁻¹ for all clusters of the 5 wt% PIB-LPSCl samples, suggesting that the PIB binder addition did not disrupt the LPSCl structure. However, clusters orange and brown exhibited a peak at 1071 cm⁻¹, likely attributed to the C–C stretch mode of the PIB polymer binder. The size of these two clusters was on the scale of 10 μm, consistent with the structural defect observed from the SEM micrograph shown in Fig. 2. Thus, the K-means clustering analysis of the Raman mapping demonstrated that the solution processing by toluene with the PIB binder did not change the LPSCl structure. However, the addition of the PIB binder led to local phase segregation due to immiscibility between the binder and LPSCl.

The relaxation of the polymer binder may affect the percolation of ion-transport pathways through the free-standing LPSCl thin film under different compressive stress values. While the pressure-dependent ion transport properties have been well explored for cold-pressed sulfide SEs, studies of free-standing sulfide SE separators are lacking. ²⁵ The pressure dependent ionic transport property can be elucidated by the impedance spectroscopy at various stack pressure values. The EIS was taken at a stack pressure of 20 MPa and again after decreasing the stack pressure to about 0.5 MPa. Upon decreasing the stack pressure, absolute resistance increased. This was to be expected since increased stack pressures are known to decrease resistance associated with interfacial contact within sulfide SE pellets. Nyquist plots of these samples are shown in Supplementary Fig. S5. At 0.5 MPa, 0 wt% PIB-LPSCl retained 64 ± 19% of its ionic conductivity at 30 MPa. 1 wt%, 2 wt%, and 5 wt% retained 49 ± 1%, 44 ± 2%, and 31 ± 2%, respectively.

To probe the pressure dependence of conductivity in more detail, a range of pressures was studied. Conductivity as a function of stack pressure is shown in Fig. 5a. Interestingly, the ionic conductivity of 5 wt% PIB-LPSCl showed a stronger dependence on stack pressure than the samples containing less binder. Such a phenomenon can be explained as follows. The non-polar PIB binder tends to distribute at the polar LPSCl grain boundaries due to the mutual exclusion between the two objects. As the PIB-LPSCl pellet was compressed, sulfide SE particles formed intimate contacts with one another to make Li⁺ percolation pathways. Consequently, PIB chains located between particles were deformed from their equilibrium conformations to accommodate the SE particles. As the compressive force was relieved, intramolecular repulsions drove the SE particles apart to allow for the deformed chains to minimize their free energy (i.e., maximize conformational entropy). This sulfide SE particle separation resulted in a loss of interparticle contact. Thus, ion conduction pathways were lost, and conductivity decreased to a greater extent than in samples with lower PIB content.

Zoom In Zoom Out Reset image size

The results of the temperature-dependent ionic conductivity data and fits corroborate these findings. The activation energy of each sample was found by fitting experimental data to the Vogel-Fulcher-Tammann (VFT) equation. The VFT equation is given in Eq. 7 where σ₀ is a prefactor, T is temperature, T₀ is the Vogel temperature, and B_R is the quantity E_a/R. E_a is the activation energy for ion conduction in the material, and R is the gas constant. ²⁶

$$\begin{eqnarray}&&\begin{array}{c}\sigma \left(T\right)={\sigma }_{0}\,{e}^{\left(\displaystyle \frac{-{B}_{R}}{T-T0}\right)}\end{array}\end{eqnarray} \tag{ 7 }$$

For polymer electrolytes, T₀ represents the ideal glass transition temperature, the temperature at which configurational entropy is zero. ²⁷ For inorganic SEs, T₀ currently bears no physical meaning. ²⁸ T₀ was fixed at a value of 800 K, such that the E_a found for neat LPSCl was near 0.34 eV, a value previously found by Arrhenius-type fitting methods. ^29,30 Arrhenius fits of collected conductivity data are shown in Fig. S6, and corresponding fit parameters are given in Table SI. The raw data is plotted with the fit results in Fig. 5b. The fit parameters, extracted activation energies, and R² values are given in Table I.

In addition to the samples containing toluene-treated LPSCl (0–5 wt% PIB-LPSCl), untreated LPSCl (neat LPSCl) was also tested as a control. According to the fit values, E_a of 0, 1, and 2 wt% PIB-LPSCl was not significantly different than that of the neat LPSCl. However, E_a of 5 wt% PIB-LPSCl was significantly greater than the other samples.

Li stripping/plating tests were conducted with Li@Cu symmetric cells in PEEK holders under an argon atmosphere at room temperature using one-hour charge/discharge cycles. The samples consisted of a neat LPSCl control cell (black), stacked flakes of 1 wt% PIB-LPSCl (gray), seven stacked layers of 2 wt% PIB-LPSCl thin film cell (blue), and two stacked layers of 5 wt% PIB-LPSCl thin film cell (red). The first ten cycles are shown in Fig. 6a, where the potential has been normalized by the pellet thickness (1.2 mm, 0.81 mm, 0.60 mm, and 0.25 mm for the neat LPSCl, 1 wt% PIB-LPSCl, 2 wt% PIB-LPSCl, and 5 wt% PIB-LPSCl, respectively). The 1 wt% PIB-LPSCl cell exhibited asymmetric polarization profiles and shorted fourth cycle. It is likely the lithium metal extruded partway through the mosaic-style pellet during fabrication due to the pellet's rough nature. The 5 wt% PIB-LPSCl thin film cell exhibited the largest magnitude overpotential due to greater electrolyte resistance within the composite. The 2 wt% PIB-LPSCl thin film cell exhibited an overpotential very similar to that of the neat LPSCl control.

Zoom In Zoom Out Reset image size

After the cells underwent 100 cycles at 40 μA cm⁻², a critical current density (CCD) test was conducted. The results of this test are given in Fig. 6b. When the current density was increased to 80 μA cm⁻², the PIB-LPSCl cells began to show instability. Upon increasing the current density to 118 μA cm⁻², the PIB-LPSCl cells shorted in the first two cycles. At the same time, the neat LPSCl sample began to show increased polarization. During the following cycle at 118 μA cm⁻², the neat LPSCl sample soft-shorted. Increased polarization preceding unstable polarization is known to be indicative of a contact-loss-driven short. ³¹ As Li metal is stripped, vacancy sites are generated at the Li/SE interface. As Li stripping continues, these vacancy sites accumulate, forming larger pores. The effective interfacial area is thereby decreased, leading to increased current density in the areas remaining intact. This phenomenon has been reported previously. ^32,33 An illustrated example of void induced short circuit mechanism is shown in Fig. 7. It is not immediately clear how the 2 wt% PIB-LPSCl cell failed. The Nyquist plots before and after cell failure are given in Supplementary Fig. S8. The CCD of the neat LPSCl was found to be 118 μA cm⁻². Potential strategies to increase the CCD include the use of an interlayer between the Li metal and SE and the use of lithiophilic composite anodes. ^34–36 For example, Wu et al. found current density at the Li/SE interface was homogenized by including a thin silver-carbon composite interlayer, which improved cycling stability against Li metal. ³

Zoom In Zoom Out Reset image size