Visualization of electrochemically driven solid-state phase transformations using operando hard X-ray spectro-imaging

In situ techniques with high temporal, spatial and chemical resolution are key to understand ubiquitous solid-state phase transformations, which are crucial to many technological applications. Hard X-ray spectro-imaging can visualize electrochemically driven phase transformations but demands considerably large samples with strong absorption signal so far. Here we show a conceptually new data analysis method to enable operando visualization of mechanistically relevant weakly absorbing samples at the nanoscale and study electrochemical reaction dynamics of iron fluoride, a promising high-capacity conversion cathode material. In two specially designed samples with distinctive microstructure and porosity, we observe homogeneous phase transformations during both discharge and charge, faster and more complete Li-storage occurring in porous polycrystalline iron fluoride, and further, incomplete charge reaction following a pathway different from conventional belief. These mechanistic insights provide guidelines for designing better conversion cathode materials to realize the promise of high-capacity lithium-ion batteries.

• The rutile FeF 2 phase was used to represent all the possible rutile related Fe 2+ -containing phases.

Materials:
All chemicals were used as received without further purification.  The operando experiments were performed using perforated 2032-type coin cells with ~4 mm holes on both sides of the cell cases. The holes were sealed using Kapton tapes. The holes need to be small to ensure a small cell impedance. The FeF 3 samples (polyhedra and MWs 1:1 by weight) were mixed with carbon black and PVDF binder in a weight ratio of 3:5:2 in NMP. The resulting slurry was pasted onto thin aluminum foils (~8 µm thickness) or carbon papers (~110 µm thickness) and then dried in vacuum to make electrodes. The thin aluminum foils and thin carbon papers are quite transparent to hard X-rays but still robust enough for handling, which is critical to the operando experiment. The density of samples on the electrodes was checked using an 15 optical microscope or a scanning electron microscope before packing into coin cells. The operando cells were assembled in an argon-filled glovebox using Li metal as the counter/quasi-reference electrode and polyethylene films soaked with 1 M LiPF 6 EC/DMC (1/1 by volume) electrolyte as the separator. No additional electrolyte was added during the cell packing in order to minimize the X-ray attenuation caused by the electrolyte liquid. The asmade cells were aged for a few hours and checked using electrochemical impedance spectroscopy before being used in the operando experiments, during which the cell was held by a custom-modified coin cell holder and discharged and charged at rate of ~1/15 C (1 C = 712 mA/g). The coin cell holder was purchased from MTI Corporation and attached to a stainless steel rod standing on the motorized stage. The positive lead of the cell holder was modified to make contact with the coin cell from the side so that the X-ray was not blocked. We also put a piece of perforated stainless steel foil between the positive lead and the coin cell to apply a gentle pressure to make better electrical contact.
Operando Hard X-ray Spectro-Imaging: The operando hard X-ray spectro-imaging experiments were performed using the full-field transmission Xray microscope (FFTXM) at beamline X8c, National Synchrotron Light Source (NSLS), Brookhaven National Laboratory (BNL). The TXM operates using a bending magnetic source and a Si (111) double-crystal monochromator. It routinely operates in absorption contrast mode over a wide energy range from 5 to 11 keV with spatial and energy resolutions of ~25 nm and ΔE/E = 10 −4 , respectively.
In the operando experiment, the operando coin cell was mounted on an motorized X, Y, Z, θ stage and aligned so that the X-ray beam could transmit through the two holes in the cell. A field of view of 40 × 40 μm 2 with a 2048 × 2048 CCD camera was used. The cell was continuously cycled in galvanostatic or potentiostatic mode and the X-ray absorption-contrast images (X-ray transmitted through the sample) and reference background images (X-ray passing through air) were collected in sequence under dynamic conditions. To track the phase transformations in the electrode, a full series of TXM images were collected at each states of discharge and charge.
Each TXM image series was collected by scanning across the Fe K-edge (7112 eV) from 7091 to 7285 eV, with a 16 step size of 2 eV, and taking one TXM image at each energy step, which generated 1024 × 1024 XANES spectra with 2 × 2 binned pixels or 512 × 512 XANES spectra with 4 × 4 binned pixels depending on the choice of camera binning. The exposure time for each image was chosen depending on the intensity of the beam and was typically 8−10 seconds for 2 × 2 camera binning and 4−5 seconds for 4 × 4 camera binning during our experiments. Each chemical phase map took ~8 minutes to finish when using 4 × 4 camera binning and the output pixel size is ~80 nm.

Data Processing and Chemical Map Construction:
Each set of TXM images was first aligned using the Xradia Controller Software to correct the positioning errors of the motor stage, X-ray optics, or sample. Then all data were analyzed using a customized program developed in house using Matlab R2011b.
We first carried out background normalization for all TXM images using the unique reference background images (X-ray passing through air) collected at each energy to get the absorption-contrast images. One example is shown in Supplementary Figure1a, which is a field-of-view taken from an operando cell discharged to ~2.0 V.
Then we could extract the XANES spectrum (X-ray attenuation versus energy raw data) at each pixel (1024 × 1024 pixels for binning 2, 512 × 512 pixels for binning 4). Here we used the single pixel spectrum (red open circles in Supplementary Figure1b) extracted from point S1 in Supplementary Figure1a to explain the data normalization procedures and compare the effectiveness of our methods with previously reported methods 4−7 .
XANES spectra need to be normalized and scaled in order to be directly compared with each other and correctly fitted using standard reference spectra. The normalization is done using the data points in the pre-edge and post-edge regions 4−7 , which involves five major steps: 1) Fit the pre-edge spectrum to a linear function and subtract it from the spectrum over the entire range of energy; 2) Identify the threshold energy E 0 , which is the maximum of the 1 st derivative of the spectrum; 3) Fit the post-edge spectrum to a linear function; 4) Determine the edge-jump value, which is the difference between the pre-edge function and post-edge function at the threshold energy E 0 ; 5) Normalize the spectrum using the edge-jump value to make the pre-edge become ~0 and the post-edge become ~1.
The normalized spectrum can then be fitted to standard reference spectra to determine the ratio between different phases. This is because that based on Beer's Law, the total X-ray attenuation at each pixel can be considered as the sum of X-ray attenuation from each constituent phase with attenuation coefficient μ and thickness t, which can be written as: where I 0 is the incident X-ray intensity and I t is the X-ray intensity after it passes through the sample. Note that attenuation coefficient μ and −ln(I t /I 0 ) are energy dependent [as the −ln(I t /I 0 ) versus energy plot is the XANES spectrum] and the rutile FeF 2 phase was used to represent all the possible rutile related Fe 2+ -containing phases.
This is a reasonable approximation because it was reported that the Li x FeF 3 (when x ≈ 1.0) phase contains structural features that are found in the rutile FeF 2 structure (23). As other battery components in the pathway of the beam (such as electrolyte, carbon, PVDF binder, and separator) also attenuate the X-ray during the operando experiment, their contribution (denoted by A bkg ) is non-trivial and should also be taken into consideration. The modified equation can be written as: Previous methods depend heavily upon the strong X-ray absorption of large-sized samples (10−20 µm) and considered the X-ray absorption of the materials under study approximately equal to the total X-ray absorption 4−7 .
This approximation is no longer valid for the smaller and weakly absorbing samples investigated herein. As a result, those methods failed to correctly normalize the XANES spectra for smaller and less X-ray absorbing samples, such as the porous FeF 3 MWs examined herein ( Supplementary Figure1a). An example of such improperly normalized spectra is shown in Supplementary Figure1e, which is clearly off the scale compared with the standard reference spectra (Supplementary Figure2). When the fitting was carried out using these improperly normalized spectra at all the pixels, the quality of the resulting chemical phase map is unsatisfactory (see Supplementary Figure1b), because very few pixels could be fitted correctly to pass the R-value filter (misfit filter).
We solved this normalization problem by approximating the internal background X-ray absorption (A bkg , black circles in Supplementary Figure1d) using the X-ray attenuating information readily available from the area that does not contain the FeF 3 sample but all the other components in the operando cell, such as electrolyte, polymeric binder, carbon black, current collector, and separator (the black box in Supplementary Figure1a). We first subtracted the internal background spectrum (black circles in Supplementary Figure1d) from the total X-ray attenuation (red circles in Supplementary Figure1d) and then carried out the data normalization following the aforementioned five-step procedure, which yielded correctly normalized spectrum as shown in Supplementary   Figure1f. Using our custom-developed program, this normalization procedure could be conveniently applied to the spectra at all pixels.
The correctly normalized spectrum at each pixel was then fitted with the linear combination of three µt values. The ratio of the weighing factor is an analogue of the thickness fraction and therefore represents the volume fractions of solid state phases containing different Fe oxidation states. The fitting was carried out by minimizing the R value (a measure of misfit) for each spectrum at each pixel, which is defined as: where Ei is 7091 eV, Ef is 7285 eV, dataE is the normalized spectrum at each pixel for the given energy E, and refE is the possible fitting reference value that is a linear combination of X-ray attenuation of FeF 3 , FeF 2 , and Fe. show that our new data processing procedures could generate a higher quality chemical phase map compared with the previously reported methods. These new data processing procedures were consistently employed to yield the chemical phase maps shown in Figure 2  Laboratory. The measurements were performed in transmission mode using a Si (111) double-crystal monochromator, which was detuned to ~35% of its original maximum intensity to eliminate the high order harmonics in the beam. A reference X-ray absorption spectrum of Fe (K-edge 7112 eV) was simultaneously collected using a standard Fe foil. Energy calibration was done using the first inflection point of the Fe K-edge spectrum as the reference point. The X-ray absorption data were processed and analyzed using IFEFFIT-ATHENA. Standard reference spectra from commercial FeF 3 , FeF 2 , and Fe powders were also collected in order to carry out linear combination analysis to determine the ratio between different Fe oxidation states. The rutile FeF 2 phase was used to represent all the possible rutile related Fe 2+ -containing phases. This is a reasonable approximation because it was reported that the Li x FeF 3 (when x ≈ 1.0) phase contains structural features that are found in the rutile FeF 2 structure S8 .
The operando experiments were performed using perforated 2032-type coin cells with holes on both sides of the cell cases. The holes were sealed using Kapton tapes. The FeF 3 MWs were mixed with carbon black and PVDF binder in a weight ratio of 7:2:1 in NMP. The resulting slurry was pasted onto thin aluminum foils (~25 µm thickness) to make electrodes. The operando cells were assembled in an argon-filled glovebox using Li metal as the counter/quasi-reference electrode and polyethylene films soaked with 1 M LiPF 6 EC/DMC (1/1 by volume) electrolyte as the separator. The as-made cells were aged for a few hours and checked using electrochemical 20