Imaging Sodium Dendrite Growth in All‐Solid‐State Sodium Batteries Using 23Na T 2‐Weighted Magnetic Resonance Imaging

Abstract Two‐dimensional, Knight‐shifted, T 2‐contrasted 23Na magnetic resonance imaging (MRI) of an all‐solid‐state cell with a Na electrode and a ceramic electrolyte is employed to directly observe Na microstructural growth. A spalling dendritic morphology is observed and confirmed by more conventional post‐mortem analysis; X‐ray tomography and scanning electron microscopy. A significantly larger 23Na T 2 for the dendritic growth, compared with the bulk metal electrode, is attributed to increased sodium ion mobility in the dendrite. 23Na T 2‐contrast MRI of metallic sodium offers a clear, routine method for observing and isolating microstructural growths and can supplement the current suite of techniques utilised to analyse dendritic growth in all‐solid‐state cells.

The samples were packaged in an argon sealed pouch for all experiments, and no signs of sodium metal oxidation were present during the experiments or 10 days after experimentation, suggesting the packing and stability of the samples under experimentation was effective over a long period (10 days, as shown in Fig. S1b).
Preparation of Na | Na-β''-Alumina | Na cells: Discs of Na-β''-Alumina measuring 20 mm in diameter and 2 mm thick were purchased from Ionotec and their surfaces prepared by polishing and heat-treatment as described previously. [1] Na-β''-Alumina was chosen for this study as it is a realistic candidate for Na anode SSBs. It has a very high ionic conductivity of 2 mS cm -1 , is commercially available, and is very well characterised in the literature. Na metal electrodes, 18 mm in diameter, were cut from Na ingot and after having any oxide layer carefully removed with a scalpel, were pressed against the ceramic electrolyte in an argonfilled glovebox. The symmetric cell was then sealed within a pouch cell utilising aluminium current collectors.
The cell size and characteristics are archetypical for studying dendrite formation mechanisms in ASSBs, which allows plating and stripping behaviour simultaneously on either side of the cell and removes any unnecessary cathode complexity. When the spatial resolution of a technique is limited (such as in neutron imaging), it is useful to use cells which are as large as possible, therefore, when imaging the whole sample one can achieve a better relative spatial resolution. For this is the reason larger cells are typically used to study dendritic growth processes in ASSBs.
Galvanostatic cycling of Na | Na-β''-Alumina | Na: Cycling was conducted under a stackpressure of 1 MPa. The cell's working electrode was stripped at a current density of 0.5 mA cm -2 moving capacity of 9.5 mAh cm -2 in 19 steps of 0.5 mAh cm -2 charge passed. These conditions were chosen to create significant contact loss at the working electrode / solid electrolyte interface [1] which is evident by the approximately 1 V increase in polarisation over the course of the stripping (Fig. 1a) and by the voids observed in the interpolated Knight shift 23 Na MRI image (Fig. 1d). Upon reversing the current and plating the working electrode, the voltage was observed to drop almost immediately to 0 V, indicating a short-circuit due to dendrite growth.

Scanning Electron Microscopy (SEM) and X-ray Computed Tomography:
The pristine and short-circuited Na | Na-β''-Alumina | Na cells were sealed into pouches after the current collectors were removed in an argon-filled glovebox and scanned inside a Zeiss Xradia Versa 510 X-ray microscope. For each tomogram, 3201 equiangularly distributed projections were taken over 360 0 , with a beam energy of 120 keV and a resulting voxel size of 4.66 µm. The projections were processed and reconstructed into tomograms using the Zeiss Scout-and-Scan Reconstructor software. Post-processing of the tomograms was completed using ImageJ for cropping and reslicing. [2] The short-circuited cell was then disassembled in an argon environment and the Na electrode were removed. The ceramic was then cross sectioned across a spalling feature, located by XCT, with the ultrasonic cutter with a tungsten carbide blade (SONOTEC). The crosssectioned cell was transferred into a Zeiss Merlin scanning electron microscope using an airtight transfer device (Gatan) for SEM and EDX analysis. Secondary electron images were collected using a probe current of 200 pA and an acceleration voltage of 5 kV.  Figure S2: (a) The Knight-shifted MRS spectra of the pristine and short circuit symmetrical cells. These spectra were achieved on the post short-circuit cell and were taken less than 5 minutes after cycling (red), after the MRI experiments (blue), and 10 days after the MRI experiments (purple), with the inset showing the three spectra overlaid. The lack of any discernible differences suggests the cell was not exposed to any oxygen/water and the MRI experiment did not affect the structure (i.e. localised heating effects). (b) Overlay of the Knightshifted MRS spectra of the symmetrical cell in the pristine state and after a short circuit, showing no obvious differences, this suggests that conventional NMR cannot be used to observe the dendrite. (c) The spin-echo bulk T2 exponential decays, measured by MRS, of the bulk 23 Na metal (4 ms) and an isolated dendrite (11 ms). The possible deviations in the T2 between the MRS and MRI data (Fig. 4a and c) are attributed to the 1 MPa pressure applied to the MRI cycled cell. The dendrite was isolated in the ceramic by polishing the metallic electrodes from a shorted cell. Figure S3: The 2D Knight-shifted 23 Na MRI at τ increments varying from 2.5 to 11.5 ms for the cell after short circuit (zero-filled to 256 x 256 points). The analogous 128 x 32 images were used to generate the T2 weighted image given in Fig. 3c. Figure S4: The gradient-echo intensity 128 x 32 images of (a) the pristine cell and (b) the cell after short circuit. The dashed red circle highlights the spallation feature (the images are presented in the −z,y projections, compared to the corresponding spin-echo images in Fig. 1). The images have 8 25 mm 2 FOVs, were collected with 512 transients, and 1s recycle delay with a gradient refocusing time of 590 ms. The raw data were zero-filled to 256 x 64 and apodised in each dimension using a sinbell squared function before image reconstruction. The long 1s delay was chosen to alleviate any potential thermal effects. As the imaging was not performed in-situ and rather for the proof of principle the recycle delay was not adjusted to 5T1 of 23 Na in the metallic phase (T1 of bulk 23 Na metal <100 ms). [3] In principle, one could reduce the imaging time using this methodology to <20 minutes by using an appropriate recycle delay.

Notes
Supporting research data has been deposited in the Oxford Research Archive and is available under this DOI: 10.5287/bodleian:1amymAzYZ