Three-dimensional (3D) characterizations of battery electrodes and active materials are essential to understanding their microstructure.This in turn provides essential insight into the cell performance.During battery cycling, particle morphological change, including cracking and disintegration, can lead to cell degradation [1].These morphological degradations can range from micrometers to nanometers in scale.Micro-computed tomography (CT) and Ga + based focused ion beam-scanning electron microscopy (FIB-SEM) tomography are widely used for battery analysis, but they face challenges either at high resolution or large field of view respectively.Plasma based FIB-SEM fills the gap between these two techniques with nm scale spatial resolution and hundreds of m field of view [2].Although the tomography techniques are well developed, gaining quantitative information from the representative 3D volume data is still challenging and diverse data processing methods must be developed.For a 3D data set of battery system, semantic segmentation is needed to extract the volume fraction of each component.On the other hand, instance segmentation is required for particle level morphology in order to have an in-depth understanding of the electrode's structureperformance relationship [3].Thus, traditional image processing methods cannot guarantee a satisfying result for such statistical analysis.Machine learning (ML) is critical to fulfilling this requirement [4].In this work, we demonstrated an approach with the aid of ML to reveal the microstructure and nanostructure of a cathode electrode and individual particles from a lithium-ion battery using PFIB-SEM.A used commercial lithium-ion battery was disassembled, and its electrode stack was first mechanically cut and polished.Then a stack of cross-sectional SEM images was collected using a ThermoFisher Helios 5 PFIB UXe (the data is collected by Dr. Adam Stokes and Dr. Thomas F. Blum from Thermo Fisher Scientific).The step size between SEM slices is 200nm.An example cross section is shown in Figure 1a.Cracks were observed within the secondary cathode grains and detailed information can be seen in the high-resolution SEM image.As shown in Figure 1b, stacking the cross-sectional images yields a 3D structure with physical volume of 148.5 m x 52.5 m x 45.4 m.The high ion currents achievable using PFIB allow this relatively large FIB tomography volume to be acquired within eight hours.The larger volume in turn allows for a more statistically significant evaluation of the battery microstructure and therefore performance.As shown in Figure 1c, multi-class semantic segmentation was applied to the 3D volume to separate the particle materials (including cracks), carbon binder domain, pores, and current collector out from each other.The final volume fractions of each component in the sample are shown in Figure 1d, and the spatial distributions of the components over their distance to current collector center are also provided.Particle analysis inside the electrode can provide local information about particle damage, including the local particle structure and local morphology.To achieve this kind of statistical analysis, particles in the electrode must be separated from each other and measured individually.Herein, we applied a deep learning method to conduct instance segmentation.More than 1,000 particles were extracted out as individual instances.Figure 2a shows an example cross-section and its segmentation result, and Figure 2b shows a 3D rendering of segmented result of the volume.This segmentation result enables statical analysis of the particle materials.Figure 2c demonstrates the equivalent spherical diameter distribution of the particles inside this volume.Figure 2d provides an example of single particle analysis.The properties of a single particle such as its volume or surface area can also be calculated.In summary, we developed a new method with the aid of ML to reveal the 3D morphologic and structure information of individual particles within a thick battery cathode via PFIB-SEM.This approach unveils the local morphology of the electrode and thus provides feedback to the battery design, manufacturing, and testing process.PFIB-SEM tomography combined with ML data processing is versatile technique that can be applied to many other complicated materials systems where conventional segmentation methods fall short.Potentially, this method can be applied to understand local heterogeneity within one battery to dig out the cell degradation mechanism.Additionally, it can also be used to compare the particle morphology of the same type of batteries before and after cycling to find out the courses to capacity fade.