Diffraction Image Processing for 4D-STEM to Improve ASTAR Pattern Matching

The emergence of new energy materials is related to the development of highly controlled poly-crystal materials exhibiting specific and interesting phase transformation, electronic/ionic conductivity, and optical properties. The scanning TEM is one of the most actively developing analytical methods to characterize these polycrystals from microscopic to atomic scales. The interest in the hyperspectral STEM approach that gathers structural and chemical information in a 4D image stack was mainly based on imaging-spectroscopy techniques such as STEM-EDX and STEM-EELS. Last decade, thanks to a new generation of direct electron detectors, imaging algorithms, precession technique, and highly coherent electron beam, a new approach emerged called 4D-STEM in which diffraction patterns in parallel or convergent beam are acquired in hyperimage stack. In parallel beam condition, automated crystal orientation mapping (ACOM) turns out to be a new powerful tool to characterize polycrystalline materials at the nanoscale by mapping crystallographic properties.<br/><br/>This 4D-STEM method using ASTAR-ACOM system allows to build maps of crystalline phase and orientation, using scanning nano-diffraction with precession with nanometer resolution. The recent use of high-speed cameras, pixelated detectors such as CMOS cameras and hybrid-pixel detectors enabled larger areas to be scanned, and a higher resolution close to 1 nm. However, using a CMOS camera in the column implies strong changes in the acquired images in comparison to the use of a NanoMegas conventional external optical camera, as the quality of the image improves with the increased electron sensitivity and the resolution. As the 4D-STEM ASTAR suite has been optimized for images acquired with the Stingray optical camera (phosphorescent screen), the data preparation should be adapted to fit with images acquired using CMOS camera as Oneview Gatan camera.<br/><br/>The goal of the data preparation methods proposed here is to improve the quality of ASTAR pattern-matching using a dataset of diffraction patterns acquired with a CMOS Oneview camera. The high sensitivity of the CMOS camera and the data filtering developed here modify the diffraction images leading to a compromise between image quality and template matching result quality.<br/><br/>First, modifications inside diffraction patterns are estimated through image quality metrics such as signal-over-noise ratio (SNR), peak signal-over-noise (PSNR), structural similarity index measure (SSIM), mean absolute error (MAE), and root-mean-square error (RMSE). Second, the quality of the pattern-matching process on filtered and reconstructed images is evaluated using index and orientation reliability, defined in the ASTAR software. We demonstrate that the experimental data preparation provides great advantages for the pattern-matching quality result, as it reduces noise overfitting, improves structural similarity index measure, and increases the orientation reliability from average values of 10-15 before to 20-30 after filtering.<br/><br/>The data reduction method applied in this work is a registration and reconstruction code for diffraction images acquired using a CMOS camera, which we use to filter and record the diffraction signal and reconstruct the images upstream of the ASTAR suite pattern-matching software. The essential information of each reflection of a dataset (200*200*512*512) such as intensity, size, and position are recorded in a few minutes with an accuracy of the order 10^-2 px, with a data reduction factor of the order 10².<br/><br/>In this work, we show that the mapping of crystal structures and orientations provides essential information for the determination of individual particle lithiation mechanisms of cathode materials for Li-ion and Na-ion batteries, as Na_xMnV(PO₄)₃ was used in this study. In addition, this image processing method gives more confidence in phase determinations in such polycrystalline materials.