Speaker
Description
Ferroelectrics, characterized by spontaneous polarization and reversible switching, play a crucial role in various applications such as non-volatile FeRAM, ferro-TFET, and catalysis. However, the influence of environmental factors on ferroelectric domain dynamics remains poorly characterized. This work aims to investigate the impact of temperature and background gas on the domain mapping of BTO, considering the challenges posed by sample warping (~150nm thickness) and reduced signal-to-noise ratio for linear analysis methods like PCA.
To address these challenges, deep learning techniques are employed to capture noise and non-linearities in the data. Previous research in our group has utilized autoencoders to learn domain structures from STEM images, and affine transforms to extract symmetry information. In this study, we extend the approach by simulating diffraction patterns through windowing and Fourier Transform, commonly used in electron microscopy. An autoencoder augmented with an affine grid is trained to learn the symmetries and periodicities of the FFT windows. The methodology is applied to characterize phases, sample warping, and contamination in an ideal ultra-high vacuum sample. Transfer learning is then employed to analyze lower-quality scans with background gas, leveraging the knowledge gained from the trained model.
The use of symmetry informed autoencoders enables more efficient analysis compared to manual phase mapping, removing human bias and providing a pathway for real-time analysis of brightfield images.
