Latent Mechanisms of Polarization Switching from In Situ Electron Microscopy Observations

Abstract

In situ scanning transmission electron microscopy enables observation of the domain dynamics in ferroelectric materials as a function of externally applied bias and temperature. The resultant data sets contain a wealth of information on polarization switching and phase transition mechanisms. However, identification of these mechanisms from observational data sets has remained a problem due to a large variety of possible configurations, many of which are degenerate. Here, an approach based on a combination of deep learning‐based semantic segmentation, rotationally invariant variational autoencoder (VAE), and non‐negative matrix factorization to enable learning of a latent space representation of the data with multiple real‐space rotationally equivalent variants mapped to the same latent space descriptors is introduced. By varying the size of training sub‐images in the VAE, the degree of complexity in the structural descriptors is tuned from simple domain wall detection to the identification of switching pathways. This yields a powerful tool for the exploration of the dynamic data in mesoscopic electron, scanning probe, optical, and chemical imaging. Moreover, this work adds to the growing body of knowledge of incorporating physical constraints into the machine and deep‐learning methods to improve learned descriptors of physical phenomena.

Why it is Unfortunate that Linear Machine Learning “Works” so well in Electromechanical Switching of Ferroelectric Thin Films

“…VAEs have received significant applications in materials microscopy. [41][42][43][44][45][46] Adv. Mater.…”

“…Utility functions to simplify deployment and implementation were provided in the M3‐Learning group research package DeepMatter. [ 46 ] This package is released open source and is pip installable.…”

Disentangling Ferroelectric Wall Dynamics and Identification of Pinning Mechanisms via Deep Learning

“…Previously, we demonstrated this approach for systems with evolution of structural units in 2D materials imaged via scanning transmission electron microscopy [ 75 ] and for the emergence of order in systems of rod‐like nanoparticles. [ 63 ] Here, we extend this approach for probing time dependent dynamics and incorporate multilayer rVAE as a natural counterpart to multiclass DCNN to disentangle the domain wall interaction mechanisms such as pinning.…”

“…Here, we explore the time‐dependent domain wall dynamics in ferroelectric materials via latent representation of the time‐dependent data. [ 61–63 ] We propose the feature‐engineering approach that effectively encodes time‐dependent wall dynamics and demonstrate the applicability of rotationally invariant variational autoencoder to establish the salient features of the domain wall dynamics. The individual elements of this workflow as well as the integrated approach per se can be universally applied to the multitude of PFM applications.…”

Bayesian optimization in continuous spaces via virtual process embeddings

“…We subsequently explore the encoding of the generated functions into a low dimensional latent space via the variational autoencoder described in our previous works. 25,[32][33][34][35] Here, the trajectories generated as described above act as the input to the VAE. The VAE then builds the smooth encoding of trajectories, where the trajectories are mapped to an n-dimensional continuous latent space.…”