Physics-informed models of domain wall dynamics as a route for autonomous domain wall design via reinforcement learning

Benjamin R. Smith, Bharat Pant, Yongtao Liu, Yu Chen Liu, Jan Chi Yang, Stephen Jesse, Anahita Khojandi, Sergei V. Kalinin, Ye Cao, Rama K. Vasudevan

研究成果: Article同行評審

摘要

Understanding the dynamics of domain walls in ferroelectrics is critical both for fundamental reasons of studying interfacial dynamics in disordered media, as well as practical engineering of metastable states with enhanced properties. Piezo response force microscopy (PFM) enables both imaging and writing of ferroelectric domain walls via a biased scanning probe. However, control over positioning of individual domain wall segments to engineer domain wall structures over large areas reproducibly, and particularly, quantification of associated mechanisms remains challenging. Here, we present a reinforcement learning based experimental workflow deployed on an autonomous PFM platform that enables automated data collection of domain walls interacting with pinning sites. The autonomous experiment is used to construct a physics-informed surrogate model of local domain wall response in response to applied electric fields by the PFM tip in prototypical (110) PbTiO3 thin films, and the results are further verified using phase-field simulations. The surrogate enables generation of ‘phase diagrams’ of the domain wall, conditional on initial structure. Subsequently, reinforcement learning is used to optimize tip-modification trajectories for obtaining desired domain wall structures in simulated environments utilizing the surrogate model for the environment dynamics. This workflow shows how automated data collection and autonomous agents can be orchestrated towards realizing domain wall manipulations with precision in scanning probe studies, and how such surrogates can aid in understanding domain wall interactions in ferroelectrics.

原文English
頁(從 - 到)456-466
頁數11
期刊Digital Discovery
3
發行號3
DOIs
出版狀態Published - 2024 2月 7

All Science Journal Classification (ASJC) codes

  • 化學(雜項)

指紋

深入研究「Physics-informed models of domain wall dynamics as a route for autonomous domain wall design via reinforcement learning」主題。共同形成了獨特的指紋。

引用此