Experimental search for high-temperature ferroelectric perovskites guided by two-step machine learning

Abstract

Experimental search for high-temperature ferroelectric perovskites is a challenging task due to the vast chemical space and lack of predictive guidelines. Here, we demonstrate a two-step machine learning approach to guide experiments in search of xBiO3–(1 − x)PbTiO3-based perovskites with high ferroelectric Curie temperature. These involve classification learning to screen for compositions in the perovskite structures, and regression coupled to active learning to identify promising perovskites for synthesis and feedback. The problem is challenging because the search space is vast, spanning ~61,500 compositions and only 167 are experimentally studied. Furthermore, not every composition can be synthesized in the perovskite phase. In this work, we predict x, y, Me′, and Me″ such that the resulting compositions have both high Curie temperature and form in the perovskite structure. Outcomes from both successful and failed experiments then iteratively refine the machine learning models via an active learning loop. Our approach finds six perovskites out of ten compositions synthesized, including three previously unexplored {Me′Me″} pairs, with 0.2Bi(Fe0.12Co0.88)O3–0.8PbTiO3 showing the highest measured Curie temperature of 898 K among them.

Accelerated Search for BaTiO3‐Based Ceramics with Large Energy Storage at Low Fields Using Machine Learning and Experimental Design

“…There has recently been much interest in employing machine learning (ML) and optimization methods to guide experimental synthesis to find materials with targeted properties. [8][9][10][11][12][13][14][15][16][17][18] The state-of-art in this field is to use an iterative approach, which is largely data-driven, starting from a set of features or material descriptors based on material knowledge to construct a surrogate model learned from data. [19,20] Ideas and methods from decision theory and experimental design then provide the means to make optimal decisions of the experiments or materials to test next.…”

Data‐driven glass/ceramic science research: Insights from the glass and ceramic and data science/informatics communities

“…17 According to Meredig, AI is most credible to domain experts when it is interpretable, not a black box. An example of AI-driven experimental design is Balachandran et al's work on high-temperature ferroelectrics, wherein the goal was to identify compositions with high Curie temperatures that also should crystallize in the perovskite phase.…”

“…Lifetime and degradation science 40 is an approach to these long lifetime, complex system, degradation pathway problems, that adds data science and big data analytics approaches to the more traditional, hypothesis-driven, laboratorybased research methods. One of the grand challenges for PV has been achieving these long lifetimes while extending the lifetime of PV modules to 50 years.…”

“…An example of AI-driven experimental design is Balachandran et al's work on high-temperature ferroelectrics, wherein the goal was to identify compositions with high Curie temperatures that also should crystallize in the perovskite phase. 17 According to Meredig, AI is most credible to domain experts when it is interpretable, not a black box. AI must also be able to handle typical issues in materials design, such as small, sparse materials data; uncertainty; and high dimensionality of many experimental design spaces, while leveraging the laws of physics.…”

“…Achieving this requires an understanding of the multiple simultaneous and sequential degradation mechanisms that are active. Lifetime and degradation science 40 is an approach to these long lifetime, complex system, degradation pathway problems, that adds data science and big data analytics approaches to the more traditional, hypothesis-driven, laboratorybased research methods.…”