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

Area of Science:

• Materials Science
• Solid State Chemistry
• Computational Materials Science

Background:

• Discovering high-temperature ferroelectric perovskites is hindered by vast chemical spaces and limited predictive models.
• Experimental synthesis is challenging as not all compositions form the desired perovskite structure.

Purpose of the Study:

• To develop a machine learning strategy for efficiently guiding experimental searches for novel high-temperature ferroelectric perovskites.
• To predict compositions within the xBi2O3-(1-x)PbTiO3 system that exhibit high ferroelectric Curie temperatures and form perovskite structures.

Main Methods:

• A two-step machine learning approach combining classification and regression with active learning.
• Iterative refinement of models using outcomes from both successful and failed experimental syntheses.
• Prediction of compositional parameters (x, y, Me', Me″) for targeted perovskite synthesis.

Main Results:

• Successfully identified six novel perovskite compositions out of ten synthesized.
• Discovered three previously unexplored {Me'Me″} pairs.
• Identified 0.2Bi(Fe0.12Co0.88)O3-0.8PbTiO3 as a lead candidate, exhibiting a Curie temperature of 898 K.

Conclusions:

• The machine learning approach significantly enhances the efficiency of discovering high-temperature ferroelectric perovskites.
• This strategy effectively navigates the complex compositional landscape, reducing experimental trial and error.
• The identified materials and predictive models pave the way for future research in ferroelectric materials design.