The Search for BaTiO₃-Based Piezoelectrics With Large Piezoelectric Coefficient Using Machine Learning.

Area of Science:

• Materials Science
• Solid State Physics
• Computational Materials Science

Background:

• Piezoelectric materials, particularly those based on barium titanate (BaTiO3), are crucial for various electronic applications.
• Discovering new compositions with enhanced piezoelectric properties, specifically a large piezoelectric coefficient (d33), remains a key research objective.

Purpose of the Study:

• To employ a data-driven strategy to accelerate the discovery of novel BaTiO3-based piezoelectric materials with high d33.
• To develop and validate a computational framework combining surrogate modeling and adaptive design for materials discovery.

Main Methods:

• Utilized a surrogate model to predict d33 values and associated uncertainties for potential piezoelectric compounds.
• Implemented an adaptive design strategy, iterating five times, to select optimal new compounds for synthesis based on model predictions.
• Synthesized four new compounds in each iteration, guided by two distinct design selection criteria.

Main Results:

• Identified (Ba0.85Ca0.15)(Ti0.91Zr0.09)O3 as the best newly discovered compound, exhibiting a d33 of 362 pC/N.
• The best compound in the training data, BCT-0.5BZT, showed a significantly higher d33 of approximately 610 pC/N.
• The developed surrogate model accurately described most of the d33 data, but struggled to precisely fit the exceptionally high value of BCT-0.5BZT.

Conclusions:

• The data-driven adaptive design strategy is effective in exploring the materials space for piezoelectric compounds.
• The exceptional performance of BCT-0.5BZT suggests limitations of purely data-driven models for extreme property prediction.
• Future research should integrate physics-based insights with data-driven approaches for more robust materials discovery.