Machine-learning of piezoelectric coefficients for wurtzite crystals

ABSTRACT

A handful of piezoelectrics cover all of today’s technologies, so there is a need to expand the range of accessible piezoelectric properties to enable future fundamental and technological advances. Focusing on Space Group 186, we have computed the piezoelectric properties of 47 materials, and deployed machine-learning models to reveal the quantities that control them. We have found reasonable linear models for the case of wurtzites, from a dataset of only 19 entries. Some of these models are based on easily retrievable features, while others involve features requiring additional computations. In the case of non-wurtzite materials, the statistical learning has not found any meaningful correlations, since the diversity of structures and compositions in a set of 26 non-wurtzite entries obliterates the influence of features on piezoelectric coefficients. The linear models presented for the wurtzite coefficients can potentially be used to inform future searches for wurtzite-structured alloys with higher piezoelectric coefficients.

KEYWORDS:

• Piezoelectrics
• machine learning
• wurtzite
• Aluminum Nitride

Acknowledgments

The data collection portion of this work was carried out with support from the National Science Foundation Grant No. DMREF-1534503.

Disclosure statement

No potential conflict of interest was reported by the authors.