To switch or not to switch - a machine learning approach for ferroelectricity.

Area of Science:

• Materials Science
• Physics
• Chemistry

Background:

• Experimental data in materials science are increasingly complex, often depending on multiple interdependent parameters.
• Analyzing multidimensional datasets for material properties like ferroelectric switching presents significant challenges.

Purpose of the Study:

• To introduce a new two-dimensional approach for representing and analyzing material hysteresis.
• To demonstrate how this method improves the interpretation of ferroelectric switching phenomena.
• To show the applicability of this approach in machine learning for materials data analysis.

Main Methods:

• Developed a two-dimensional data representation for hysteretic response.
• Utilized ferroelectric polarization as a model system.
• Applied machine learning algorithms (clustering, neural networks) to the new data representation.

Main Results:

• The two-dimensional approach allows for more transparent differentiation between phenomena like charge trapping and ferroelectricity.
• The method successfully differentiates between various origins of hysteresis.
• Machine learning models accurately inferred sample temperature from hysteresis morphology.

Conclusions:

• The proposed two-dimensional representation offers a more robust and insightful method for analyzing complex material hysteresis.
• This approach facilitates advanced data analysis and machine learning applications in materials science.
• It provides a pathway to deeper understanding of material properties influenced by multiple stimuli.