Accelerating Structure-Property Relationship Discovery with Multimodal Machine Learning and Self-Driving Microscopy

Area of Science:

• Materials Science
• Nanoscience
• Machine Learning

Background:

• Conventional microscopy relies on manual sampling, limiting data diversity and discovery potential in functional materials.
• Correlating nanoscale structure with functional properties requires advanced, efficient data acquisition techniques.

Purpose of the Study:

• To develop an integrated framework for autonomous microscopy and adaptive data acquisition using machine learning.
• To create a structure-property relationship map for functional materials, exemplified by halide perovskite films.

Main Methods:

• Integration of autonomous microscopy with dual-novelty deep kernel learning (DN-DKL) for adaptive data acquisition.
• Utilizing a dual variational autoencoder (VAE) for representation learning and embedding structure-property data.
• Application of conductive atomic force microscopy for investigating halide perovskite films.

Main Results:

• The framework efficiently collected large spectral datasets by guiding microscopy to novel regions.
• Distinct hysteresis behaviors in halide perovskite films were linked to specific nanoscale structural motifs.
• Identified grain boundary junction points exhibiting bias-dependent hysteresis and asymmetric grain boundaries suppressing charge transport.

Conclusions:

• The developed framework accelerates scientific discovery in functional materials by combining self-driving microscopy and machine learning.
• Establishes a general strategy for leveraging AI to uncover complex structure-property relationships.
• Provides new insights into the factors influencing charge transport and hysteresis in perovskite films.