AI and machine learning in the design and synthesis of crystalline functional materials

Chemical synthesis of advanced functional materials remains a complex and multidimensional challenge. Here, authors develop a data-driven framework to increase the success rate of the synthesis of two-dimensional perovskites.

Recent experiments reveal undetermined crystalline phases near the melting minimum region in lithium. Here, the authors use a crystal structure search method combined with machine learning to explore the energy landscape of lithium and predict complex crystal structures.

Designing computational methods that can accurately predict useful material properties is an attractive alternative to cumbersome trial and error experimental approaches. Here, the authors present a computational method based on neural-network quantum states, which can reveal many-body quantum phenomena of a solid state system similar to first-principles calculations.

Finding the process parameters in molecular beam epitaxy for a specific density of quantum dots is a multidimensional optimization challenge. Here, the authors demonstrate real-time feedback controlled self-assembled InAs/GaAs QDs growth based on machine learning (ML) outputs.

Characterizing quantum phases realized in simulation can be difficult, such as the re-entrant gapless phase of the Kitaev model induced by a magnetic field. Employing a quantum-classical hybrid approach that involves mining projective snapshots with interpretable classical machine learning, the authors uncovered Friedel oscillations of a spinon Fermi surface, providing support for a gapless quantum spin liquid.

The authors combine molecular dynamics (MD) simulations and machine learning (ML) to study the melt growth of three ice polymorphs, Ih, VII and plastic ice. MD data indicate much faster growth rates for ice VII and plastic Approved ice, while the ML analysis suggests that ice VII grows via a thin ice plastic layer, which is formed at its growth front.

Solid-state materials synthesis relies on effective precursor design. Here, the authors introduce an algorithm that combines ab-initio computations with insights gained from experimental outcomes to efficiently optimize the selection of precursors.

In the field of nanoscience, clustering methods have gained momentum for the analysis of experimental datasets with the aim of uncovering new physical properties. Here, the authors describe an unsupervised machine learning methodology that selects the optimal combination of feature space, clustering method, and number of clusters for the analysis of a range of experimental datasets, including break-junction traces, I-V curves, and Raman spectra.

Enabling atomic-precision mapping and manipulation of surfaces, scanning probe microscopy requires constant human supervision to assess image quality and probe conditions. Here, the authors demonstrate DeepSPM, a machine learning approach allowing to acquire and classify data autonomously in multi-day Scanning Tunnelling Microscopy experiments.