Element selection for crystalline inorganic solid discovery guided by unsupervised machine learning of experimentally...

Area of Science:

• Materials Science
• Inorganic Chemistry
• Computational Chemistry

Background:

• Element selection dictates synthetic chemistry outcomes, influencing material properties.
• Existing data on isolable materials is vast, hindering efficient discovery of new chemistries.
• Predicting new crystalline materials relies on understanding chemical structure and bonding.

Purpose of the Study:

• To develop a machine learning model for identifying promising element combinations for new crystalline materials.
• To guide the exploration of quaternary phase fields with two anions for novel lithium solid electrolytes.
• To accelerate the discovery of advanced materials for energy storage.

Main Methods:

• Unsupervised machine learning to analyze patterns in element combinations yielding crystalline inorganic materials.
• Prioritization of quaternary phase fields containing two anions for synthetic exploration.
• Collaborative workflow integrating computational prediction and experimental synthesis.

Main Results:

• The machine learning model successfully captured complex similarity patterns between element combinations.
• The model guided the identification of Li3.3SnS3.3Cl0.7, a novel lithium solid electrolyte.
• The new material exhibits a low-barrier ion transport pathway due to its defect-stuffed wurtzite structure.

Conclusions:

• Unsupervised machine learning is effective in navigating vast chemical spaces for materials discovery.
• This approach accelerates the identification of promising candidates for solid-state electrolytes.
• The discovered Li3.3SnS3.3Cl0.7 demonstrates potential for next-generation battery technologies.