Inorganic Materials Synthesis Planning with Literature-Trained Neural Networks.

Area of Science:

• Materials Science
• Computational Chemistry
• Artificial Intelligence

Background:

• Accelerating materials design and discovery relies on novel data sources.
• Existing synthesis planning methods utilize historical, experimental, and computed data.
• Connecting scientific literature to synthesis insights remains a challenge.

Purpose of the Study:

• To develop an automated, unsupervised method for extracting inorganic synthesis information from scientific literature.
• To generate synthesis insights for inorganic materials using natural language processing and machine learning.
• To predict precursors for perovskite materials and screen novel compounds for synthesizability.

Main Methods:

• Utilizing word embeddings from language models on natural language text.
• Applying a named entity recognition (NER) model.
• Training a conditional variational autoencoder (CVAE) on recognized entities to generate synthesis information.

Main Results:

• Successfully predicted precursors for two perovskite materials using data from over a decade prior to their synthesis.
• Demonstrated that the model learns material representations linked to synthesis properties.
• Showcased that the model's predictions complement existing thermodynamic knowledge.

Conclusions:

• The developed method effectively connects scientific literature to inorganic synthesis insights.
• The AI-driven approach can predict material precursors and aid in synthesizability screening.
• This technique holds potential for accelerating the discovery of novel inorganic materials.