Closed-Loop Solid-State Synthesis Planning for Materials Discovery With Large Language Models

Area of Science:

• Materials Science
• Computational Chemistry
• Artificial Intelligence

Background:

• Accelerating materials discovery is hindered by challenges in developing reliable synthesis routes.
• Extracting and structuring synthesis information from scientific literature is complex.

Purpose of the Study:

• To establish a large language model-based framework for predicting and optimizing material synthesis conditions.
• To accelerate the design and discovery of new materials through data-driven approaches.

Main Methods:

• Systematic extraction of synthesis information (compounds, precursors, parameters) from 4407 solid-state synthesis papers.
• Development of a retrieval-augmented generation (RAG) model to generate candidate synthesis recipes.
• Benchmarking generated recipes against literature data and experimental validation.

Main Results:

• The framework achieved strong agreement between generated and experimentally reported synthesis conditions.
• Successfully synthesized novel oxy-selenide solid-state electrolyte candidates using the model's predictions.
• Demonstrated iterative refinement of synthesis parameters, leading to phase-pure products with minimized trial-and-error.

Conclusions:

• The developed framework offers a data-driven, feedback-optimized route for accelerating synthesis design.
• This approach provides a generalizable paradigm for integrating language models into experimental materials research.
• Enables efficient discovery and synthesis of complex materials.