A Large-Scale Nanocrystal Database with Aligned Synthesis and Properties, Enabling Generative Inverse Design.

Area of Science:

• Materials Science
• Artificial Intelligence
• Nanotechnology

Background:

• Nanocrystal synthesis relies heavily on trial-and-error due to complex parameter-property relationships.
• Scarcity of high-quality datasets hinders deep learning applications in inverse design for nanocrystals.

Purpose of the Study:

• To develop an AI model for extracting structured nanocrystal synthesis data from unstructured literature.
• To create a large-scale database of nanocrystal synthesis-property relationships.
• To enable generative inverse design of nanocrystal synthesis routes.

Main Methods:

• Developed NanoExtractor, a large language model (LLM) with data augmentation for literature data extraction.
• Constructed a Nanocrystal Synthesis-Property (NSP) database with ~160,000 entries.
• Developed NanoDesigner, an LLM for generative inverse synthesis design.

Main Results:

• NanoExtractor achieved 92% accuracy, outperforming other LLMs.
• The NSP database contains extensive aligned synthesis-property data.
• NanoDesigner achieved an F1 score of 0.85 and was experimentally validated.
• Identified a critical precursor concentration for MgF2 nanocrystal synthesis.

Conclusions:

• Bridged the gap between unstructured literature and data-driven synthesis for nanocrystals.
• Established a human-AI collaborative paradigm for accelerated materials discovery.
• Demonstrated the practical utility of AI in designing specific nanocrystal systems.