Artificial Intelligence Predicted OSDAs Enable Direct Synthesis of Interlayer-Expanded Zeolites.

Area of Science:

• Materials Science
• Chemistry
• Crystallography

Background:

• Zeolite crystallization is a complex, metastable process.
• Directed synthesis of specific zeolite frameworks is challenging due to poorly understood mechanisms.
• Organic structure-directing agents (OSDAs) are crucial for controlling zeolite framework formation, but their discovery relies heavily on inefficient trial-and-error screening.

Purpose of the Study:

• To develop a novel, domain knowledge-informed machine learning model for predicting OSDAs.
• To overcome the limitations of traditional descriptor-based machine learning models in screening OSDAs for novel zeolite frameworks.
• To enable the efficient and directed synthesis of new zeolite materials.

Main Methods:

• Development of a domain knowledge-informed machine learning model (ECNU-Zeoformer).
• Integration of an end-to-end architecture with active learning strategies.
• Prediction of OSDA-zeolite binding energies for effective OSDA selection.

Main Results:

• Successful synthesis of three novel zeolites: ECNU-30, ECNU-34, and ECNU-40.
• The ECNU-Zeoformer model demonstrated superior predictive performance compared to traditional methods.
• The model exhibited excellent generalizability across different zeolite framework topologies.

Conclusions:

• The developed machine learning model effectively predicts OSDAs, enabling the discovery of new zeolites.
• This approach significantly advances the directed synthesis of zeolites by replacing trial-and-error screening with accurate computational prediction.
• The ECNU-Zeoformer represents a breakthrough in materials discovery for novel zeolite frameworks.