$ exttt{DiffSyn}$: A Generative Diffusion Approach to Materials Synthesis Planning

2025-09-21

AI Summary

Crystalline materials synthesis—e.g., zeolites—faces challenges including high-dimensional parameter spaces, complex structure–synthesis relationships, and lengthy experimental cycles. To address these, we propose the first generative diffusion model specifically designed for materials synthesis planning. It explicitly models the non-bijective mapping between target structures (including organic structure-directing agents) and multimodal synthesis conditions, enabling pathway discrimination and optimization under phase-competition scenarios. Trained on large-scale literature data and integrated with density functional theory (DFT) to verify thermodynamic feasibility of generated routes, the model achieves unprecedented Si/Al$_{ ext{ICP}}$ = 19.0 in UFI zeolite synthesis—substantially exceeding prior records—and markedly enhances thermal stability. This work establishes a new paradigm for data-driven, inverse design of crystalline materials synthesis.

Technology Category

• Machine Learning: Deep Generative Models & Autoencoders
• Natural Language Processing: Code Generation / Program Synthesis from Natural Language
• Planning, Routing, and Scheduling: Planning with Language Models

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Web Mining and Content Analysis: Web data generation and simulation
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications

Abstract

The synthesis of crystalline materials, such as zeolites, remains a significant challenge due to a high-dimensional synthesis space, intricate structure-synthesis relationships and time-consuming experiments. Considering the one-to-many relationship between structure and synthesis, we propose $ exttt{DiffSyn}$, a generative diffusion model trained on over 23,000 synthesis recipes spanning 50 years of literature. $ exttt{DiffSyn}$ generates probable synthesis routes conditioned on a desired zeolite structure and an organic template. $ exttt{DiffSyn}$ achieves state-of-the-art performance by capturing the multi-modal nature of structure-synthesis relationships. We apply $ exttt{DiffSyn}$ to differentiate among competing phases and generate optimal synthesis routes. As a proof of concept, we synthesize a UFI material using $ exttt{DiffSyn}$-generated synthesis routes. These routes, rationalized by density functional theory binding energies, resulted in the successful synthesis of a UFI material with a high Si/Al$_{ ext{ICP}}$ of 19.0, which is expected to improve thermal stability and is higher than that of any previously recorded.

Problem

• Generating probable synthesis routes for crystalline materials
• Capturing complex structure-synthesis relationships in materials science
• Differentiating among competing phases for optimal synthesis planning

Innovation

• Generative diffusion model for materials synthesis planning
• Generates synthesis routes from structure and template
• Captures multi-modal structure-synthesis relationships