Deep Learning Models for Colloidal Nanocrystal Synthesis.

Area of Science:

• Materials Science
• Nanotechnology
• Artificial Intelligence

Background:

• Colloidal synthesis of nanocrystals involves complex chemistry, making it difficult to link synthesis parameters to material properties.
• Despite advances, precise control over nanocrystal size and shape remains a significant challenge in materials science.

Purpose of the Study:

• To develop a deep learning (DL) model for predicting nanocrystal size and shape based on synthesis parameters.
• To establish correlations between chemical reaction parameters and the physical properties of synthesized nanocrystals.
• To create a versatile tool for expediting the discovery and optimization of nanocrystal synthesis.

Main Methods:

• A DL-based synthesis model was developed using a dataset of 3508 recipes for 348 nanocrystal compositions.
• Nanocrystal size and shape data were extracted from transmission electron microscope images using a segmentation model trained with a semi-supervised algorithm.
• Reaction intermediate-based data augmentation and elaborated descriptors were employed to enhance model performance.

Main Results:

• The DL model accurately predicted nanocrystal size with a mean absolute error of 1.39 nm.
• The model achieved an 89% average accuracy in classifying nanocrystal shapes.
• The model demonstrated knowledge transfer capabilities across different nanocrystal types and identified key chemical influences on size.

Conclusions:

• The developed DL model provides a powerful and efficient method for predicting and optimizing nanocrystal synthesis.
• This approach significantly accelerates the development of high-quality nanocrystals by elucidating structure-property relationships.
• The study highlights the potential of AI in advancing materials synthesis and design.