A Deep Learning Framework Discovers Compositional Order and Self-Assembly Pathways in Binary Colloidal Mixtures.

Area of Science:

• Materials Science
• Colloidal Science
• Machine Learning Applications

Background:

• Binary colloidal superlattices (BSLs) are promising for advanced materials synthesis via self-assembly.
• Mechanistic understanding of 3D BSL self-assembly is limited by characterization challenges for complex structures.
• Existing methods lack systematic, single-particle level analysis of BSL formation.

Purpose of the Study:

• To present a novel framework for quantitative characterization of colloidal crystal structures during BSL self-assembly.
• To systematically describe BSL formation at the single-particle level using branched graphlet decomposition and deep learning.
• To elucidate the influence of particle size ratio and interparticle potential on BSL self-assembly.

Main Methods:

• Utilized branched graphlet decomposition to create high-dimensional neighborhood graphs for each particle.
• Quantified structural and compositional order, including defects, using these neighborhood graphs.
• Employed deep autoencoders to translate graphs into low-dimensional manifolds for relationship inference.

Main Results:

• Demonstrated the framework on in silico DNA-functionalized particle systems.
• Showed that small interparticle size disparities (radius ratios 0.8-0.95) promote defect-free BSLs (e.g., BCC-CsCl, FCC-CuAu, IrV).
• Identified nonclassical nucleation pathways where BSLs form from dense amorphous precursors in size-disparate mixtures.

Conclusions:

• The developed framework enhances mechanistic understanding of binary colloidal mixture self-assembly.
• It facilitates the engineering and growth of defect-free BSLs by optimizing particle size and interactions.
• This approach opens new avenues for designing multifunctional materials through controlled colloidal self-assembly.