UC Berkeley

Colloidal self-assembly (colloidal SA) is the process by which particles in solution spontaneously organize into an ordered structure. The spontaneous self-organization central to colloidal SA enables "bottom-up" materials synthesis, which would allow for manufacturing advanced, highly ordered crystalline structures in an inherently parallelizable and cost-effective manner. Thus, colloidal SA can create new avenues for highly scalable, economical manufacturing of novel metamaterials with unique optical, electrical, or mechanical properties. Colloidal SA is an inherently stochastic (i.e., random) process prone to kinetic arrest due to particle Brownian motion. This leads to variability in materials synthesis and possibly high defect rates, which can severely compromise the viability of using SA to manufacture advanced materials reproducibly. Successful implementation of colloidal SA thus critically hinges on the ability to avoid defective, kinetically arrested configurations and consistently reach highly-ordered, often defect-free states that tend to exist within global minima on the free energy landscape.

The thermodynamic and kinetic driving forces that govern colloidal SA thus need to be precisely modulated -- by actively exploiting intermolecular forces, selective template or surface geometries, and/or external fields such as temperature and pressure -- to direct colloidal SA systems consistently and efficiently towards mass-producible structures and materials. Two major strategies for this precise modulation are particle design, which involves designing the colloidal SA system such that specific inter-particle interactions ensure the high probability realization of a desired configuration, and control, which seeks to modulate external actuators systematically based on real-time measurements in order to induce global colloidal SA configuration changes. The primary objective of this thesis is to enable more effective particle design and control of colloidal SA systems. To this end, this thesis investigates strategies based on machine learning and optimal control for quantifying and classifying colloidal SA system states, learning tractable stochastic dynamical models of colloidal SA dynamics, and learning control policies that dynamically change external actuators to guide colloidal SA. The insights gained from these methods provide a deeper mechanistic understanding of colloidal SA and contribute to an ever-developing archive of methods that can be used or expanded upon to achieve reproducible colloidal SA.