Abstract
Supraparticles (SPs) are assemblies of smaller particles, and they form an interesting material class. One way through which these structures can be formed is self-assembly (SA) in spherical confinement, and what makes them unique is that they combine the properties of the smaller particles with collective properties arising from the length scale on which these smaller particles are ordered. Additionally, the limited number of particles in an SP enables them to form structures that are not found in bulk systems. An example of this is icosahedral symmetry, which is the equilibrium structure for SPs up to several hundreds of thousands of particles. Although these icosahedral structures have been investigated through computer simulations and several experimental techniques have been used to analyze them in 3D, the number of experimental datasets published is so limited that no statistically relevant conclusions have been drawn so far. The experimental technique most commonly applied to study them is scanning electron microscopy (SEM), but with this, only quantitative information about the surface of the SPs can be obtained. By using a combination of 3D confocal and stimulated emission depletion (STED) microscopy on extremely well-index-matched (within 0.002) fluorescent core-shell, colloidal silica spheres (of 442-478 nm in diameter with polydispersities below 1%), we obtained full 3D real-space datasets of tens of SPs within several hours. The structures were classified based on bond order parameters and deviations from local centrosymmetry, using an unsupervised machine learning model. From this, we are able to correctly classify structures that are commonly misidentified using SEM. Additionally, the quantitative real-space analysis gave experimental insights into the SA pathway and defect formation mechanisms of mostly icosahedral SPs.