A Statistical Understanding of the Growth and Etching of Anisotropic Colloidal Gold Nanoparticles Assisted by Machine Learning

Colloidal gold nanoparticles (AuNPs) have been widely studied due to their various chemical and biomedical applications. Recently, anisotropic AuNPs, such as Au nanorods, nanodisks, and triangular nanoprisms, have received much attention, due to their shape-specific, tunable plasmonic properties ideal for a variety of potential applications. While reliable colloidal pathways to the synthesis of many anisotropic AuNPs have been developed, the mechanisms leading to the formation of these shapes remain a matter of debate. To better elucidate the mechanism of the formation of anisotropic Au nanoparticles, statistical level information on the evolution of the shapes of AuNPs during their growth is necessary. Previously, shapes of AuNPs and their evolution during colloidal growth and etching have been studied by transmission electron microscopy (TEM). However, few previous studies were able to analyze the shapes of AuNPs at a population level (>1000 nanoparticles), due to the inefficiency of manual analysis of TEM data. We have previously developed AutoDetect-mNP, an algorithm that can automatically extract and classify the shapes of nanoparticles from TEM data with minimum human intervention, enabling the analysis of the shapes of thousands of nanoparticles in a matter of minutes. In this work, applying AutoDetect-mNP, we will demonstrate the mechanism and thermodynamics governing the formation of anisotropic Au NPs based on statistical analysis, using Au triangular nanoprisms as a model system. By analyzing the shapes of Au triangular nanoprisms at different stages of their growth and etching, on a statistically significant level, we will attempt to elucidate the formation of anisotropic AuNPs as a function of the relative reactivities and stability of different crystal facets. In this way, we hope to provide a systematic and quantitative understanding of how anisotropic shapes form during the growth of AuNPs. Further, we believe that this method of statistical analysis of nanoparticles shapes can be generalized to nanoparticle systems beyond AuNPs, including, for example, semiconductor quantum dots, and hence provides a potential solution to a range of fundamental questions in nanosciences.