Nucleation in atomic crystallization remains poorly understood, despite advances in classical nucleation theory. The nucleation process has been described to involve a nonclassical mechanism that includes a spontaneous transition from disordered to crystalline states, but a detailed understanding of dynamics requires further investigation. In situ electron microscopy of heterogeneous nucleation of individual gold nanocrystals with millisecond temporal resolution shows that the early stage of atomic crystallization proceeds through dynamic structural fluctuations between disordered and crystalline states, rather than through a single irreversible transition. Our experimental and theoretical analyses support the idea that structural fluctuations originate from size-dependent thermodynamic stability of the two states in atomic clusters. These findings, based on dynamics in a real atomic system, reshape and improve our understanding of nucleation mechanisms in atomic crystallization.

Liquid‐phase transmission electron microscopy (TEM) offers a real‐time microscopic observation of the nanometer scale for understanding the underlying mechanisms of the growth, etching, and interactions of colloidal nanoparticles. Despite such unique capability and potential application in diverse fields of analytical chemistry, liquid‐phase TEM studies rely on information obtained from the limited number of observed events. In this work, a novel liquid cell with a large‐scale array of highly ordered nanochambers is constructed by sandwiching an anodic aluminum oxide membrane between graphene sheets. TEM analysis of colloidal gold nanoparticles dispersed in the liquid is conducted, employing the fabricated nanochamber array, to demonstrate the potential of the nanochamber array in quantitative liquid‐phase TEM. The independent TEM observations in the multiple nanochambers confirm that the monomer attachment and coalescence processes universally govern the overall growth of nanoparticles, although individual nanoparticles follow different growth trajectories.