Machine Learning-Assisted Rapid Scalable Synthesis and Assembly Evolution of Lithographic Metal-Oxo Clusters

The controllable synthesis of metal-oxo clusters (MOCs) is often limited by competing reaction pathways and strongly coupled variables, which hinder the synthetic efficiency and scalability. Here, we develop a machine-learning-assisted reaction-space optimization strategy that defines an effective synthetic window through data-driven modeling and key parameter identification, enabling precise control over complex reaction systems. Within this framework, an unprecedented racemic Sn₁₂Ti₄ cluster rapidly formed within several seconds at room temperature, and 266 g of phase-pure crystalline product was obtained within 15 min. Under the structure-directing effect of trimethylolpropane ligands, tin species underwent rapid hydrolysis-condensation, while titanium atoms were incorporated through oxo bridge reconstruction. The system rapidly converged to the thermodynamically favored Sn₁₂Ti₄ structure, as supported by mass spectrometry and theoretical calculations, enabling highly selective synthesis within seconds. Through modulating the steric hindrance and electronic properties of terminal ligands, truncated tetrahedral Sn₁₂Ti₄ units achieved assembly evolution from isolated clusters to supramolecular two-dimensional (2D) layers and three-dimensional (3D) networks via halogen or hydrogen bonding interactions and eventually homochiral metal-organic frameworks (MOFs). Furthermore, the Sn₁₂Ti₄ cluster exhibited interesting lithographic applications to achieve a high-resolution performance with a line width of 9.83 nm. This work establishes a generalizable machine-learning-assisted route toward rapid, scalable synthesis of functional metal-oxo clusters.