Data-driven curation of organometallic precursor boiling points via natural language processing

The development of high-performance semiconductor devices relies on precise control of thin-film deposition techniques, notably Chemical Vapor Deposition (CVD) and Atomic Layer Deposition (ALD). Central to these processes are organometallic precursors, whose boiling points control volatility and vapor delivery conditions, thereby constraining processing windows and precursor feasibility. However, boiling point data for these precursors are scattered across unstructured literature and underrepresented in commercial databases. To address this gap, we developed a Natural Language Processing (NLP) pipeline to systematically extract, validate, and organize boiling point data from 5,442 peer-reviewed articles since 2000. Using the ChemDataExtractor toolkit, we curated 1,902 candidate data points, followed by rigorous multistage filtration to remove duplicates, non-standard nomenclature, non-organometallic compounds, and erroneous entries. The validated dataset comprises 75 organometallic precursors with boiling points from 21 °C to 181 °C. Comparative analysis with CAS Registry and SciFinder revealed that over 80 % of these data points were previously undocumented, underscoring the novelty of the dataset. This work demonstrates the utility of NLP-driven literature mining in building chemically precise databases and offers a scalable framework to accelerate precursor characterization and materials discovery in semiconductor manufacturing and ceramic thin-film applications.