Ruthenium nanofilms were synthesized on hydrogen-terminated Si(100) substrates using tricarbonyl(trimethylenemethane)ruthenium, Ru(TMM)(CO)3, as the precursor under varying conditions to explore growth mechanisms and electronic properties. In the absence of reactants, pulse dosing at 150–260 °C revealed three distinct regimes: adsorption of Ru fragments above ∼150 °C, formation of conductive RuSix islands (∼20 nm) with narrow bandgaps near 190 °C, and thick metallic Ru films via CVD above 230 °C. Nanoscale characterization using STM/scanning tunneling spectroscopy (STS), Kelvin probe force microscopy, and XPS confirmed morphological and electronic modifications, while statistical analysis of STS spectra via unsupervised clustering supported regime differentiation. Complementary experiments introduced N2/O2, N2/H2, and N2 reactants at 190 °C to mimic atomic layer deposition (ALD)-like submonolayer growth. Reactant chemistry strongly influenced nucleation and bonding states, producing diverse surface phases such as RuSix and RuOx. Machine-learning (ML) workflows employing principal component analysis -based dimensionality reduction and clustering identified distinct domain counts as a function of growth environment, including RuO2 (∼30% coverage under N2/O2), consistent with XPS observations. This data-driven approach demonstrates that ML can objectively capture reactant-dependent variations, enabling optimization of Ru ALD processes and nanofilm uniformity.
Bolotov et al. (Tue,) studied this question.