Unraveling the adhesion characteristics of ruthenium as an advanced metal interconnect material using machine learning potential

Abstract

As the dimensions of semiconductor devices continue to shrink and the complexity of the manufacturing process increases, metal interconnects that link different parts of integrated circuits have become a key factor in determining device performance, speed, and power efficiency. Until recently, copper (Cu) has been used as a metal interconnect material, but due to a sharp increase in resistance at sub-10 nm, ruthenium (Ru) is considered a promising candidate for advanced interconnect materials. In order to employ Ru as the interconnect material, it is necessary to secure adhesion characteristics with amorphous SiO₂, which is used as a representative insulator, but there is little understanding of the interfacial adhesion especially within an atomistic perspective. This study combines machine learning potential and steered molecular dynamics to provide atomic-level understanding of the adhesion properties of Ru/SiO₂ interfaces. It was found that the presence of hydroxyl groups on the surface of SiO₂ significantly affects the adhesion and the removal of hydrogen atoms from the hydroxyl groups is remarkably effective in increasing adhesion, even under excessive conditions. The analysis of the bonding characteristics between Ru and interfacial atoms of SiO₂ suggests that the degree of bonding between Ru and oxygen atoms is crucial for adhesion, and that the adhesion characteristics can be predicted through the bond order of interfacial atoms.