Predicting the morphology of cobalt, copper, and ruthenium on TaN for interconnect metal deposition.

Downscaling of metal interconnects has become a bottleneck in the back-end-of-line processing of CMOS semiconductor devices. The conductivity of the commonly used Cu metal drastically reduces at nanoscale, limiting its use as the devices shrink, as the metal starts forming non-conductive islands. This has led to the requirement for new interconnect metals such as Co and Ru, as they have a much higher tendency to form conductive films with horizontal growth at nanoscale. Understanding how the morphology of interconnects depends on the metals' atomistic properties and their interactions at diffusion barrier layers is necessary to continue the rapid development of interconnects. In this study, we have used first-principles density functional theory (DFT) relaxations, ab initio molecular dynamics, and neural network machine learning potentials (MLPs) to investigate how the morphology of Cu, Co, and Ru differs on TaN substrates. We investigate the binding of single metal atoms and four atom clusters to obtain the metals' substrate binding energy and metal-metal interaction energies. The morphology of the metals was then investigated by 15 ps molecular dynamics simulations with DFT and 5 ns with MLPs using larger metal structures that can display 2D and 3D morphologies. Comparing the binding energies with the obtained morphology allows us to demonstrate how the balance of metal-substrate and metal-metal interactions determines the morphology, while the MLP simulations allow longer timescale processes to be included. These insights help in the development of morphology predictors, allowing a rapid method for screening new interconnect materials with targeted horizontal growth on substrates used in semiconductors.