Unveiling strain-dependent adhesion behavior at TaN-Cu interface using machine learning interatomic potential

doi:10.1016/j.apsusc.2025.162558

2025-04-01

PRE

DOI：10.1016/j.apsusc.2025.162558

原文链接

原文求助

摘要

En 中文

Copper (Cu) is commonly used as a low-resistance interconnect in semiconductors, but it tends to diffuse into silicon (Si) and silicon dioxide (SiO2). To mitigate this, diffusion barriers like tantalum nitride (TaN) are used, designed to endure thermal and electromigration stresses. Understanding adhesion failure at the TaN/Cu interface under strain is key for reliable long-lasting performance. In this study, we investigated TaN/Cu interfacial adhesion using steered molecular dynamics (SMD) simulations with CHGNet, a machine learning interatomic potential (MLIP) trained on 75,000 configurations of TaN, Cu, and TaN/Cu interfaces. SMD simulations were performed on various Cu and TaN interfaces under biaxial strains from-30 % to 30 %. Our results show that the maximum adhesion force depends on the specific TaN and Cu surfaces, with biaxial compressive strain generally increasing the force required for detachment. This is supposed to be due to the increase in atomic packing density at the TaN/Cu interface caused by the biaxial compressive strain. Additionally, the TaN(111)/Cu (111) interface exhibited a superior average adhesion maximum force of 0.074 eV/A3 compared to other surfaces under various biaxial strain conditions. This is presumed to occur because the TaN(111)/Cu(111) interface has a larger lattice mismatch, leading to greater structural deformation, which increases the initial interface energy and results in a higher maximum adhesive force. We believe that our findings can guide the selection of an optimal interface between TaN and Cu. In addition, this study highlights the methodology for finding stable interfaces under various biaxial stress conditions similar to operating environments, offering valuable insights for further research into the reliable performance of diffusion barriers.

Keyword:

Adhesion behavior

TaN-Cu interface

Steered molecular dynamics

Machine learning interatomic potential

Strain dependency