First-Principles-Informed Unified Transport Model for Resolving Parameter Non-Identifiability in Sub-10 nm Ruthenium Interconnects

Abstract

As backend interconnect dimensions approach the sub-10 nm regime, ruthenium (Ru) is increasingly attractive as a barrier-lean conductor, yet linking process conditions to transport performance remains difficult because transport parameters extracted from resistivity-thickness scaling are strongly non-identifiable. In particular, surface specularity (p), grain-boundary reflection (R), and effective dead-layer thickness (t_dead ) are highly coupled and can produce similarly accurate yet physically inconsistent fits. To resolve this ambiguity, we present a density functional theory (DFT)-anchored compact transport framework that resolves this ambiguity by imposing DFT-derived and microstructure-informed parameter bounds. The model unifies Fuchs–Sondheimer surface scattering, Mayadas–Shatzkes grain-boundary scattering, dead-layer area loss, and a smooth quantum-crossover correction for ultrathin Ru. Simultaneous calibration to three experimentally distinct Ru process conditions (conventional, sputtered, and textured ALD; 6 nm to 50 nm) yields root-mean-square errors below 0.5 µΩ · cm while increasing the local Fisher identifiability metric for the coupled p–R–t_dead subspace relative to unconstrained fitting. The extracted trends establish a process-relevant scaling hierarchy: lowering R through texture engineering provides the largest resistivity benefit (∼30%), dead-layer reduction is secondary but substantial, and specularity improvements show diminishing returns beyond p ≈ 0.3. The resulting resistivity-thickness-dead-layer design maps provide quantitatively constrained targets for materials selection and design-technology co-optimisation (DTCO) of next-generation Ru interconnects.