Sacchi, SaraSaraSacchiVaranasi, AnirudhAnirudhVaranasiDegraeve, RobinRobinDegraeveVici, AndreaAndreaViciMolinaro, GiorgioGiorgioMolinaroFranco, JacopoJacopoFrancoRoussel, PhilippePhilippeRousselKaczer, BenBenKaczerMerckling, ClementClementMerckling2026-06-082026-06-0820260038-1101https://imec-publications.be/handle/20.500.12860/59605Time-Dependent Dielectric Breakdown (TDDB) remains a critical reliability challenge in advanced CMOS technologies using thin /High-k oxides. While extensive research focused on the formation of conductive filaments and the physics and statistics of soft breakdown and hard breakdown events, the intermediate wear-out phase — where a localized leakage path gradually increases in conductivity — has not been thoroughly analyzed or modeled. Firstly, this work addresses this gap by experimentally isolating and analyzing the wear-out phase with a Machine learning-assisted analysis, revealing key statistical features of wear-out and its dependence on stress voltage. Secondly, a Monte Carlo-implemented Markov model is used to simulate the localized degradation of a one-defect percolation path by means of a thermally activated defect creation and deactivation/annealing process, governed by an Arrhenius-like transition probability function. Simulations qualitatively reproduce the observed experimental degradation trends, with discrepancies in voltage dependence and initial defect accumulation, highlighting the need for a more nuanced approach, including statistical distributions of atomic bond strengths.engJournal article10.1016/j.sse.2025.109266WOS:001614997700002