Failure mechanisms in semiconductor devices under coupled thermal and mechanical loading involve complex multiphysics interactions across multiple length and time scales, challenging predictive fatigue modelling. Experimental studies show that the thermo-mechanical response of electroplated metallisation at high heating rates deviates from conventional wafer-curvature predictions. Heating rates up to 10⁶ K/s lead to higher stress levels with heating exhibiting an approximately linear stress–temperature response, yet the stress-temperature response remains below purely thermo-elastic predictions [1]. The thermo-mechanical hysteresis indicates increasing dissipation at higher heating rates, reflecting rate-dependent inelastic mechanisms. Although the global stress state is compressive at elevated temperatures during heat pulses, intergranular creep pores form, suggesting localized tensile hotspots [2]. Microstructural analyses show increased kernel average misorientation and grain reorientation, reflecting intense localised plastic activity [3]. These findings highlight the importance of coupled thermo-mechanical effects across scales and the necessity of microstructure-resolved modelling. A finite element–based multiscale framework is developed to analyse cyclic thermo-mechanical degradation. Device-level simulations on a dedicated test structure, called Polyheater [4], are implemented in COMSOL Multiphysics, linked to in-situ experiments and coupled to microstructure-resolved sub-models reconstructed from EBSD data. The material response is described by an anisotropic, rate-dependent plasticity model including viscous effects, grain-boundary sliding as localised creep, and kinematic and local isotropic hardening correlated with dislocation density evolution. Intergranular fatigue is captured using a diffusion-based pore growth model coupled to the mechanical response using a damage field, with stress redistribution and localisation. Efficient cyclic lifetime prediction is achieved via a universal, non-damage-specific cycle-jump algorithm. Results demonstrate that a purely device-level perspective is insufficient. Microstructure-resolved simulations reveal localized tensile hotspots driving intergranular fatigue, demonstrating that coupled effects and microstructural heterogeneity govern damage evolution. The presented framework provides a predictive, multiscale tool for assessing fatigue in semiconductor metallisations under extreme thermal loadings and va