Process-integrated in-situ characterization and multiscale modeling of cure-dependent thermo-mechanical behavior of epoxy molding compounds for accurate warpage prediction in advanced semiconductor packaging

Accurate prediction of cure-induced thermo-mechanical deformation in polymer-based composites critically depends on reliable characterization of material property evolution under process-relevant curing conditions. For warpage prediction in advanced semiconductor packages, a liquid-type, highly silica filled epoxy molding compound (EMC) was characterized under molding-relevant thermal and pressure histories. A process-integrated experimental–computational framework is presented that couples in-situ dielectrometry and embedded fiber Bragg grating (FBG) sensing during compression molding with molecular-dynamics (MD)-based multiscale modeling to obtain a self-consistent set of cure- and temperature-dependent thermomechanical properties. Gelation under the applied thermal/pressure history was identified by in-situ dielectrometry. Embedded FBG sensing quantified the post-gel effective cure shrinkage referenced to gelation point as ∼0.102%. A dual-configuration FBG strategy using two concentric cylindrical specimens under ∼1 MPa enabled inverse reconstruction of post-gel Young's modulus evolution, reaching ∼768 MPa at the end of molding dwell. Poisson's ratio evolution of the epoxy network was obtained from stepwise crosslinked MD simulations and upscaled to the EMC composite level using Mori–Tanaka homogenization, separating cure-dependent ν ( α ) during curing from temperature-dependent ν ( T ) during cooling. In addition, the MD-based multiscale predictions show agreement with experimentally quantified effective cure shrinkage and the reconstructed post-gel modulus evolution under the same molding history. Viscoelastic relaxation during cool-down was represented using DMA-derived master curves and time–temperature superposition. The property set was validated through an EMC/Al bi-strip warpage experiment and process FE simulation (3.326 mm predicted vs. 3.393 mm measured; ∼1.97% deviation), enabling quantitative assessment of key contributors to warpage.