Reliability analysis for heterogeneous millimeter-wave RF front-end package using a hybrid FEA-machine learning model

Purpose The purpose of this study is to implement a hybrid finite element analysis (FEA) and machine learning framework. A deep neural network (DNN) surrogate model was trained on parametric FEA data for rapid multiobjective optimization. The Anand viscoplastic model captured SAC305 solder behavior, while the Darveaux fatigue model predicted joint lifetime using a submodeling approach. Experimental validation included thermal warpage measurements (25–150 °C) and microscopic analysis of fabricated prototypes using X-ray and scanning electron microscopy to correlate simulation predictions with actual manufacturing quality. Design/methodology/approach To address thermo-mechanical reliability challenges in heterogeneous millimeter-wave radio frequency (RF) front-end packages for 5G/6G systems, this study develops an efficient co-design methodology. The work aims to predict SAC305 solder joint fatigue life while optimizing geometric parameters (chip thickness, high-density interconnect (HDI) thickness and solder spacing) to balance thermal performance and mechanical stress, bridging the gap between theoretical modeling and manufacturing realities in multimaterial high-density interconnect substrates. Findings The DNN achieved high accuracy (temperature MAE 1.77 K, displacement MAE 0.0001 mm). Optimization yielded chip thickness 0.07 mm, HDI thickness 0.7 mm and solder spacing 0.6 mm. Fatigue analysis predicted 476 cycles for critical corner joints. Warpage measurements confirmed simulation trends (21.2 µm concave to 34.3 µm convex). However, manufacturing defects – including via misalignment and incomplete filling – were identified, suggesting actual reliability may be lower than predicted and highlighting critical process control requirements. Originality/value Proposing a validated hybrid FEA-ML framework for heterogeneous RF packaging, correlating fatigue predictions with manufacturing defects to provide process control guidelines for solder joint reliability.