Model order reduction techniques are developed and utilized to make numerical simulations more efficient. The use of Reduce Order Models (ROM) also enables data exchange with external parties without disclosing the sensitive information present in a Full-Order Model (FOM). It is crucial to optimize for both the efficiency and accuracy of a ROM to keep a minimal deviation from the FOM. The complexity of a ROM-based simulation depends on the definition of the ROM as well as its connection with the remaining FOM. This paper investigates the effect of different ROM-FOM interface definitions for a test case consisting of an electronic package-on-PCB assembly. A virtual Design of Experiments (DoE) was carried out with a total of 41 cases considering three different locations and up to four different constraint equations for the ROM-FOM interface. The effect on the accuracy and time-efficiency of the ROM-based thermomechanical simulations are compared with respect to the full Finite Element (FE) model. The deformable configuration for the interface generally showed the most accurate results, while the rigid configuration was the most efficient across the board. The beam configuration did not always follow an expected trend based on the order of elasticity values of the assigned materials. Based on the deformation results and the time associated with ROM generation and use-pass, multiple optimal solutions from the DoE are discussed.

The objective of this work is to develop a microstructure-based simulation approach to assess the fatigue life of solder joints that are used by the microelectronics industry. The developed approach can generate solder joints with random grain morphologies by means of 3D Voronoi tessellation. The anisotropic material behavior of each grain is described by the Garofalo creep equation combined with Hill's definition of the equivalent stress for anisotropic materials. Grain boundaries are implemented as interface elements, with an isotropic creep constitutive model. The stochastic variability in the creep response of solder joints was qualitatively estimated by generating 100 unique solder joints containing 5 to 9 grains, each having a random material orientation. These joints were independently loaded with a realistic stress level for microelectronic products during thermal cycling. The volume-averaged creep strain energy density in the solder joints was used to predict the fatigue life of the solder joints. The results showed a factor of ~4 difference in expected lifetime of the individual solder joints. Next, nine randomly picked solder joints from the above-mentioned pool of 100 were sandwiched between a silicon die and a printed circuit board to form a simulation model of a Wafer-Level Chip-Scale package (WLCSP). The creep strain energy density in the joints was computed for 34 unique cases of the WLCSP. A factor of ~2.5 between the highest and lowest estimate for the solder joint life was found. The slope of the corresponding Weibull distribution equals ~6, which falls within the slopes typical reported for solder joint reliability of WLCSPs.