In this work, an investigation has been performed on hygrothermally induced die stresses in flip chip assemblies caused by moisture absorption by the underfill encapsulant. Silicon test chips were first applied to perform a variety of measurements of moisture and thermally induced die stresses in flip chip on laminate assemblies. The sample die stresses were first measured after underfill encapsulation and cure, and then subsequently after long term storage (10 years) at room temperature and ambient humidity. The assemblies were then exposed to and 85 °C and 85% RH high humidity harsh environment for various durations, and the die stresses were evaluated as a function of the exposure time. Finally, reversibility tests were conducted to see whether the effects of moisture uptake were permanent. After long term storage, the experimental measurements showed that the normal stresses in the flip chip die relaxed significantly, while the shear stresses exhibited only small variations. In addition, the 85/85 hygrothermal exposure had strong effects, generating tensile die normal stress changes of up to 30 MPa in the flip chip assemblies. Thus, the initial compressive die normal stresses due to flip chip assembly were found to relax significantly during the moisture exposure. Upon fully redrying, it was observed that the moisture-induced stress changes were fully recovered.
The results of the experimental measurements were subsequently correlated with predictions from finite element numerical simulations. When performing moisture diffusion modeling, the conventional method is to use a thermal analogy based on the similarity of governing equations of heat transfer and moisture diffusion. However, this method has some drawbacks including giving incorrect results when dealing with time- and temperature-dependent problems or discontinuities in the moisture concentrations at material boundaries. In this study, we have used a new feature in ANSYS v14 to perform coupled multi-physics simulations of the moisture diffusion process without the aforementioned limitations. The simulation results were found to show strong correlations with experimental measurements.
