Epoxy molding compound (EMC) is a thermosetting polymer filled with inorganic fillers such as fused silica. EMC has been used extensively as a protection layer in various semiconductor packages. The warpage and the residual stress of packages are directly related to the thermomechanical properties of EMC. As the size of semiconductor packages continues to shrink, prediction of the warpage and residual stress becomes increasingly important. The viscoelastic properties of EMC are the most critical input data required for accurate prediction. In spite of the considerable effort devoted to warpage prediction, accurate prediction of warpage remains a challenging task. One of the critical reasons is the inappropriate assumption about the bulk modulus – time and temperature “independent” bulk modulus, which is not valid at high temperatures. In this thesis, a novel experimental method, based on an embedded fiber Bragg grating (FBG) sensor, is developed and implemented to measure a complete set of linear viscoelastic properties of EMC just from a single configuration. A single cylindrical EMC specimen is fabricated, and it is subjected to constant uniaxial compression and hydrostatic pressure at various temperatures. Two major developments to accommodate the unique requirements of EMC testing include: (1) a large mold pressure for specimen fabrication; and (2) a high gas pressure for hydrostatic testing while minimizing a temperature rise. The FBG embedded in the specimen records strain histories as a function of time. Two linear viscoelastic properties, Young’s modulus and Poisson’s ratio, are first determined from the strain histories by the analytical relationship between the BW change and the properties. A unique sequential iterative procedure is proposed to completely negate the effect of measurement uncertainties during complex non-linear regression. The other two elastic properties, Shear modulus and Bulk modulus, are also calculated from the relationship among the linear elastic constants. The master curves are produced, and the corresponding shift factors are determined.
Validity of three major assumptions associated with the linear viscoelasticity - thermorheological simplicity, Boltzmann superposition and linearity - are verified by supplementary experiments. The effect of the time-dependent bulk modulus on the warpage is also discussed