Characterization of Non-Conductive Adhesives

This study aims to establish a global-local modeling methodology for determining the residual contact stress developed during fabrication of flip-chip-on-flex (FCOF) microelectronics systems. The assembly consists of a silicon die with gold bumps bonded with a non-conductive adhesive (NCA) on to gold-plated copper bumps on a flex substrate. Manufacturing variabilities cause a nonuniformity in the bump heights, leading to some bumps that are “tall” and some that are “short.” The fabrication process needs to achieve a significant amount of compressive initial contact stress in all the bumps, to achieve an acceptable level of electrical contact resistance. Furthermore, this stress level forms the initial condition for cyclic relaxation of the stress (and corresponding progressive loss of contact resistance) due to temperature cycling throughout the life cycle of the assembly. A key issue to be investigated is the nonuniformity of the contact stresses due to the variabilities in the height of the metal contact bumps. The method is demonstrated for a selected NCA. The fabrication process consists of mechanical compression to bring all the bumps into contact, thermal curing of the adhesive during which it undergoes chemical shrinkage, removal of the mechanical compressive force and cool-down to room temperature. The modeling complexities include the geometric complexity, as well as nonlinearities due to elastic-plastic properties and large deformations of the metal bumps, evolution of contact surfaces between the two bumps, and nonlinear thermomechanical properties of the adhesive as it cures. Modeling strategies used to capture the nonlinearities include “contact elements” to prevent interpenetration at the contact surfaces, elastic-plastic models to account for metal plasticity, “element birth and death” to account for the solidification of the polymer NCA. The entire bonding process is modeled with a global-local model to reduce the computational complexity. The results of the global model serve as the input for the local model. Key findings include: the accuracy of the simulation is very sensitive to the accuracy of the gold and flex constitutive models used; the inclusion of viscoelastic properties for the epoxy has a significant effect on simulations; and better stress development comes from a higher concentration of short bumps than tall bumps.