Abstract Overmolded Flip Chip Quad Flat No-lead (FCQFN) is a low cost flip chip on leadframe package where there is no need for underfill, and is compatible with Pb free or high Pb metallurgy. A robust leadframe design, quality solder joint formation and an excellent molding process are three factors needed to assemble a high performance FCQFN. It combines the best of both wirebonded QFN and wafer chip scale devices. For example, wafer chip scale has low resistance, but inadequate thermal performance (due to absence of thermal pad), whereas wirebonded QFN has good thermal performance (i.e., heat dissipated through conductive die attach material, through the pad and to the board) but higher resistance. Flip chip QFN combines both positive aspects – that is: low resistance and good thermals. One of the common defects for molded packages across the semiconductor industry is the occurrence of mold voiding as this can potentially affect the performance of a device. This paper will discuss how mold voiding is mitigated by understanding the mold compound behavior on flip chip QFN packages. Taking for example the turbulent mold flow observed on flip chip QFN causing mold voids. Mold compound material itself has a great contribution to mold voids, hence defining the correct attributes of the mold compound is critical. Altering the mold compound property to decrease the mold compound rheology is a key factor. This dynamic interaction between mold compound and flip chip QFN package configuration is the basis for a series of design of experiments using a full factorial matrix. Key investigation points are establishing balance in mold compound chemistry allowing flow between bump pitch, as well as the mold compound rheology, where gelation time has to be properly computed to allow flow across the leadframe. Understanding the flow-ability of mold compound for FCQFN, the speed of flow was optimized to check on its impact on mold voids. Mold airflow optimization is also needed to help fill in tighter bump spacing but vacuum-on time needs to be optimized as well.
Read full abstract