2.5D and 3D applications using through silicon vias (TSVs) are increasingly being considered as a packaging alternative. Miniaturization and high performance product requirements are driving this move – even though in many cases the cost of both 2.5D and 3D is still high.
The primary applications for 2.5D interposers with TSVs are GPUs/CPUs, high-end ASICs, and FPGAs. Adoption for FPGAs has already started. The key to the performance gains in recently introduced FPGAs is the partitioning of an FPGA die into four “slices” that are mounted on a silicon interposer or what Xilinx calls its Stacked Silicon Interconnect technology. Applications for interposers include tablets, gaming, and high-end computing and network systems. The drivers are mainly partitioning large die, integrating single chips into a module, reducing die size where substrate density is the constraint, and using the interposer to minimize the stress on large die that are fabricated with extra-low-k (ELK) dielectrics.
The primary applications for 3D solutions are stacked memory cubes and memory plus logic. The true 3D nature of stacking all active silicon allows better miniaturization, but yield issues can quickly drive the cost unacceptably high.
This analysis examines the cost drivers for 2.5D and 3D applications. Activity based cost models will be used to analyze the complete cost of fabricating and assembling active die on a silicon interposer and active die stacking on other active die. Total product cost impact - not just the cost of a specific activity - is the focus of this analysis. Since yields play a major role in cost, a sensitivity analysis of the different yields including die yield before wafer probe, die yield after wafer probe, TSV yield, interposer yield, assembly yield, substrate yield, etc. will be presented. The critical yield points in the manufacturing flow and dominant activity cost drivers (equipment, material, and /or labor) will be presented as well as suggested minimum thresholds for 2.5D and 3D technology to be a cost effective technology.
A Bit String Content Aware Chunking Strategy for Reduced CPU Energy on Cloud Storage
