To KGD or not to KGD? That is the question!

Abstract

When planning and designing MCM products or products that contain MCMs, designers are faced with numerous questions including Should I use Known Good Die (KGD)? and If I want to use KGD, what price premium is reasonable to pay? These questions arise because in most cases, but especially for complex MCMs, tradeoffs exist between paying a premium for KGD to increase first pass module yield and performing rework on modules without KGD to raise final module yields to acceptable levels-if possible. Further complicating the issue is the fact that the price premium for KGD is decreasing over time. As a result, an economic model is required that compares the module level cost implications of not using KGD and reworking the module vs. using KGD at a variety of price premiums. In this paper, an economic analysis for building an existing product, the Cardio 486 manufactured by Seiko-Epson will be presented. This is a 486 computer system with 2 M DRAMs in a PCMCIA sized enclosure. The size constraints dictate that MCM technology be used. The paper will begin by establishing a baseline cost estimate for the existing Cardio 486 with KGD and/or packaged, tested ICs. Next, this model will be extrapolated to another product which is hypothetically in the planning stages; a Pentium-based PCMCIA sized computer with 64 M DRAMs. For this model, cost estimates for building the products with non-KGD using rework will be compared to building the product with KGD. The models will be presented with a range of prices for both KGD and non-KGD to determine break-even points for the various approaches.