Yield-based system partitioning strategies for MCM and ASEM design

Abstract

We propose a new quantitative approach to MCM and ASEM system partitioning under yield, I/O, and area constraints. The objective is to minimize the off-chip communication/wiring cost while improving the yield of an MCM or ASEM (using smaller partitions without violating I/O and area constraints). We formulate this as a 0/spl minus/1 quadratic programming problem with linear constraints. This paper describes the quantitative model and results from detailed partitioning of large benchmark VLSI circuits.<>