Circuit yield optimization by analyzing performance statistics

A new method for yield optimization is presented that considers the high computational costs for circuit simulations and the typical parameter partitioning of integrated device models in deterministic and statistical parameters. It is based on a systematic concept for performance function macromodeling, that allows a flexible trade-off between simulation costs and modeling accuracy. Using quadratic performance models, an analytical expression for the yield gradient is derived by analyzing the performance statistics. Based on this yield gradient, an efficient deterministic optimization procedure will be proposed, which maintains the individual view of performance properties.