Min Norm Failure Vector Guided Yield Optimization Method for Nanometer SRAM Design

摘要

High-sigma analysis is important for estimating the probability of rare events. SRAM usually require extremely low rate. Further more, nanometer SRAM design is challenging due to ever increasing process variation, this is especially true for near-threshold voltage design. That means basic yield analysis is not enough, we also need a yield optimization method. In this paper, a new systematic methodology is proposed to tackle the above issues. Importance sampling method with an efficient online surrogate model is used for yield analysis. MNFV (min-norm failure vector) is used to optimize the yield by increasing the overall distance from the failure boundary. Experiments show that with commercial 65nm technology our yield driven design can work under 0.6 vdd with 100% yield.