Presentation
Seeking the Yield Barrier: High-Dimensional SRAM Evaluation Through Optimal Manifold
DescriptionEfficient yield estimation is a central issue in modern submicrometer circuits.
In this work, we propose a novel sampling method, named Optimal Manifold Important Sampling (OPTIMIS), which extends the classic norm minimization from optimal vectors to optimal manifolds. We implement a normalization flow, which uses a parallel updating scheme to update itself to approximate the ground truth failure probability. Also, for the first time, we show the close connection between the surrogate-based and importance sampling-based yield estimation under our framework. Experiments of several SRAM problems show the superiority of our method with up to 2.5x than the SOTA methods.
In this work, we propose a novel sampling method, named Optimal Manifold Important Sampling (OPTIMIS), which extends the classic norm minimization from optimal vectors to optimal manifolds. We implement a normalization flow, which uses a parallel updating scheme to update itself to approximate the ground truth failure probability. Also, for the first time, we show the close connection between the surrogate-based and importance sampling-based yield estimation under our framework. Experiments of several SRAM problems show the superiority of our method with up to 2.5x than the SOTA methods.
Event Type
Research Manuscript
TimeThursday, July 13th11:25am - 11:40am PDT
Location3010, 3rd Floor
EDA
Design for Test and Silicon Lifecycle Management