Statistical Analysis and Optimization Methodology for VLSI Circuits


Kelvin Le


Statistical analysis and optimization methodology, specifically statistical static timing analysis, was a hot topic and drew a lot of attention during 2003 and 2007. Many research papers were published and multiple solutions were developed from various EDA vendors, including several startup companies. However, none of these solutions were widely adopted in the market due to their intrinsic complication, difficulty in adoptions, and lack of support from the entire ecosystem. Methods such as AOCV and POCV were adopted as low cost alternatives to address various needs of variability modeling.

In recent years, there are renewed interests in statistical methodology. On one hand, the pursuit for low power design for mobile/IoT drives the circuit to operate at near threshold, which leads to exponential increasing in cell timing variability. On the other hand, continued downward scaling of transistors leads to increased impact of wire/via variability to timing, especially for high performance operations. This talk reviews various technologies developed for statistical static timing analysis, including statistical cell characterization, statistical parasitic extraction, parametric delay calculation and timing propagation, full-chip sensitivity analysis and variation-driven ECO. Given the renewed interests in statistical analysis, we believe these technologies will be useful to serve the needs of modern VLSI circuit design


Kelvin Le is a software engineer at Google. He has been in Google since Nov 2019. Before that, he worked at Synopsys since 2011, where he was one of the key developers for PrimeTime and PrimeYield. Before joining Synopsys, he worked at Extreme-DA since 2004 as one of the founding members. Extreme-DA was acquired by Synopsys in 2011. Kelvin Le received his Ph.D. degree in Electrical and Computer Engineering from Carnegie Mellon University in 2006, andM.S. and B.S. degrees in Electrical Engineering from Shanghai Jiaotong University, in 2001 and 1999, respectively.

Kelvin Le's research interests include electronic design automation, machine learning, and distributed systems. He is the inventor of Parametric On-chip Variation (POCV), and one of the key contributors to Liberty Variation Format (LVF). Kelvin Le has 15+ publications and holds 10+ US patents. He is also one of the authors of the book Statistical Performance Modeling and Optimization.