Abstract
Corner-based Timing Analysis (CTA) becomes more and more pessimistic along with the shrinking feature size. This trend has urged the need of Statistical Static Timing Analysis (SSTA). However, this new generation of timing analysis has not yet widely adopted in the industry due to various weaknesses. The path-based SSTA framework proposed in this thesis computes path delay distributions by propagating iteratively mean and variance of cell delay with the help of conditional moments. These moments, condi-tioned on input slope and output load, are stored in a statistical timing library. This framework performs as fast as parametric methods while not losing too much accuracy compared to Monte Carlo simulations, which meets the objective of the research. Another contribution of this thesis is the improvement of the techniques to do timing characterization. We use input signals based on log-logistic distributions and inverters as output load to capture slope and load variations. In addition, the runtime of characterization could be greatly saved by the reducing dimension technique, which would be validated in the near future. In the part of applications, our SSTA engine shows significant delay gains with respect to CTA. The discrepancy of critical paths orderings obtained respectively by SSTA and CTA is explained as well. Finally, a study of cell-to-cell correlation is given.