False Path Aware Timing Yield Estimation under Variability

Effects of fluctuations in circuit timing due to process and environmental variations are becoming increasingly important as we move into sub-45 nm technology. Since the delay of each gate is dependent on its input vectors, the timing yield, the probability that the circuit meets the given timing constraint, varies with different primary input patterns. Traditional timing yield estimation approaches assumed worst case delay models for each gate over all its input vectors, which results in much pessimism. To overcome the aforementioned problems, this paper proposes a Monte Carlo based approach which can obtain a much tighter lower bound on the circuit timing yield compared to the existing timing yield estimation techniques. Specifically, our approach builds multiple input-vector-dependent variation-aware delay models for each logic gate, and considers the impact of false paths, both static and dynamic false paths, which are carefully selected from the likely timing-critical paths under variability. We demonstrate gradual improvement in the estimated timing yield in the simulation results, and show that the timing yield computed using traditional worst-case delay models is highly pessimistic.