A statistical static timing analysis incorporating process variations with spatial correlations

To evaluate effects of process variations on circuit delay accurately, this study proposes a Statistical Static Timing Analysis (SSTA) which incorporates process variations with spatial correlations. The algorithm applies a second order delay model that taking into account the non-Gaussian parameters - by inducting the notion of 'conditional variables', the 2D non-linear delay model is translated into 1D linear one; and by computing the tightness probability, mean, variance, second-order moment and sensitivity coefficients of the circuit arrival time, the sum and max operations of non-linear and non-Gaussian delay expressions are implemented. For the ISCAS89 benchmark circuits, as compared to Monte Carlo (MC) simulation, the average errors of 0.81%, -0.72%, 2.23% and -0.05%, in the mean, variance, 5% and 95% quantile points of the circuit delay are obtained respectively for the proposed method. The runtime of the proposed method is about 0.21% of the value of Monte Carlo simulation. The experimental results prove that the high accuracy of the SSTA is reliable.