A fast approximation technique for power grid analysis

In this paper, we present a fast approximation algorithm for computing IR drops in a VLSI power grid. Assuming that the grid does not have pathological defects, the algorithm can estimate IR drops to within 5% average error, with a run time of less than one second per million nodes. Incremental recomputations with new current source values are even faster. The IR drop profiles have excellent correlation with simulated values, making this approach a viable platform for building automatic grid optimization algorithms.