A new faster algorithm for iterative placement improvement

We present a new faster design-style independent iterative placement improvement algorithm. Randomized simulated annealing based algorithms produce good result. But on very large designs, the inherently long run-time makes it prohibitive to use randomized algorithm. On the other hand deterministic improvement methods do not produce as good a result as the simulated annealing based algorithms. Moreover, none of the existing placement improvement techniques addresses the non row-based design style. In this paper, we combine the advantages of both the random and deterministic approach to develop a new faster placement improvement algorithm. Experimental results show that our algorithm performs much better than existing placement improvement algorithm. On some benchmarks, our algorithm is as much as 8× faster than that on Domino with a significant reduction in total net length