Accurate estimation of wirelength and congestion is one of the fundamental issues in VLSI physical design. Current probabilistic estimation methods fail to produce accurate results since they ignore congestion-related detouring and effects of the number of vias and bends. In this work, we propose a practical stochastic routing probability distribution model which includes the effects of blockage and the number of bends. The new model is tested by comparing the estimated routing probability distribution with the actual routing results of a commercial detailed router. An iterative congestion map construction algorithm based on the new probabilistic model is proposed for accurate wire-length and congestion map estimation. The results show that our proposed methods can improve the total wirelength estimation accuracy (i.e., reduce estimation error) by 90% on average with respect to the traditional RSMT wirelength estimate. Our methods also produce more accurate congestion maps than the previous congestion estimation method of without significant runtime overhead.
展开▼
