A novel net-degree distribution model and its application to floorplanning benchmark generation

This paper characterizes the behaviour of multi-terminal nets in a circuit to provide an accurate estimation of interconnect net-degree distribution. Based on Rent's rule, we derived a new net-degree distribution model, called the weighted exponential model. Numerical evaluations show that weighted exponential model is more accurate than any other currently available net-degree distribution models. It generates from 18% to 87% smaller average absolute errors in the prediction of the total number of nets and average net degree. Moreover, it was observed that the internal fraction factor f, that represents the ratio of the number of new internal nets to the total number of new nets, similarly to Rent's rule, exhibits Regions II and III. With a refined value of f, our model can provide even more accurate estimation of net-degree distribution. A method to generate floorplanning benchmarks using weighted exponential model is presented. Thanks to the accuracy of new model and the introduction of a block-degree distribution, the significantly improved benchmark generator, BGen, outperforms a prototype tool we have previously developed using Zarkesh-Ha et al's model.