Disjoint Region Partitioning for Probabilistic Switching Activity Estimation at Register Transfer Level

Abstract

This paper presents a partition method for probabilistic switching activity estimation of combinational circuits described at register transfer level (RTL). Probabilistic estimation of switching activity requires large and complex models that could be unfeasible for large circuits; therefore, circuit partitioning becomes a necessary step to address the analysis. Nevertheless, partition methods imply approximations that produce inaccurate results. We present a partition method based on disjoint signals that minimizes the error and, in addition, it is easy to carry out. Results show important reductions on the binary decision diagrams (BDD) of the probabilistic model as well as low errors. Furthermore, the BDD reduction ratio shows a tendency to increase with large circuits; whilst error seems to decrease with the circuit size.