State assignment for low-power FSM synthesis using genetic local search

The power consumption of a finite state machine can be reduced if fewer internal nodes change states during operation. We assign the symbolic states representation so that the most probable state machine transitions change fewer state bits. We calculate the state transition probabilities from the given input signal probabilities and the Markov chain state probabilities (estimated from probability simulation). We reduce the Hamming distances between the binary representations of frequently transitioning pairs of states. Using a genetic local search algorithm, we can quickly find near optimal state assignments for finite state machines with less than fifty states and good assignments for larger benchmarks. The average power consumption is estimated using a sequential transition density simulator on gate-mapped representation of the design. The gate´s switching density is measured for each edge in the state transition graph and then scaled using the state transition density. Results using the MCNC benchmarks show that low-power state assignment produces encodings that consume about 20% less power than current “area optimizing” state encoding programs. When these encodings are synthesized into multilevel gate representations using SIS, we find that the area is also reduced