Factorizing FSM's with modify and restore method

Abstract

Implementation of large finite-state machines (FSMs) as smaller interacting machines, by factorizing them and interconnecting the factored and factoring FSM's in such a way so as to maintain the functionality of the original machine usually leads to an improvement in the performance (that is, reduction in delay) of the original machine. Exact factors, if present in an FSM, can result in the most effective way of factorization. However, it has been found that most of the FSMs are not exact factorizable. In this paper, we have presented a scheme called the modify and restore (MAR) method which attempts to make FSMs exact factorizable even if the original FSM is not directly exact factorizable. This is done by making minor changes in the next state space of the original FSM while maintaining the functionality of the FSM by a restoring logic. We have tested the effectiveness of our method of factorization followed by state assignment for both two-level and multilevel implementations. Experimental results on the MCNC benchmark examples have shown significant improvement in delays of the final realization.