FSM re-engineering and its application in low power state encoding

Abstract

We propose Finite State Machine (FSM) re-engineering, a performance enhancement framework for FSM synthesis and optimization procedure. We start with any traditional FSM synthesis and optimization procedure; then reconstruct a functionally equivalent but topologically different FSM based on the optimization objective ; and conclude with another round of FSM synthesis and optimization (can be the same procedure) on the newly constructed FSM. This allows us to explore a larger solution space that includes synthesis solutions to the functionally equivalent FSMs instead of only the original FSM, making it possible to obtain solutions better than the optimal ones for the original FSM. Guided by the result of the first round FSM synthesis, the solution space exploration process can be rapid and cost-efficient. To demonstrate this framework, we develop a genetic algorithm and a fast heuristic to re-engineer a low power state encoding procedure POW3 [1]. On average, POW3 can reduce the switching activity by 12% over non-power-driven state encoding schemes on the MCNC FSM benchmarks. We then re-engineer these benchmarks by the proposed genetic algorithm and heuristic respectively. When we apply POW3 to the re-engineered FSMs, we observe an additional 8.9% and 6.0% switching activity reduction. This translates to an average of 7.9% energy reduction with little area increase. Finally, we obtain the optimal low power coding for benchmarks of small size from an integer linear programming formulation. We find that the POW3-encoded original FSMs are 27.0% worse than the optimal, but this number drops to 6.7% when we apply POW3 to the re-engineered FSMs.

FSM Encoding for BDD Representations

“…So by proposing to add heuristically some redundant states to a CSFSM to make it a power-of-two, we are not in a position too different from the traditional case where we encode a CS-FSM without knowing what state transition graph (maybe with redundancies) would be the best starting point, and therefore we start arbitrarily, say, with the state-minimized one. A recent work (Yuan, 2005) proposes 'FSM reengineering' as a technique to re-construct an FSM that is functionally equivalent to a given one to enable optimization tools to find better solutions to synthesis problems, and applies it to state encoding for low power.…”

A Technology Mapping of FSMs Based on a Graph of Excitations and Outputs

“…Internal state assignment is a vital stage of FSM synthesis, and is usually directly associated with the result that we expect to obtain. The goal of optimization may be the chip area required for the FSM [4], [5], the speed of operation of the automaton [5], [6], or minimization of the power consumption [7]- [9]. Issues concerning the minimization of power consumption have recently become particularly important.…”

Discrete-State Encoding in Hybrid-State Systems for Intelligent Vehicle Control and Estimation

“…Different code assignments such as Gray code [12] or Johnson encoding [13] try to minimize the number of bits changing during state transitions, whereas performancebased coding schemes, such as one-hot coding, aim to maximize the speed of the implementation on field-programmable gate arrays by assigning as many bits to the code as there are discrete states. While these pioneering studies and more recent counterparts such as [14] and [15] dealt mainly with FSMs in isolation, the hybrid-state hierarchy and the DSS-CSS interaction is the focus of this paper.…”