Self Modifying Finite Automata (SMFA) based state machine implementation for lower energy

Many finite state machines (FSMs) in embedded systems exhibits strong locality in state transitions. Traditional state assignment algorithms consider all FSM states´ outputs and next states to assign a code for each state. Self-modifying finite state machine(SM-FSM) takes the FSM strong locality characteristics to dynamically change the traditional finite state machine during runtime. This not only reduces the FSM search space, but also reduces the code space and allows code sharing among the dynamic behaviors. This paper proposes an architecture to support SMFA and evaluates the gain from SMFA on state machines with different characteristic.