Flexible, ultra-low power sensor nodes through configurable finite state machines

Abstract

Due to the recent popularity of context-sensitive applications, there is a growing need for reliable, long-lifetime ubiquitous sensor nodes. The severe energy-efficiency requirements of these energy-scarce devices require complementing traditional circuit-level energy saving techniques, with architecture-level methods. Traditional approaches such as exploiting parallelism have however limited impact in sensor node processors, due to their control-dominated and event-based, irregular data processing workload patterns. Executing event-based tasks in specialized finite state machines relieves the on-board microcontroller, however, at the penalty of reduced post-manufacturing configurability. An architecture proposal for configurable finite state machines assisting sensor node processors is presented, which allows saving energy through task off-load while maintaining system flexibility. Simulations demonstrate 46% energy savings when compared to a sensor node that executes tasks in a microcontroller. This gain comes at relatively minor area overhead.