Rapid Performance and Power Consumption Estimation Methods for Embedded System Design

Abstract

As embedded systems increase in complexity, rapid performance estimation methods, appropriate for use early in the design process, are becoming more and more necessary. These methods can produce significant decreases in execution time, power consumption and system cost. However, to be practicable, a design space exploration (DSE) process must be capable of evaluating several design alternatives quickly. This paper focuses on ways to accelerate performance and power consumption evaluation for embedded systems. Three methods: statistical simulation (SS), analytical modeling and detailed simulation (AMDS) and analytical modeling and statistical simulation (AMSS), offering both speed and accuracy for detailed cycle-accurate micro-architecture simulation, are presented and compared. Experimental results indicate that these methods produce interesting simulation acceleration factors. In addition, the error margin is on average less than 3.8%, reaching 8% in the worst case