A Low Power-consuming Embedded System Design by Reducing Memory Access Frequencies with Multiple Reference Tables and Encoding the Most Executed Instructions

Abstract

In designing an embedded system, system performance and power consumption have to be taken carefully into consideration. In this paper, we use the locality of running programs to reduce the great number of memory access to save the power and maximize the performance in the embedded system design. We encode the frequently executed instructions as shorter code words and then pack continuous code words into a pseudo instruction. Once the decompression engine fetches one pseudo instruction, it can extract multiple instructions. Therefore, a number of memory access times can be efficiently reduced because of space locality. However, if the program size of applications increases, the number of the most frequently executed instructions grows up. This situation results in the degradation of system performance and power consumption. To solve this problem, we also propose a method of using multiple look-ahead tables to make most frequently executed instructions have shorter encoded code words to improve the performance and power. From our simulation results, our method with one reference table which contains the most frequently executed 512 instructions reduces the number of times memory is accessed by about 60% Moreover, when two reference tables that contain 256 instructions in each table are used, the memory access ratio is 16.73% less than the ratio resulting from one 512-instruction reference table. According to the simulation results, our proposed methods based on the frequencies of executed instructions result in low power consumption and performance improvement.