A Study of Performance and Power Consumption on Multicore Embedded Systems

• Main Authors: Er-Hsi Tsao, 曹爾習
• Other Authors: Chang-Hong Lin
• Format: Others
• Language: zh-TW
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/68hy7s

碩士 === 國立臺灣科技大學 === 電子工程系 === 106 === As the application of ARM based embedded system has become more common in our daily life, video and multimedia applications have also become necessary requirements for embedded systems, such as video streaming, surveillance, automobile systems, etc. Besides, embedded systems often require small form factor, light weight, low power, and multi-function features that are sometimes conflicting in nature. This thesis discusses how ARM based embedded system can utilize the multi-core architecture, with the new NEON SIMD technology to maximize embedded system’s function and performance, as to improve the multimedia user’s visual experience. However, ARM CPU’s arithmetic logic unit (ALU) and register are mainly designed for 32 bit or 64 bit algorithm. If not used in conjunction with NEON architecture, the CPU will expend a lot of time on processing 8 bit and 16 bit multimedia’s video pixel data and respective color depth application, which will cause the device unable to achieve balance between performance and power consumption, and thus have adverse effect on the entire embedded system’s objective of multi-function and low power consumption. In this test experiment, the embedded system is based on ARM Dual Cortex-A9 CPU with NEON , and the OS is Android 4.4 Kitkat. The JTAG debugging interface is used to control the operation of multi-core on and off. As the testing results of the implementation and verification, we can see that when the dual-core is on, whatever the benchmark test scores or playing high resolution videos, the system only consume about 50% of power, but can enhance more performance of about 30% to 80%. The multi-core architecture can significantly make up for the single core CPU in the processing of complex applications, and increase the embedded system functions more efficiency.