%0 Journal Article
%A CHEN Lan
%A HAO Xiao-ran
%A LIU Chen-ji
%A NI Mao
%A SUN Hao
%T A Hybrid Memory System for Edge Computing
%D 
%R 10.13190/j.jbupt.2019-091
%J Journal of Beijing University of Posts and Telecommunications
%P 103-109
%V 43
%N 2
%X In order to satisfy the low power requirements of IoT devices, a low power hybrid main memory system composed of dynamic random access memory and phase change memory is proposed. The hybrid main memory is managed by the hybrid memory control modules which are added to the memory controller. An improved dual queue algorithm is proposed to filter out memory pages with more write requests from phase change memory. Because of the poor write performance of phase change memory, the page migration module migrates the selected pages from phase change memory to dynamic random access memory. And then the updated page address is recorded in the address mapping table for subsequent memory access. Experimental results show that compared with the traditional main memory composed entirely of dynamic random access memory, the hybrid main memory reduces the power-delay product by 43.9% on average.
%U https://journal.bupt.edu.cn/EN/10.13190/j.jbupt.2019-091