Abstract

As a critical part of a stored-program digital computer, the main memory system supports instructions and data to be consumed and produced by processing cores. High storage density and low manufacturing cost are the main design goals of the dynamic random access memory (DRAM) technology, making DRAM a primary option for main memory storage on contemporary computer systems that demand ever-increasing storage capacity. However, this focus on storage density and low cost leads to slow access time, low bandwidth, and high performance sensitivity on access patterns compared to the static random access memory (SRAM) technology, which is mainly used for on-chip storage. Thus, both the performance and energy efficiency of the main memory system are critical in improving the energy efficiency of the entire computer system. Also, emerging technologies such as 3D through-silicon vias and phase change memory have started to impact the structures and roles of main memory systems. This chapter categorizes and reviews recent research efforts to improve the performance or energy efficiency of contemporary memory systems. First, we describe conventional memory access scheduling policies which generate DRAM commands based on pending memory requests; we then review more advanced techniques that are focused on either improving performance or effectively managing DRAM power. Research works exploiting emerging technologies are introduced, and their impacts on future memory systems are analyzed. Then, we cover the proposals to modify memory modules or memory device architectures that reflect the access characteristics of future manycore systems.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call