TUESDAY June 26, 10:30am - 12:00pm | Room 3024
TOPIC AREA: DESIGN, DESIGN
KEYWORD: ARCHITECTURE & SYSTEM DESIGN
EVENT TYPE: SPECIAL SESSION

SESSION 15
Memory-centric Architectures: Industry Perspective from Embedded Systems to High Performance Computing
Chair:
John Kim - Korea Advanced Institute of Science and Technology, Daejeon, Republic of Korea
Organizer:
John Kim - Korea Advanced Institute of Science and Technology, Daejeon, Republic of Korea
For the past 50 years, systems architecture has mostly focused on feeding data to the processing units, including CPUs and GPUs, and the memory hierarchy have focused on allowing slow memory to work with fast processors. As Moore’s Law slows down, faster processors do not continue to provide improved system performance. Instead, memory-centric architectures flip the existing model by bring the processing to the data, thereby reducing latency, lowering system power, and improving throughput. These architectures require rethinking all aspects of existing memory systems, from physical structures to workloads. Memory-centric organization is not only applicable for high-performance computing but also embedded and mobile systems. In this special session, we will have three invited speakers from the industry to discuss the impact of memory-centric architectures on different systems; talks will cover mobile workloads and their impact on mobile systems, near-data processing and accelerators, as well as high-performance computing and energy-efficient DRAM.

15.1Memory Characterization for Mobile Workloads
 Speaker: Daehyun Kim - Samsung Electronics America, Inc., Seoul, Republic of Korea
 Author: Daehyun Kim - Samsung Electronics America, Inc., Seoul, Republic of Korea
15.2Memory-centric Accelerators in High-performance Systems
 Speaker: Nuwan Jayasena - Advanced Micro Devices, Inc., Santa Clara, CA
 Author: Nuwan Jayasena - Advanced Micro Devices, Inc., Santa Clara, CA
15.3Energy-Efficient DRAM for High-bandwidth Applications
 Speaker: Mike O'Connor - NVIDIA Corp., Austin, TX
 Author: Mike O'Connor - NVIDIA Corp., Austin, TX