Research

Research Highlights

[Professor Rhu Minsoo’s team wins the Best Paper Award’ at the International Symposium on High-Performance Computer Architecture (HPCA)]

[Professor Rhu Minsoo’s team wins the Best Paper Award’ at the International Symposium on High-Performance Computer Architecture (HPCA)]
 
IMG 0100
<(from left) Certificate of Award, Award Ceremony, PhD candidates Bongjoon Hyun (first author), Taehun Kim, and Dongjae Lee>
 
The research team led by Professor Rhu Minsoo from the school of EE announced that they have won the Best Paper Award at the IEEE International Symposium on High-Performance Computer Architecture (HPCA), one of the premier international conferences on computer architecture. This is the first time a research team from a domestic university has received the Best Paper Award at an international top-tier conference in the field of computer architecture. It is a prestigious honor given to a single top submission out of 410 papers submitted.
 
The research team led by Professor Rhu Minsoo, consisting of PhD candidates Bongjoon Hyun (first author), Taehun Kim, and Dongjae Lee from the Department of Electrical Engineering, won the Best Paper Award with their proposal of a simulation framework named uPIMulator. This framework is based on the commercialized Processing-In-Memory (PIM) technology of UPMEM.
 
Technologies that have been gaining significant attention recently such as Large Language Models like ChatGPT and recommendation systems require a high amount of memory bandwidth (the amount of data that can be moved in and out of memory at one time). Traditional CPU and GPU-based systems face limitations in meeting the increasing demand for memory bandwidth due to physical constraints.
To address the issue of limited memory bandwidth, Processing-In-Memory (PIM) technology, which integrates computing units within the memory itself, has started to gain attention. PIM technology is receiving recognition not only in academia but also in the industry, with the unveiling of PIM prototype products such as Samsung Electronics’ HBM-PIM and SK Hynix’s AiMX, as well as commercialization examples through UPMEM’s UPMEM-PIM product, demonstrating its potential.
 
However, the current state of Processing-In-Memory (PIM) technology is relatively in its early stages compared to the level of development in hardware architectures like CPUs or GPUs, and there is a need for extensive research on a wide range of hardware structures. Simulators that mimic hardware are frequently used in both academia and industry to explore various hardware design spaces, but research on simulators for PIM is comparatively lacking.

IMG 0101

Professor Rhu Minsoo’s research team explored various hardware structures that could improve the performance, robustness, and security of PIM through the development of a simulator based on the commercial PIM technology, UPMEM-PIM, undergoing design and verification processes. The significance of this research lies in the detailed analysis of PIM hardware structures and the exploration of various design directions through a simulator grounded in actual PIM products. The developed simulator is currently available as open-source (https://github.com/VIA-Research/uPIMulator), contributing to the research and development community.
 
This research was conducted with the support of the Korean government (Ministry of Science and ICT), the National Research Foundation of Korea, the Institute for Information & communications Technology Planning & Evaluation, and Samsung Electronics.
 

https://arxiv.org/abs/2308.00846

uPIMulator: https://github.com/VIA-Research/uPIMulator

 

​​​​