A research team headed by Professor Jungrae Kim of the Department of Semiconductor Systems Engineering has won the Best Paper Award at the IEEE/ACM International Symposium on Computer Architecture (ISCA) 2026.
ISCA is the world’s most prestigious conference in computer architecture and AI semiconductor systems. With a history spanning more than half a century, the conference is often described as the “Olympics of AI semiconductor architecture,” and even having a paper accepted is considered a major academic achievement.
The award marks the first time in ISCA’s history that a Korean university has received the Best Paper Award. No Korean institution had previously received the award or even had a paper selected as a Best Paper candidate. This achievement further strengthens the global standing of Sungkyunkwan University and Korea’s semiconductor research community.
ISCA 2026 was held from June 27 to July 1 in Raleigh, North Carolina, bringing together approximately 1,000 researchers and industry experts from around the world. Leading global technology giants, including NVIDIA, Meta, Google, and Microsoft, as well as prominent universities worldwide, participated in the conference.
A total of 845 state-of-the-art research papers were submitted. Following a rigorous review process, 161 papers were accepted, and Professor Kim’s research was selected as the conference’s Best Paper.
The award-winning paper, titled “Cerberus: Cross-Layer ECC Co-Design for Robust and Efficient Memory Protection,” presents an innovative technology that improves the reliability and efficiency of High Bandwidth Memory, or HBM.
HBM is a critical high-performance memory technology for artificial intelligence systems and has become a major focus of global technological competition amid the rapid expansion of AI. At its core, Cerberus detects and addresses the small data errors that occur while computers store and transfer information more quickly and intelligently than conventional approaches.
As HBM is pushed toward its physical performance limits, errors occurring during data storage and transmission are becoming more frequent. These errors can reduce manufacturing yield and cause serious reliability and efficiency problems in large-scale AI infrastructure. Hardware failures already interrupt a substantial proportion of large-scale AI training workloads, significantly reducing the likelihood that long-running training jobs will be completed successfully.
Conventional systems protect data through several independent error-correction mechanisms: within the memory device, across the data interface, and at the overall system level. Because these mechanisms were developed and operated separately, they require duplicated redundancy and may introduce unnecessary delays and protection holes.
Cerberus, developed by Professor Kim’s research team, unifies these separate protection mechanisms and enables them to share error-correction information. Based on an HBM4-class memory configuration, the proposed technology reduces the storage space dedicated to error-correction redundancy by approximately 33% while simultaneously improving data reliability and system efficiency.
Professor Kim’s research team has established a strong international track record in memory reliability and security. Its previous research has repeatedly been selected as a Best Paper Award finalist at leading international computer architecture conferences, including ASPLOS, HPCA, SC, and DATE.
The ISCA 2026 Best Paper Award further demonstrates the technical excellence, originality, and academic impact that the research team has built over the years.
Professor Jungrae Kim said: “As international competition for technological leadership in artificial intelligence continues to intensify, it is deeply meaningful that our research team has become the first in Korea to receive the Best Paper Award at ISCA, the world’s most prestigious computer architecture conference.”
He added: “I hope this research will make a meaningful contribution to improving the reliability of HBM, an essential technology in the AI era. I also hope it will provide a strong technological foundation for Korea’s memory semiconductor industry, a key pillar of the national economy, to maintain its leadership in the global market.”
This research was supported by the Institute of Information & Communications Technology Planning & Evaluation under programs funded by the Korean government.