Monday, November 23, 2026

SK hynix: "Processing-in-Memory Can Resolve AI Data Bottlenecks"

Input
2026-09-17 11:19:30
Updated
2026-09-17 11:19:30
Eui-Cheol Lim, SK hynix vice president in charge of Solution AT, is presenting the company's processing-in-memory (PIM) technology at the AI Infrastructure Summit held at the Santa Clara Convention Center in California on the 16th (local time). Yonhap News

[Financial News] SK hynix has presented a vision for resolving AI data bottlenecks by leveraging processing-in-memory (PIM) technology, which performs computations directly within memory semiconductors.
Eui-Cheol Lim, SK hynix vice president in charge of Solution AT, said at the AI Infrastructure Summit in California on the 16th (local time) that PIM could overcome the limitations of AI infrastructure centered on graphics processing units (GPUs) and high-bandwidth memory (HBM).
Unlike conventional architecture, which transfers data from memory to AI accelerators such as GPUs for computation and then sends it back to memory, PIM uses computing units inside memory to handle simple operations directly.
Samsung Electronics also unveiled PIM for on-device AI at Hot Chips, a semiconductor conference held last month.
Lim noted that the AI industry currently relies mainly on static random-access memory (SRAM) when ultra-fast inference is required. "SRAM-based systems provide high bandwidth, but they have clear limitations in terms of capacity and cost," he said.
"PIM provides approximately 300 times the capacity of SRAM in the same area while maintaining high internal bandwidth," he explained.
Regarding issues long cited as limitations of PIM, including insufficient memory capacity and higher unit costs resulting from larger circuit areas, Lim emphasized that they could be addressed through hybrid bonding. This technology separately fabricates the memory die and computing die and connects them vertically.
He also introduced the concept of scaling PIM beyond individual chips to the server-rack level.
Lim stressed that the industry must move beyond an era of single-infrastructure systems that process all AI services using only a GPU-and-HBM combination. He called for hardware innovation across the memory hierarchy, including PIM and high-bandwidth flash (HBF), along with software optimization and integrated design.
[email protected] Lee Seok-woo, International Affairs Specialist Reporter