SK hynix unveils self-cooling iHBM chips to combat AI overheating

SK hynix on Tuesday unveiled iHBM, a new high bandwidth memory technology that embeds a proprietary cooling element directly inside the HBM package to reduce heat generation in AI computing environments. The company said the integrated cooling elements, placed within the high-speed data interface between the HBM base die and AI processor dies, lower thermal resistance by more than 30 percent compared with existing designs. SK hynix plans to apply iHBM beginning with HBM5 and subsequent products targeting high-performance computing and AI data center applications.

SK hynix on Tuesday revealed its new high bandwidth memory technology iHBM, which embeds a proprietary cooling element inside the HBM package to substantially reduce heat generation in artificial intelligence AI computing environments. The company said heat management has become a critical bottleneck as HBM stacks more layers and operates at higher speeds to keep pace with surging AI workload demand. Heat concentration is most acute at the die-to-die physical layer D2D PHY , the high-speed data interface between the HBM base die and AI processor dies. Controlling power density in that zone has emerged as a key differentiator in next-generation HBM development. iHBM addresses the problem by placing integrated cooling elements ICE , directly within the D2D PHY area. ICE uses electrically non-conductive but thermally conductive silicon material to form a dedicated heat path inside the package, replacing the indirect routing of heat through core dies used in conventional HBM. SK hynix said the approach lowers thermal resistance by more than 30 percent compared with existing designs and maintains stable operation under high-temperature, high-load conditions. The company said iHBM is also designed for manufacturability. It is built on the company's Advanced Mass Reflow Molded Underfill-based wafer-level packaging process, which has already been validated in mass production. High design compatibility with existing system-in-package environments means customers can adopt the technology without major redesign work. SK hynix plans to apply iHBM beginning with HBM5 and continuing with subsequent products targeting high-performance computing and AI data center applications. Lee Kang-wook, senior vice president and head of packaging development at SK hynix, said iHBM combines memory design and advanced packaging expertise to minimize heat generation and that the company would use it to reinforce its leadership in AI memory. This article was published with the assistance of generative AI and edited by The Korea Times.