SK Hynix unveils iHBM cooling for high‑bandwidth AI memory

SK hynix has unveiled a new memory technology designed to tackle a critical challenge in artificial intelligence computing: heat. The company announced its iHBM solution, which embeds integrated cooling elements directly inside high-bandwidth memory (HBM) packages to manage rising temperatures as chips become faster and more densely stacked.

Heat management has emerged as a significant bottleneck for HBM, a specialized form of memory stacked next to AI accelerator chips. As demand for AI processing surges, HBM stacks are growing taller and operating at higher speeds, generating more heat that limits further performance scaling. The most acute heat concentration occurs at the die-to-die physical layer (D2D PHY), the high-speed interface connecting the HBM to the processor.

The new iHBM technology addresses this by placing silicon-based integrated cooling elements directly within the D2D PHY area. These elements are thermally conductive but electrically non-conductive, creating a dedicated path for heat to escape without interfering with the chip's circuitry. The company states this structural approach reduces thermal resistance by more than 30 percent compared to conventional...