![]() |
| An architectural concept diagram of SK hynix's iHBM package. /SK hynix |
SK hynix has unveiled "iHBM," a groundbreaking technology that integrates custom cooling mechanics directly inside High Bandwidth Memory packages. The strategy is designed to tackle the escalating thermal challenges driven by intensifying AI computations, cementing the company's competitive edge in next-generation AI memory architectures.
On May 26, SK hynix showcased its Integrated HBM (iHBM) technology engineered to efficiently mitigate high-density heat generation. While recent HBM advancements have yielded rapid performance gains through taller die stacking and faster processing speeds, the accompanying thermal loads have surged simultaneously, elevating the critical importance of stable heat management. In particular, localized thermal density across the die-to-die physical layer (D2D PHY) interface connecting the HBM to the GPU has emerged as a major architectural hurdle.
While conventional HBM models route internal heat outward across the core dies, the iHBM architecture introduces Integrated Cooling Elements (ICE) directly inside the D2D PHY region—the primary source of thermal concentration—to forge an independent, dedicated heat dissipation path.
These ICE structures are built from non-conductive, silicon-based materials featuring exceptionally high thermal conductivity. SK hynix explained that this specialized configuration slashes structural thermal resistance by over 30 percent compared to existing setups, ensuring highly stable operational performance even under sustained, high-temperature workloads.
The new package also demonstrates high viability for volume manufacturing. By utilizing the market-proven MR-MUF mass-production packaging process, SK hynix ensures highly reliable high-volume yields. Furthermore, the architecture maintains high design compatibility with standard system integration ecosystems, allowing corporate clients to adopt the technology without facing disruptive system overhauls.
SK hynix is currently evaluating plans to implement iHBM technology starting with its upcoming next-generation lines, such as HBM5. The objective is to resolve the acute thermal density challenges inherent to ultra-high bandwidth memories demanded by modern AI data centers and High-Performance Computing (HPC) infrastructures, ultimately maximizing system stability and power efficiency.
"The iHBM solution minimizes thermal risks by seamlessly fusing core memory design with advanced packaging breakthroughs," said Lee Kang-wook, Senior Vice President of Packaging Development at SK hynix. "We will continue to proactively address client demands in the AI era to solidify our market leadership in advanced AI memory."
Lee Ji-sun
1
2
3
4
5
6
7