SK hynix presents iHBM: integrated cooling for future HBM5 memories

SK hynix has announced iHBM, a new packaging solution designed to improve the thermal management of future memories intended for AI and HPC workloads. The technology will be introduced in the upcoming HBM generations, including future HBM5, in response to the increasing challenges related to heat dissipation in high-density systems.

As the number of stacks and operational speeds of HBM memories increases, temperature control has become one of the most critical aspects for ensuring reliability and performance. In particular, one of the most sensitive points concerns the Die-to-Die Physical Layer (D2D PHY), that is, the interface that connects HBM memory to the GPU or AI accelerator. This area is characterized by a high concentration of heat and increasingly difficult-to-manage power densities.

The iHBM solution introduces a different structural approach compared to traditional systems. Current HBM uses an indirect dissipation method, where heat is transferred through the main die, or "core die". In the new design developed by SK hynix, cooling elements, called ICE (Integrated Cooling Elements), are integrated directly into the D2D PHY area, creating an additional path for thermal dispersion.

According to data released by the company, this configuration allows for a 30% reduction in thermal resistance, helping to maintain stable chip operation even under high temperatures and severe thermal pressures.

In addition to the design aspect, SK hynix also emphasizes the manufacturing advantages. The company claims that its Wafer Level Packaging (WLP) process, based on the proprietary Mass Reflow Molded Underfill (MR-MUF) technology, will enable high-volume production of memories equipped with iHBM without compromising reliability and yield.

Another highlighted element concerns compatibility with existing System-in-Package (SiP) architectures. According to the company, adopting the new thermal solution will require limited changes to current designs, thus facilitating integration for clients involved in developing next-generation AI accelerators and HPC platforms.

With iHBM, SK hynix aims to further strengthen its position in the memory market for AI, a rapidly expanding segment due to the growing demand for infrastructures for data centers and systems dedicated to artificial intelligence.