SK hynix reportedly moves forward with iHBM memory to tackle HBM heat challenges
SK hynix is reportedly taking new steps toward its next-generation high-bandwidth memory technology, known as iHBM, as the company looks for ways to solve one of the biggest problems facing advanced AI and data center hardware: heat.
High-bandwidth memory, or HBM, has become a critical component in modern GPUs and AI accelerators because it delivers extremely fast data transfer rates while saving space. Instead of placing memory chips side by side like traditional DRAM, HBM stacks multiple DRAM layers vertically on top of a base die. This compact design allows the memory to sit close to the GPU through a silicon interposer, helping improve performance and efficiency.
However, stacking memory layers creates a major thermal challenge. As more DRAM layers are added, heat can become trapped inside the stack, especially near the middle. Since cooling hardware such as heat spreaders, cold plates, or liquid coolers typically sits above the package, heat from the lower and inner layers must travel through several materials before it can be removed. This can limit performance, reliability, and future scaling.
According to a new industry report, SK hynix has begun discussions with both domestic and international materials and equipment suppliers about its iHBM plans. The talks reportedly include mass production strategy, commercialization, and potential applications for reducing heat generation and improving thermal management in future HBM products.
The key innovation behind iHBM is the use of integrated cooling elements, also referred to as ICE. These cooling structures are expected to be placed within the die-to-die physical layer, known as D2D PHY, inside the memory package. By adding cooling support directly into the HBM structure rather than relying only on external cooling, SK hynix aims to lower thermal resistance significantly.
The report claims that iHBM could reduce thermal resistance by up to 30%. If achieved in commercial products, this would be an important step for future AI chips, high-performance computing systems, and data center accelerators, all of which depend heavily on faster and denser memory.
At the moment, SK hynix is focused on HBM4 production, but iHBM is expected to play a larger role in the generation after that, likely HBM5. As AI workloads continue to grow, memory bandwidth and power efficiency are becoming just as important as GPU performance. This makes better HBM cooling a priority for chipmakers and system designers.
SK hynix is already one of the leading companies in the HBM market, and its experience with advanced packaging could help it bring iHBM to production. The company is expected to use its Mass Reflow-Molded Underfill technology, also known as MR-MUF, as part of the iHBM manufacturing process. Because this technology is already part of its existing HBM production approach, building a supply chain for iHBM may be more straightforward than starting from scratch.
The move also highlights how competitive the memory industry has become. As AI processors demand more bandwidth, higher capacity, and better energy efficiency, memory makers are under pressure to deliver more advanced HBM solutions. Simply stacking more DRAM layers is no longer enough; future memory must also handle heat more effectively.
If SK hynix successfully commercializes iHBM, it could give next-generation AI accelerators a better thermal foundation, allowing them to maintain high performance under demanding workloads. For data centers, improved HBM cooling could also help reduce power and cooling costs, which are becoming major concerns as AI infrastructure expands worldwide.
While iHBM is still in the planning and development stage, the reported supplier discussions suggest that SK hynix is preparing early for the next phase of high-bandwidth memory. With HBM4 already on the roadmap and HBM5 expected to push performance even further, integrated cooling may become a key feature in the future of advanced memory packaging.
