SK hynix Unveils iHBM Cooling Tech, Cutting AI Memory Thermal Resistance by Over 30%
Updated
Updated · Tom's Hardware · May 26
SK hynix Unveils iHBM Cooling Tech, Cutting AI Memory Thermal Resistance by Over 30%
8 articles · Updated · Tom's Hardware · May 26
SK hynix said its new iHBM architecture embeds cooling elements directly into HBM packages, targeting the heat spikes that trigger throttling in AI memory systems.
The design places non-conductive silicon ICEs around the die-to-die PHY—the high-speed link between HBM and AI processors—creating a direct heat-dissipation path at the hottest point.
SK hynix said the approach reduces thermal resistance by more than 30%, helping memory sustain stable operation under high-temperature, high-load AI data center workloads.
The company plans to introduce iHBM starting with next-generation products such as HBM5, and said it can be manufactured with existing wafer-level packaging and MR-MUF processes without major system redesigns.
As rivals push new cooling tech, can SK hynix's iHBM maintain its lead in the AI memory race?
With HBM stacks growing taller and hotter, is integrated cooling the final fix or just a temporary solution?
SK hynix iHBM Slashes AI Memory Thermal Resistance by 30%: Integrated Cooling Redefines HBM for Next-Gen Accelerators
Overview
SK hynix has recently introduced its iHBM thermal control technology, a major innovation designed to tackle the growing heat challenges in advanced AI processors and high-performance computing. As AI accelerators become more powerful, they generate more heat due to increased power density and memory stacking. iHBM addresses this by integrating advanced memory design and packaging technologies, aiming to reduce heat resistance by 30%. Leveraging its proven MR-MUF packaging process, SK hynix enables stable, high-volume production of iHBM chips. With high compatibility for existing systems, iHBM allows customers to adopt this solution easily, reinforcing SK hynix’s leadership in the evolving AI memory market.