SK hynix has unveiled a new high bandwidth memory technology, iHBM, designed to reduce overheating in AI computing systems. The company says that as HBM packages stack more layers and run at higher speeds to meet growing AI demand, heat management becomes a key bottleneck, particularly in the die-to-die physical layer (D2D PHY), the high-speed interface between HBM base dies and AI processor dies. SK hynix adds that heat concentration in this area creates a challenge for controlling power density in next-generation HBM designs.
In iHBM, SK hynix embeds integrated cooling elements (ICE) directly inside the HBM package, positioned within the D2D PHY region. The cooling element uses electrically non-conductive but thermally conductive silicon material to create a dedicated heat path inside the package. The approach is intended to replace conventional designs where heat is routed indirectly through core dies.
SK hynix states the change lowers thermal resistance by more than 30 percent versus existing designs, supporting more stable operation under demanding, high-temperature conditions. The company also says the solution leverages market-proven related technology to help reduce barriers to adoption.
