Samsung Electronics is developing an 8-layer HBM4E product that meets Nvidia's requirements in an effort to secure a key role as a supplier for its custom high-bandwidth memory, NVHBM. The product reduces stacking height from the originally planned 12-layer and 16-layer versions, while Nvidia's target speed specification is 17-18Gbps, about 20% faster than Samsung's initial HBM4E sample speed of 14.4Gbps.
According to ChainCatcher, the 8-layer design is intended for Nvidia's publicly disclosed NVLink-optimized NVHBM and is seen as a way to ease manufacturing difficulty and yield pressure compared with the traditional approach of increasing stack layers to raise capacity. It is also viewed as part of Nvidia's strategy to address memory shortages.
NVHBM is expected to be used in Nvidia's next-generation AI GPU Rubin Ultra, which is scheduled to launch next year and will expand GPU-to-GPU connectivity from up to 72 to 576. Industry sources said Samsung has already demonstrated the industry's highest-level speed in HBM4 verification and may have an advantage in the speed competition.
In the custom HBM market, Samsung is considered more competitive than SK Hynix and Micron because NVHBM requires both DRAM and base die production capabilities based on logic chips, which Samsung can provide in an integrated manner.
Alex
