Samsung and Nvidia Team Up on 8-Layer HBM4E Solution

Samsung Electronics is collaborating with Nvidia to develop an 8-layer HBM4E solution, aiming for speeds of 17-18 Gbps/pin. This represents a 20% increase from Samsung's initial HBM4E samples, which reached 14.4 Gbps. The product will be used for NVHBM, a custom AI chip solution designed by Nvidia.

Nvidia has requested that Samsung adopt an 8-layer HBM solution this time, abandoning its previously planned 12-layer and 16-layer configurations. By reducing the number of layers, Samsung can lower the difficulty of back-end processes such as wafer thinning and stacking packaging, which helps improve yields and expand production capacity.

This move is seen as Nvidia's strategy to address HBM supply shortages. The company has released NVHBM, an HBM solution designed for custom AI chip customers using the NVLink Fusion platform. Compared to standard HBM4E solutions, NVHBM offers increased bandwidth of up to 30%, reduced power consumption by 15%, and increased available compute area on the main chip by up to 25%.