Samsung unveils 36GB HBM3E 12H DRAM, potentially advancing memory capabilities

Samsung has introduced a "groundbreaking" achievement – the HBM3E 12H. This potentially represents a significant milestone as the industry's inaugural 12-stack HBM3E DRAM, boasting "the highest-capacity HBM product to date."

The HBM3E 12H delivers a bandwidth of up to 1,280 gigabytes per second (GB/s), coupled with a storage capacity of 36 gigabytes. Both aspects have seen an improvement of over 50% compared to the 8-stack HBM3 8H. Samsung has employed advanced thermal compression non-conductive film (TC NCF) in the 12-layer configuration, maintaining the same height specification as the 8-layer counterpart to meet existing HBM package requirements.

Samsung's Executive Vice President of Memory Product Planning, Yongcheol Bae, stated:

The industry’s AI service providers are increasingly requiring HBM with higher capacity, and our new HBM3E 12H product has been designed to answer that need, This new memory solution forms part of our drive toward developing core technologies for high-stack HBM and providing technological leadership for the high-capacity HBM market in the AI era.

This technological advancement is poised to offer additional benefits, particularly as the industry addresses chip die warping issues associated with thinner die in higher stacks. According to the tech giant, its commitment to reducing the thickness of its NCF material has resulted in the industry's smallest gap between chips at seven micrometers (µm) and the elimination of voids between layers. Consequently, there has been an enhancement in vertical density, exceeding 20% compared to the HBM3 8H.

The use of Samsung's TC NCF extends to improving the thermal properties of the HBM. This is achieved by enabling the deployment of bumps in various sizes between the chips. During the chip bonding process, smaller bumps are placed in signaling areas, while larger ones are designated for effective heat dissipation.

The adoption of the HBM3E 12H in AI applications is expected to result in a 34% increase in the average speed for AI training, with the potential to expand the number of simultaneous users of inference services by over 11.5 times.

Samsung has commenced sampling of the HBM3E 12H to customers, with mass production scheduled for the first half of this year.

Source