Samsung’s HBM3E 12H achieves {industry}’s largest capability HBM with groundbreaking 12-layer stack, elevating each efficiency and capability by greater than 50%

Superior TC NCF know-how enhances vertical density and thermal properties

Samsung positioned to satisfy demand for high-performance and high-capacity options within the AI period


Samsung Electronics, a world chief in superior reminiscence know-how, right now introduced that it has developed HBM3E 12H, the {industry}’s first 12-stack HBM3E DRAM and the highest-capacity HBM product so far.


Samsung’s HBM3E 12H supplies an all-time excessive bandwidth of as much as 1,280 gigabytes per second (GB/s) and an industry-leading capability of 36 gigabytes (GB). Compared to the 8-stack HBM3 8H, each facets have improved by greater than 50%.


“The {industry}’s AI service suppliers are more and more requiring HBM with greater capability, and our new HBM3E 12H product has been designed to reply that want,” mentioned Yongcheol Bae, Govt Vice President of Reminiscence Product Planning at Samsung Electronics. “This new reminiscence resolution types a part of our drive towards creating core applied sciences for high-stack HBM and offering technological management for the high-capacity HBM market within the AI period.”


The HBM3E 12H applies superior thermal compression non-conductive movie (TC NCF), permitting the 12-layer merchandise to have the identical peak specification as 8-layer ones to satisfy present HBM package deal necessities. The know-how is anticipated to have added advantages particularly with greater stacks because the {industry} seeks to mitigate chip die warping that include thinner die. Samsung has continued to decrease the thickness of its NCF materials and achieved the {industry}’s smallest hole between chips at seven micrometers (µm), whereas additionally eliminating voids between layers. These efforts lead to enhanced vertical density by over 20% in comparison with its HBM3 8H product.


Samsung’s superior TC NCF additionally improves thermal properties of the HBM by enabling using bumps in numerous sizes between the chips. Throughout the chip bonding course of, smaller bumps are utilized in areas for signaling and bigger ones are positioned in spots that require warmth dissipation. This technique additionally helps with greater product yield.


As AI functions develop exponentially, the HBM3E 12H is anticipated to be an optimum resolution for future programs that require extra reminiscence. Its greater efficiency and capability will particularly permit clients to handle their sources extra flexibly and scale back whole price of possession (TCO) for datacenters. When utilized in AI functions, it’s estimated that, compared to adopting HBM3 8H, the typical pace for AI coaching may be elevated by 34% whereas the variety of simultaneous customers of inference providers may be expanded greater than 11.5 instances.1


Samsung has begun sampling its HBM3E 12H to clients and mass manufacturing is slated for the primary half of this 12 months.



1 Based mostly on inside simulation outcomes