Samsung Electronics has begun supplying samples of its next-generation AI accelerator memory, the HBM4E 12-layer, to global customers. Following the mass production of HBM4, the early shipment of HBM4E samples marks a strategic move to regain leadership in AI memory that had shifted during the HBM3E era.
As of May 29, industry sources report that Samsung is the first in the world to start shipping HBM4E 12-layer samples. HBM4E, a successor to HBM4, is a critical memory component that influences the computational performance and power efficiency of AI semiconductors. With the rapid growth in demand for high-performance HBM driven by the expansion of large language models and inference-based AI services, competition among customers to secure supply is intensifying.
The new product incorporates 1c DRAM and 4-nanometer logic dies, allowing Samsung to achieve both process stability and mass production capabilities. The operating speed per pin is 14 Gbps, with a maximum potential of 16 Gbps. The bandwidth for a single stack reaches 3.6 TB per second.
The capacity for the 12-layer version is 48 GB. Samsung plans to expand its lineup to include 32 GB 8-layer and 64 GB 16-layer products to meet customer demand. As AI accelerator companies increase model sizes and computational loads, HBM is emerging as a critical technology that influences overall system performance rather than just a simple component.
Market research firm TrendForce noted earlier this year that during the HBM4 validation phase, Samsung was ahead in terms of stability due to process improvements. With the growing demand for AI, major customers are likely to utilize Samsung, SK Hynix, and Micron to secure their supply chains.
Samsung emphasizes its unique advantage in turnkey capabilities that integrate memory, foundry, system LSI, and advanced packaging. The HBM4E features 1c DRAM and proprietary 4-nanometer logic dies. Given that HBM products require not only high memory cell performance but also quality in base die and packaging, Samsung highlights its comprehensive semiconductor structure as a strength.
Power efficiency and heat dissipation improvements are also key points. Samsung claims that the energy efficiency of the HBM4E 12-layer product has improved by 16% compared to its predecessor, with thermal resistance characteristics enhanced by over 14%. As power costs and heat management become critical concerns in AI data centers, both high-speed operation and energy efficiency are becoming essential evaluation criteria for customers.
Following the world’s first mass production shipment of HBM4 in February, Samsung is also expanding its supply. Since both HBM4 and HBM4E are based on the combination of 1c DRAM and 4-nanometer base dies, there is a strong sentiment in the industry regarding the likelihood of transitioning HBM4E to mass production.
The key factors will be customer certification and actual volume expansion. In the HBM market, the adoption by major AI semiconductor customers will significantly influence performance outcomes more than technology announcements. With Samsung accelerating the supply of HBM4E samples following the mass production of HBM4, there is a growing possibility of changes in the existing competitive landscape starting from the HBM4 generation.
Hwang Sang-jun, Executive Vice President of Samsung Electronics' Memory Business, stated, "By successfully completing the mass production of HBM4 and the timely supply of next-generation HBM4E samples, we have solidified Samsung's technological leadership in the market. We will lead the growth of the global AI memory market based on overwhelming technological superiority and proactive investments in production infrastructure."
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.