en.Wedoany.com Reported - On June 18, South Korea's SK Hynix announced that it has supplied 12-layer stacked HBM4E samples to major customers. This product is a next-generation ultra-high-performance DRAM for artificial intelligence applications, an upgraded high-bandwidth memory primarily used for AI accelerator chips and large-scale data center computing scenarios.
HBM4E is an enhanced product following HBM4, developed to meet demands for higher data throughput, lower power consumption, and improved heat dissipation. The 12-layer sample supplied by SK Hynix increases storage density through vertical packaging of multiple DRAM layers in a single stack, aiming to meet the high-bandwidth memory requirements of generative AI, large model training, and high-performance computing.
In terms of performance parameters, the 12-layer HBM4E sample achieves a data processing speed of up to 16Gbps per pin. Compared to previous products, this product improves energy efficiency by over 20%, helping to reduce the power consumption pressure on AI servers under high-load operation. For AI data centers, the performance of high-bandwidth memory not only affects chip data exchange speed but also directly impacts overall system energy efficiency, thermal design, and deployment costs.
SK Hynix continues to use and strengthen MR-MUF-related technologies in its packaging process. This process improves the stability of multi-layer chip structures and enhances heat dissipation by filling protective materials between stacked chips. Current Korean news reports indicate that the thermal resistance of this product is approximately 17% lower than that of HBM4, which is significant for the stable operation of 12-layer high-density stacked products.
This sample supply marks the entry of HBM4E into the customer verification phase. For AI memory, sample delivery is just an important step before mass production, followed by customer testing, performance optimization, platform adaptation, and batch supply preparation. SK Hynix has not disclosed specific customer names, but its HBM products have long been targeted at AI chip and data center clients. The progress of customer verification will directly impact the subsequent mass production timeline.
Competition in the HBM market is accelerating. Samsung Electronics announced at the end of May that it had supplied 12-layer HBM4E samples to major global customers, and Micron is also advancing its next-generation high-bandwidth memory product lineup. SK Hynix's initiation of 12-layer HBM4E sample supply will further advance the competition between South Korea's two major memory companies in the customer certification and mass production window for next-generation AI memory.
Going forward, the key focus will be on whether SK Hynix can complete performance optimization in line with customer verification schedules and drive HBM4E into large-scale supply around 2027. As AI servers continue to demand higher bandwidth, capacity, power efficiency, and heat dissipation, the progress of 12-layer HBM4E mass production will become a critical milestone for assessing the supply landscape of next-generation AI memory.
This article is compiled by Wedoany. All AI citations must indicate the source as "Wedoany". If there is any infringement or other issues, please notify us promptly, and we will modify or delete it accordingly. Email: news@wedoany.com
