SK Hynix has officially begun supplying samples of its next-generation high-performance DRAM, 'HBM4E.' This announcement comes just a month after Samsung Electronics shipped its own samples, marking a new phase in the competition for dominance in the AI memory market.
On June 18, SK Hynix revealed that it has supplied 12-layer samples of its 7th generation high-bandwidth memory (HBM4) to key customers. This product is a successor to the currently mass-produced 6th generation HBM (HBM4) and is designed to maximize data processing performance essential for AI learning and inference. Initially expected to launch in the second half of this year, the timeline was accelerated due to smooth progress in its development schedule.
The new generation boasts improved performance and energy efficiency compared to its predecessor. It achieves data processing speeds of up to 16 Gbps per pin and enhances energy efficiency by over 20% through optimized power design.
Notably, the product incorporates SK Hynix's proprietary 'Advanced MR-MUF' process, which enhances structural stability. It achieves a high capacity of 48GB based on 12-layer stacking while reducing thermal resistance by approximately 17% compared to the previous generation. This advancement allows for more effective heat management in AI computing environments.
SK Hynix stated, "We are closely collaborating with our key customers based on our accumulated HBM development capabilities and production know-how. We will ensure timely mass production to proactively deliver the value demanded by the market."
In response, Samsung Electronics is also making significant strides. At the end of May, the company showcased its technology by supplying the world's first HBM4E 12-layer samples to global customers. This follows its announcement in February of the world's first mass production shipment of HBM4, marking a remarkable move to supply the next-generation lineup just three months later.
Samsung's HBM4E combines its validated 10-nanometer 6th generation (1c) DRAM with a 4-nanometer foundry process, ensuring both productivity and process stability. The low-power design improves energy efficiency by 16% and enhances thermal resistance characteristics by over 14%. The company's 'one-stop turnkey' strategy aims to create strong synergies, positioning it favorably in securing supply chains with major tech firms.
Both companies have entered a fierce competition for the next-generation AI supply chain. With renewed bidding wars for HBM4 and HBM4E, the company that successfully passes the rigorous validation and final approval from customers is likely to gain the upper hand.
Ahn Gi-hyun, Executive Director of the Korea Semiconductor Industry Association, noted, "Both companies are progressing faster than expected in the development and validation of next-generation products. The timing of passing the final quality tests by global big tech customers and the speed of transitioning to mass production will be pivotal in determining the future leadership in the AI memory market."
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.