Samsung Electronics announced on Friday that it has begun delivering samples of its latest high-bandwidth memory (HBM) chip, the 12-layer HBM4E, calling it the world's first such product shipment. As the demand for high-end memory chips for artificial intelligence servers and processors surges, leading artificial intelligence companies such as AMD, Nvidia, and Google are all customers of Samsung.
Following the industry's first mass production and commercial shipment of its industry-leading HBM4 earlier this year, Samsung has now expanded its HBM roadmap with the launch of HBM4E samples to meet the rapidly evolving needs of artificial intelligence computing and hyperscale infrastructure.
Samsung's HBM4E delivers stable 14 Gbps pin transfer speeds, with performance scalable to 16 Gbps to meet growing data processing needs. Performance improvements of more than 20% compared to HBM4, along with memory bandwidth of up to 3.6 TB/s per stack, help maximize computing performance for large language models (LLMs) and next-generation artificial intelligence systems.
Samsung's 12-layer HBM4E offers 48GB of capacity, more than 30% more than the previous generation, and plans to expand the product line based on customer demand to include 32 GB (8-layer) and 64 GB (16-layer) configurations.
The design and process optimization of Samsung HBM4E’s memory and logic architecture also improve performance, energy efficiency and yield.
Specifically, advanced low-power design technology and optimized packaging structure have improved energy efficiency by 16% compared with the previous generation product, and thermal resistance characteristics have improved by more than 14%. These improvements also enable more efficient cooling, resulting in longer reliability and lower energy consumption in highly loaded next-generation data centers.
Samsung plans to start mass production of HBM4E according to the customer's schedule after completing the first batch of sample delivery and optimization work.