Nam Hyun-woo has worked as a staff writer at The Korea Times since 2013, mostly covering business and politics. He currently belongs to the Business Desk where he covers topics such as emerging tech, AI, ICT and Korea's chaebol community. Prior to joining the team, he was the paper's correspondent for the presidential office of Korea during the Yoon Suk Yeol and Moon Jae-in administrations.
SK hynix ships 12-high HBM4E samples to customers
SK hynix's high-bandwidth memory 4E / Courtesy of SK hynix
By Nam Hyun-woo
SK hynix said Thursday it has shipped samples of high-bandwidth memory 4E (HBM4E) for next-generation artificial intelligence (AI) processors to major customers.
The company said it delivered 12-high HBM4E samples on schedule, and will work closely with partners for mass production “in a timely manner.”
HBM4E is the seventh generation of HBM, a high-performance DRAM designed for rapid data processing and used in graphics processing units (GPUs) and other AI accelerators that require massive computing power.
The previous version, HBM4, is set to be used for Nvidia’s Vera Rubin AI supercomputer, which will be released in the third quarter. HBM4E is also expected to be used for Vera Rubin Ultra platform next year.
SK hynix said the 12-high HBM4E delivers improvements in both performance and power efficiency. The product offers a maximum data transfer speed of 16 gigabits per second (Gbps) per pin and more than 20 percent higher power efficiency compared to previous models.
The HBM4E reduces data transfer latency through its latest interface and design optimization while maintaining stable operation in high-bandwidth environments, SK hynix said.
While mass-producing HBM4, SK hynix has opted for a strategy focused on improving yields and ensuring stability, but experienced minor setbacks during the qualification process, including fine-tuning chip designs before full-scale shipment to meet Nvidia's stringent speed specifications.
The HBM4E samples use most advanced 1c-class DRAM to deliver 16 Gbps of speed and featured the company's Advanced Mass Reflow-Molded Underfill technology, enabling 12-layer HBM4E products to achieve a capacity of 48-gigabyte while ensuring structural stability.
In particular, the company said it improved heat dissipation performance by 17 percent compared to HBM4, allowing memory chips to operate more stably in high-performance computing environments.
Nvidia CEO Jensen Huang introduces the Vera Rubin architecture during a keynote speech on the sidelines of the COMPUTEX 2026 trade show in Taipei, June 1. Reuters-Yonhap
Vera Rubin Ultra is expected to use 12 HBM4E stacks per GPU, up from eight HBM4 stacks per GPU in the Vera Rubin platform. This means memory bottlenecks are expected to aggravate, and SK hynix's ability to supply the chips in a timely manner is becoming increasingly important.
SK hynix said it has provided optimized memory solutions on time throughout its HBM series and will continue working with customers to address AI system bottlenecks with HBM4E.
“SK hynix has laid the foundation to strengthen its AI leadership with HBM4E based on its market-leading technological capabilities and manufacturing expertise,” SK hynix Chief Development Officer Ahn Hyun said. “Through close collaboration with our partners, we will deliver the value needed in the market while reinforcing our technology leadership as a full-stack AI memory creator.”