'Best talks in our long partnership,' Samsung semiconductor chief says after 15-minute meeting; HBM4, SOCAMM to be supplied in full this year; foundry cooperation to extend beyond Grok to next-generation chips
Jeon Young-hyun, Samsung Electronics' co-CEO and head of its Device Solutions division, said Monday that he discussed the supply of HBM4E and HBM5 — the seventh- and eighth-generation high-bandwidth memory chips — with Nvidia CEO Jensen Huang. Jeon said the two sides talked about supplying HBM4E and other products starting next year, and also touched on joint memory development over the medium to long term.
Jeon met with Huang and senior Nvidia executives for about 15 minutes at the Shilla Hotel in Jung-gu, Seoul, on Monday. Speaking to reporters gathered outside afterward, he said, "We have been working together for a long time, and I think today was the best conversation we've had."
"We also talked extensively about long-term cooperation, including HBM4E, our foundry business and HBM5 starting next year," Jeon added. "In the short term, we need to make sure we supply HBM4 and SOCAMM — the low-power server memory module — in sufficient quantities starting this year."
Asked about the expansion of foundry cooperation, Jeon said Samsung is already working with Nvidia on autonomous driving chips and Nvidia's Grok accelerator chip using its 4-nanometer and 8-nanometer processes, and that the two companies are also discussing collaboration on the next generation of chips.
On Huang's earlier remarks that SK Hynix "has been and will continue to be Nvidia's largest memory partner," Jeon said, "We will work hard at what we do and let the results speak for themselves."
The meeting also included Madison Huang, Nvidia's senior director and Jensen Huang's eldest daughter, and Jeff Fisher, Nvidia's senior vice president. On the Samsung side, Kim Jae-jun, vice president of the memory business division, and Jeong Seong-taek, DS division vice president for planning, were also present.
Samsung Electronics is supplying HBM4, its sixth-generation product delivering speeds exceeding 11.7 gigabits per second — among the highest in the industry — for Nvidia's next-generation AI accelerator Vera Rubin, which has entered full-scale mass production. The company is also supplying SOCAMM2, based on LPDDR5X, for the Vera central processing unit.
Samsung recently completed the world's first sample shipment of HBM4E, its seventh-generation product, delivering samples to Nvidia. The HBM4E combines a cutting-edge 1c DRAM core die with a base die built on Samsung's own 4-nanometer foundry process, operating at 14 gigabits per second per pin and achieving a maximum of 16 Gbps — equivalent to a peak bandwidth of 4 terabytes per second.
go@heraldcorp.com
