Samsung Foundry to build 2nm processors with HBM for Japanese AI firm
Samsung, Gaonchips, and Preferred Network team up for custom AI processor.
Samsung Electronics has announced a partnership with Japanese AI company Preferred Networks to supply advanced processors made on its 2nm-class process technology (which is presumably also known as SF3P) and packaged using its Interposer-Cube S 2.5D packaging technology. The processors are proprietary AI accelerators for datacenters that Preferred Networks will use for upcoming projects.
The system-on-chip AI accelerator will be developed by Gaonchips, a South Korea-based contract designer of chips that partners with Arm, among others. Samsung does not disclose the specifications or target performance of the processors it will produce for Preferred Networks, but the usage of the Interposer-Cube S 2.5D packaging technology allows it to build multi-chiplet designs with HBM memory. In addition, the joint press release by the two companies implies that we might be dealing with a multi-chiplet project.
Samsung will start high-volume manufacturing of chips on its SF2 fabrication process in 2025. The node is aimed at smartphone and high-performance computing applications, so it is a good fit for multi-chiplet AI solutions for datacenters.
"We are excited to lead in AI accelerator technology with Samsung Electronics's 2nm GAA process," said Junichiro Makino, VP and Chief Technology Officer (CTO) of Computing Architecture at Preferred Networks. "This solution will significantly support Preferred Networks's ongoing efforts to build highly energy-efficient, high-performance computing hardware that meets the ever-growing computing demands from generative AI technologies, especially large language models."
Preferred Networks aims to develop vertically integrated solutions spanning hardware, software, AI models, and supercomputers. To that end, it is not surprising that the company wants to build its own custom silicon processor for its AI supply chain. What is a bit surprising is that Samsung and Preferred Networks decided to disclose the collaboration at all, as chip owners normally prefer not to reveal their foundry partners.
"This order is pivotal as it validates Samsung's 2nm GAA process technology and Advanced Package technology as an ideal solution for next-generation AI accelerators," said Taejoong Song, Corporate VP and the head of Foundry Business Development Team at Samsung Electronics. "We are committed to closely collaborating with our customers, ensuring that the high performance and low power characteristics of our products are fully realized."
Stay On the Cutting Edge: Get the Tom's Hardware Newsletter
Get Tom's Hardware's best news and in-depth reviews, straight to your inbox.
Anton Shilov is a contributing writer at Tom’s Hardware. Over the past couple of decades, he has covered everything from CPUs and GPUs to supercomputers and from modern process technologies and latest fab tools to high-tech industry trends.