[For more diverse corporate information on the startups mentioned in this article, please visit Unicorn Factory's big data platform 'DataLab'.]
On the 8th, semiconductor design platform SemiFive announced that it has begun Yangsan of HyperXcel's LLM (Large Language Model)-specialized inference semiconductor 'Verda'.
SemiFive stated on this day that Verda has started Yangsan at Samsung Foundry's 4-nanometer process. SemiFive is providing turnkey support for the entire process, from front-end design and verification of Verda to packaging, testing, and Yangsan supply.
HyperXcel's Verda is a semiconductor specialized for LLM inference operations. Its architecture uses LPDDR (Low Power Double Data Rate) instead of HBM (High Bandwidth Memory), maximizing power efficiency in LLM inference. With a 'Big Die' chip structure featuring an area of over 500㎟, SemiFive managed power, heat, and yield to support Yangsan.
This Yangsan marks the first large-scale case where SemiFive applied Samsung Foundry's most advanced 4-nanometer process. Following the initial Yangsan contract, additional purchase orders (POs) are expected as HyperXcel expands its services, leading to a continued increase in Yangsan volume.
Jo Myung-hyun, CEO of SemiFive, said, "The customer's innovative architecture combined with SemiFive's turnkey capabilities successfully executed the design-to-Yangsan process for a chip with an area exceeding 500㎟." He added, "This is a landmark achievement demonstrating competitiveness in the most advanced process. We will seamlessly continue the Yangsan growth momentum verified in the first half of the year into the second half."
[MoneyToday Startup Media Platform Unicorn Factory]
in FUTURE
