SemiFive Wins KRW 70 Billion Order From U.S. AI Fabless Company to Develop Next-Gen AI Inference Accelerator
TECHWORLD ·
✦ AI Summary
SemiFive announced on the 29th that it had signed a contract with a U.S. AI fabless company to develop a next-generation AI inference accelerator.
The contract is worth USD 52 million, or about KRW 70.3 billion, making it SemiFive's largest single deal to date and its first spec handoff project in the North American market.
SemiFive said it plans to tape out in the first half of 2027 and begin global mass production in 2028, and to use the order to demonstrate its end-to-end ASIC development capabilities in the North American market.
SemiFive announced on the 29th that it had signed a contract with a U.S. AI fabless company to develop a next-generation AI inference accelerator. The contract is worth USD 52 million, or about KRW 70.3 billion. With this major deal, SemiFive is set to accelerate its push into the North American ASIC market.
This contract is SemiFive's largest single deal to date. Its value is more than 40% of the company's annual order intake of KRW 168.4 billion last year and is equivalent to about 60% of its new orders of KRW 118.9 billion in the first half of this year. It is also SemiFive's first spec handoff project in the North American market.
In a spec handoff model, the customer presents core performance and specifications. Accordingly, the ASIC specialist carries out detailed design, packaging, testing, mass production, and software development. Spec handoff is a structure in which an ASIC design company leads the process from the requirement stage through chip implementation and mass production.
In a typical turnkey model, production begins after the customer's semiconductor design has been completed. By contrast, this spec handoff approach was also presented as a model in which the ASIC design company leads the entire process from the customer's specification stage.
In this project, SemiFive is aiming to secure both high memory bandwidth and data processing performance for AI inference chips, while also focusing on reducing power burden and total cost of ownership (TCO). To that end, it plans to adopt LPDDR6, a next-generation low-power DRAM, and PCIe Gen5 interfaces. It also aims to reduce bottlenecks in the process of transmitting large volumes of data.
SemiFive has experience carrying out multiple Big Die projects with sizes of up to 800 mm². Based on this design capability, the company plans to secure the compute throughput of a single chip and ensure operational stability.
The schedule calls for tape-out in the first half of 2027, followed by global mass production starting in 2028. After mass production begins, the company plans to expand supply to global hyperscalers and cloud service providers (CSPs).
The AI inference market is expanding, and demand is also rising for customized semiconductors optimized for specific workloads and services. With this order, SemiFive plans to demonstrate its end-to-end ASIC development capabilities in the North American market and use that to strengthen its position in the global AI semiconductor market.
Cho Myung-hyun, CEO of SemiFive, said the AI inference market is growing quickly and demand for customized semiconductors optimized for customer services and workloads is beginning to expand in earnest. He said SemiFive will push ahead with the commercialization of a next-generation inference accelerator aimed at transforming the economics of ultra-large AI models, and intends to establish itself as a key custom semiconductor development partner in the global hyperscale infrastructure market and secure an irreplaceable partner status.
Source: TECHWORLD · Park Gyu-chan
Original: https://www.epnc.co.kr/news/articleView.html?idxno=407482
References
This article was produced with the help of an automated content generation algorithm.
Source: TECHWORLD
View originalThis article was summarized and organized by BizCrush based on the original article from TECHWORLD. For exact quotations and full details, please refer to the original article.