Samsung to Manufacture Cutting-Edge 2nm AI Chips for PFN, Securing First Major Contract Over TSMC in This Semiconductor Space

Samsung managed to be the first to implement GAA transistors after years of drifting and amid a truly controversial decision-making process within the company. They had to accept that their 7nm technology was not on par with TSMC’s despite having similar foundations, and afterwards, they shifted focus entirely to 3nm technology with these transistors. Facing issues after completing the milestone, Samsung sought help from others and eventually considered GAA transistors a viable option. Now, their most advanced version, the 2nm (technically speaking, SF2) is already producing results as Samsung has snatched the first major IA chip contract for 2nm technology from TSMC, partnering with PFN.

We covered Samsung’s struggles and the assistance they received from other companies to improve their success rates. After hard work, the Korean company is now ready and has secured its first major contract with an AI company from their home country.

Samsung will manufacture PFN’s AI chips on their 2nm SF2 technology. PFN, or Preferred Networks, is a Japanese start-up focused on critical research in deep learning development, particularly for AI in IoT applications.

Although it may appear as a small start-up, PFN is one of the giants in AI at the moment. The Japanese company is currently working on a new AI project that, similar to OpenAI, requires a large number of chips.

While the exact number of chips needed is not known, PFN, projected to be Japan’s most advanced AI company by 2024, requested 2nm wafer and chip samples from both Samsung and TSMC. After reviewing the samples, PFN decided to sign a contract with Samsung to produce AI chips using the 2nm technology, likely through their next-generation accelerators.

There are several reasons why PFN chose Samsung over TSMC. First, TSMC is experiencing delays and will only have mass production in LP version by the end of next year. Second, Samsung offered PFN an attractive price, which they can afford after saving a considerable amount on R&D at the 5nm node. Lastly, Samsung already has mass production capabilities this year and no other clients yet for their new node, allowing them to guarantee a large supply of chips at a low price. This combination has led PFN to choose the Korean company.

Additionally, other companies, especially American ones, will notice that Japan’s leading AI company trusts Samsung, opening doors for them in the semiconductor market.

“The successful foundry collaboration with PFN will serve as an entry point for Samsung Electronics into Silicon Valley, United States, the largest semiconductor market,” said an industry official in South Korea.

As a result, TSMC has been left behind. Samsung is about to mass produce its 2nm SF2 node, and by mid-year, they are expected to start shipping their first orders. Intel is set to arrive around the same time with its Intel 20A, while TSMC has lost momentum and will only supply Apple by the end of next year.