Samsung Wins First 2nm Order to Produce AI Accelerator Chips

On Tuesday, July 9th, Samsung Electronics announced that it has secured an order from Japanese artificial intelligence (AI) company Preferred Networks (PFN). The order involves utilizing Samsung’s state-of-the-art 2nm process technology and advanced chip packaging services to manufacture AI accelerator chips.

This milestone represents Samsung’s first publicly disclosed order for its advanced 2nm chip foundry services, though the specific order volume has not been revealed. Samsung aims to outpace TSMC in mass-producing 2nm technology, thereby gaining a competitive advantage in next-generation semiconductor processes.

In its announcement, Samsung detailed that these AI chips will be produced using Gate-All-Around (GAA) technology and the advanced 2.5D packaging method known as Interposer-Cube S (I-Cube S). This combination is designed to enhance interconnect speed and minimize chip size. The chip design has been developed by South Korean firm Gaonchips Co.

Preferred Networks, established in 2014, focuses on AI deep learning development and has received significant investments from major corporations such as Toyota, NTT, and Fanuc. The partnership with Samsung leverages Samsung’s comprehensive capabilities in both memory and foundry services, along with its extensive technical expertise. Samsung’s ability to offer a complete solution, from high-bandwidth memory (HBM) design to production and advanced 2.5D packaging, was a key factor in this collaboration.

Junichiro Makino, Vice President and Chief Technology Officer of Computing Architecture at Preferred Networks, noted that these AI accelerator chips will be integral to developing the company’s high-performance computing hardware, particularly for applications in generative AI technologies like large language models.