Samsung launched its first AI phone, the Galaxy S24, earlier this year. Before that, the South Korean firm launched its own Large Language Model (LLM) called Samsung Gauss, which powers some of the AI features on the Galaxy S24. It is reported that Samsung will launch its own AI accelerator chip next year.
Samsung's first AI accelerator chip will be called Mach-1
According to a report from Seoul Economic Daily, Samsung's first in-house AI accelerator chip will be called Mach-1, and it will be launched in 2025. The company will reportedly try to break Nvidia's stranglehold on the AI accelerator market and restore itself as the world's biggest semiconductor chip company.
Kye Hyun Kyung, CEO of Samsung Semiconductor, announced at Samsung Electronics' 55th annual general shareholders' meeting that the company is developing its own AI accelerator chip. He said, “A specialized lab is working to create an entirely new type of semiconductor, a semiconductor designed to meet the processing requirements of future AGI.”
The technical verification of the chip has already been completed through FPGA, and the SoC design is in progress. The chip will go into production by the end of this year, and an AI system using the chip will be available early next year.
Mach-1 is tailor-made for AI inferencing. Currently, the problem with AI inferencing is that there is a bottleneck between the data transfer between the GPU, the CPU, and the memory. Mach-1 is expected to have an unprecedented structure that can reduce the data transfer speed bottleneck to 1/8th of current levels. In addition, it will be a lightweight AI chip that can use LPDDR memory and not depend only on HBM.
Samsung plans to compete with AMD and Nvidia with its own AI accelerator chip
Major tech firms, including Microsoft and Meta, use Nvidia's GPUs to build their AI infrastructure. Some other firms, including Amazon, Alibaba, AMD, IBM, Intel, and Google, have their own AI chips. Groq is a new firm founded by ex-Google employees, and it has developed its own AI chips. Samsung will join those firms next year.
Samsung is trying to get a contract to supply HBM memory for Nvidia's AI GPUs. However, it is facing tough competition from SK Hynix, which ranked first in the segment. Samsung recently established an artificial general intelligence (AGI) lab in Silicon Valley, USA, to develop AI semiconductors.
