Samsung's AI inference chip Mach-1 is about to be prototyped and trial-produced, and is expected to be based on 4nm process

On May 10, South Korean media ZDNet Korea quoted industry insiders as saying that Samsung Electronics' AI inference chip Mach-1 will soon be prototyped in the form of MPW (multi-project wafer), and is expected to be based on Samsung's own 4nm process.

The industry insider also said that the possibility of Mach-1 adopting 5nm process is not ruled out.

Samsung has set a timetable for Mach-1: mass production in the second half of this year, chip delivery at the end of this year, and delivery of inference servers based on the chip in the first quarter of next year.

At the same time, Samsung has also received a pre-order of up to 1 trillion won (IT Home Note: currently about 5.28 billion yuan) from Naver .

Although Samsung Electronics can currently provide 3nm foundry, using the more mature 4/5nm process on Mach-1 can improve the stability of project execution.

The Mach-1 chip is seen as an important step in Samsung Electronics' AI strategy. Previously, Samsung Electronics' AI-related products were mainly high-performance storage and edge AI SoCs, and it has not yet entered the field of server-level computing chips.

Mach-1 inference chip uses LPDDR off-chip memory. Samsung Electronics Co-President Kyung Gyu-hyun once said that the development goal of this chip is to reduce the data bottleneck between off-chip memory and computing chips to 1/8 through algorithms, while also increasing energy efficiency eightfold.

Samsung Electronics is currently recruiting more talents for the development of Mach-1: externally, Samsung announced 8 recruitment positions related to the chip in April; internally, Korean media reported that Samsung is transferring part of the personnel of the automotive semiconductor team of the System LSI Department to the Mach-1 project.