Samsung plans to release the Mach-1 AI accelerator chip in 2025

In January 2024, Samsung debuted its first AI phone, the Galaxy S24. Some of the AI features on the Galaxy S24 are powered by Samsung Gauss (the company’s LLM). As of now, it is reported that the company will debut its AI accelerator chip in 2025. This information comes from Seoul Economic Daily. The publication claims that the company’s first AI accelerator chip will be named Mach-1.

According to reports, the company intends to challenge Nvidia’s dominant position in the AI accelerator industry and reclaim its title as the largest semiconductor chip manufacturer in the world. At Samsung Electronics’ 55^th annual general shareholders’ meeting, Kye Hyun Kyung, CEO of Samsung Semiconductor, shared the company’s plan regarding its AI accelerator chip. “A dedicated lab is striving to develop a completely new kind of semiconductor, a semiconductor made to fulfill the processing needs of future AGI,” he stated.

Reportedly, the SoC design of the chip is in progress. The technical verification, on the other hand, has been completed through FPGA. By the end of this year, it will enter production. By next year, an AI system using the chip will debut in the market.

Notably, the Mach-1 is customized for AI inferencing. The bottleneck that exists right now in the data flow between the GPU, CPU, and memory is the issue with AI inferencing. It is anticipated that Mach-1 will have an unparalleled structure that can cut the bottleneck in data transfer speed to 1/8th of what it is at the moment. Furthermore, it will be a thin AI chip that doesn’t rely solely on HBM and is capable of using LPDDR memory.

When it comes to AI chips, Microsoft and Meta rely on Nvidia’s GPU. Other companies including Google, Amazon, Intel, AMD, IBM, and Alibaba have their own AI chips. Former Google employees launched the new company Groq, which has created its own AI processors.

Next year, Samsung is going to be one of those companies. For the development of AI semiconductors, the company recently built an artificial general intelligence (AGI) lab in Silicon Valley, USA. Samsung is attempting to secure a deal to provide HBM memory for AI GPUs made by Nvidia. However, SK Hynix, which came in first in the category, is a fierce rival.