Samsung Electronics announced on the 31st that it will collaborate with NVIDIA to establish the industry's largest 'Semiconductor AI Factory'.

The plan is to apply AI throughout the entire semiconductor manufacturing process to shorten development and mass production cycles while simultaneously enhancing production efficiency and quality competitiveness through an intelligent manufacturing platform.

The collaboration between Samsung Electronics and NVIDIA began 25 years ago when Samsung Electronics supplied DRAM for NVIDIA's first graphics card, 'GeForce'. Since then, the two companies have expanded their cooperation scope to include memory, system semiconductors, and foundry, and through this project, they will expand their relationship into an 'AI Semiconductor Alliance' encompassing the semiconductor and AI convergence ecosystem.

This project is a strategic collaboration aiming to transform the semiconductor manufacturing paradigm by combining Samsung Electronics' comprehensive semiconductor capabilities with NVIDIA's GPU-based AI technology.

Samsung Electronics plans to introduce over 50,000 NVIDIA GPUs in the coming years to expand the AI factory infrastructure and utilize the simulation platform 'Omniverse' to recreate the entire process in a digital twin environment.

The Intelligent Manufacturing Platform being pursued by Samsung Electronics is a system where AI independently analyzes, predicts, and controls the entire cycle of production data.

The core is to learn data from all stages such as design, process, equipment, and quality in real-time and automatically correct any process abnormalities detected in advance. This method, where AI judges and operates lines previously controlled by humans, essentially corresponds to an 'autonomous semiconductor factory'.

Samsung Electronics has already introduced NVIDIA's 'cuLitho' and 'CUDA-X' in some processes to correct microcircuit design distortions in real-time. Through this, the process simulation speed has improved 20 times compared to the past, and design accuracy and production speed have also been enhanced. Samsung Electronics explained, "We are maximizing operational efficiency through digital twin technology based on Omniverse, such as detecting equipment abnormalities and optimizing production schedules."

In addition, along with the establishment of the AI factory, Samsung Electronics will supply next-generation memory and foundry services such as HBM3E, HBM4, GDDR7, and SOCAMM2 to NVIDIA. In particular, HBM4 is a product combining 1c (10-nanometer class 6th generation) DRAM and 4-nanometer logic process, achieving a bandwidth exceeding 11Gbps, surpassing the existing standard (8Gbps). It is optimized for large-scale AI computation environments based on ultra-high-speed and low-power characteristics and is expected to critically contribute to enhancing NVIDIA's AI platform performance.

Currently, Samsung Electronics is supplying HBM3E to major global customers, and sample shipments of HBM4 have been completed to all requesting customers.

Samsung Electronics stated, "We are conducting preemptive facility investments to respond without disruption to the rapidly increasing demand for HBM4," and added, "We are also expanding cooperation in high-performance graphic DRAM (GDDR7) and next-generation low-power modules (SOCAMM2)."

Samsung Electronics plans to expand this technological innovation to major domestic and international production bases. The company will apply AI factories to global lines, including Pyeongtaek in Korea and Taylor in the United States, to promote the intelligence and efficiency of the entire supply chain.

At the same time, it plans to drive qualitative growth in the industrial ecosystem by strengthening cooperation with domestic fabless, materials, equipment, and EDA companies and supporting the smart factory transformation of small and medium-sized partner companies through the 'Smart Factory 3.0' project.

Collaboration in AI application fields will also proceed concurrently. Samsung Electronics is operating its own AI model based on the Megatron framework on NVIDIA GPUs and is advancing humanoid robot autonomy technology using the RTX PRO 6000 Blackwell Server Edition and Jetson Thor platform.

Additionally, the two companies have signed a Memorandum of Understanding (MoU) to verify next-generation intelligent base station AI-RAN technology and are jointly researching real-time computation and communication systems for the physical AI industry, including robots and drones.

A Samsung Electronics representative stated, "Starting with the supply of DRAM for NVIDIA graphics cards, we have continued technical cooperation in all areas, including foundry, for over 25 years," adding, "This project is a symbolic achievement of the collaboration between the two companies, representing the realization of the industry's top-level semiconductor AI factory."

Popular News