In a Digital Twin factory that precisely reproduces an advanced semiconductor fab in virtual space, engineers and artificial intelligence (AI) collaborate around the clock. Engineers simulate the logistics flow of process equipment or test new processes without stopping the actual line, optimizing Production yield. In this process, AI predicts and corrects real-world variables such as signs of equipment failure and production schedules in real time, supporting factory operation.
Samsung Electronics said on the 31st that it will join hands with Nvidia to begin full-scale expansion of such a "semiconductor AI factory." To that end, it plans to introduce more than 50,000 Nvidia GPUs (graphics processing units) over the next few years and accelerate the creation of a Digital Twin manufacturing environment based on Nvidia's simulation platform "Omniverse." It also plans to extend this know-how beyond Korea to major overseas production hubs such as Taylor in the United States, pursuing intelligence across the global semiconductor supply chain.
A Samsung Electronics official said, "We will build the industry's largest semiconductor AI factory through the synergy of our capabilities as an integrated semiconductor corporations and Nvidia's GPU-based AI technology," and added, "Through this, we will lead a paradigm shift in global manufacturing."
◇ "Process simulation 20 times faster," expanding to the entire semiconductor fab
The AI factory pursued by Samsung Electronics is a so-called "thinking" manufacturing system that manages and analyzes the entire semiconductor production process—from design to process to quality control—with AI. AI collects and learns the vast data generated during manufacturing in real time to simultaneously boost production efficiency and quality.
Samsung Electronics has already introduced Nvidia's platform into some processes to pilot an AI-based manufacturing system. By applying Nvidia's AI computing technologies "Curisso" and "CUDA-X," AI was made to predict and correct circuit distortions that can occur in advanced nodes in real time. As a result, process simulation speed became about 20 times faster than before, and design accuracy and development speed improved together.
Samsung Electronics plans to extend this technology across the entire process and advance a Digital Twin-based production system. Through this, it aims to shorten the development and mass production cycle of next-generation semiconductors and further strengthen the competitiveness of its global production lines.
◇ Toward an "AI semiconductor alliance"… HBM4 supply virtually confirmed
The cooperation goes beyond building an AI factory and strongly resembles an "AI Semiconductor alliance." While Samsung Electronics pushes manufacturing innovation based on Nvidia's technology, it has also begun in earnest a "two-way cooperation structure" to provide Nvidia with next-generation memory and foundry services.
Samsung Electronics said it plans to supply Nvidia with next-generation memory products such as HBM3E (5th-generation HBM), HBM4 (6th-generation HBM), GDDR7, and SOCAMM2, as well as foundry services. In particular, a Samsung Electronics official emphasized HBM4 as "a product that dramatically improves performance and energy efficiency, and we are in close consultation with Nvidia." Nvidia also noted, "From today's core supply cooperation on HBM3E and HBM4 to now, the two companies have maintained a strong alliance for more than 20 years," signaling that Samsung Electronics' HBM4 supply is virtually assured.
While Samsung Electronics struggled with HBM3E supply, it is achieving a reversal of momentum with HBM4. Samsung Electronics' HBM4, based on 1c (10-nanometer-class 6th-generation) DRAM with a 4-nanometer logic process, delivers ultra-high bandwidth exceeding the JEDEC standard (8 Gbps) to over 11 Gbps. A Samsung Electronics official said, "HBM4, with its ultra-high bandwidth and low-power characteristics, will speed up AI model training and inference and play a key role in enhancing the performance of Nvidia's AI platform."
◇ Expanding cooperation on AI robotics and intelligent base station technology
Cooperation between Samsung Electronics and Nvidia will expand to the broader AI field. Samsung Electronics is using Nvidia's Robotics platform "Jetson Thor" to advance AI inference and control technologies for intelligent robots, and is pursuing commercialization and autonomy technologies for Humanoid Robot based on the "RTX Pro 6000 Blackwell Server Edition" platform.
The companies will also continue cooperation in next-generation communications technologies. Samsung Electronics signed a memorandum of understanding (MoU) with Nvidia and domestic industry, academia, and research institutions for research and demonstrations of next-generation intelligent base station (AI-RAN) technology. The collaboration follows the two companies' successful verification of AI-RAN technology last year and aims to realize physical AI based on next-generation networks.
Samsung Electronics said, "The two companies' 25 years of technological cooperation has evolved into a symbolic achievement: implementing a top-tier AI factory," and added, "This AI factory build-out will lead to cooperation with domestic fabless, materials, and equipment corporations, driving qualitative growth in the national manufacturing ecosystem."