Samsung Semiconductor announced today (31st) that it will cooperate with NVIDIA to build an artificial intelligence super factory (AI Megafactory). By deploying more than 50,000 NVIDIA GPUs, Samsung will fully introduce AI into the entire manufacturing process to accelerate the development and production of next-generation semiconductors, mobile devices and robots.

Samsung pointed out that AI Factory will integrate all aspects of semiconductor manufacturing, from design and process to equipment, operations and quality control, into a single smart network, with AI real-time and continuous analysis, prediction and optimization of the production environment. Samsung AI Factory will go beyond traditional automation and is a smart manufacturing platform that can connect and analyze the huge data generated during chip design, production and equipment operation.

Samsung said that in addition to celebrating more than 25 years of cooperation between the two parties, Samsung and NVIDIA are currently jointly developing HBM4. HBM4, which uses Samsung's sixth-generation 10nm DRAM and 4nm logic base die, can achieve a processing speed of 11Gbps, significantly ahead of the JEDEC standard of 8Gbps.

Samsung will continue to provide next-generation memory solutions and foundry services covering HBM, GDDR and SOCAMM to promote the innovation and expansion of the global AI value chain.

Samsung stated that it plans to introduce NVIDIA accelerated computing technology in the next few years to expand the scale of its AI Factory and accelerate digital twin manufacturing through the NVIDIA Omniverse library. This project will be applied to one of the most complete chip manufacturing infrastructures in the world, spanning memory, logic chips, wafer foundry and advanced packaging.

Using NVIDIA cuLitho and CUDA-X libraries to enhance its optical proximity correction (OPC) process, Samsung successfully increased the calculated exposure by 20 times. As a key step in precise wafer patterning, the enhanced OPC technology enables AI to predict and correct circuit pattern errors in a faster and more accurate mode, thereby shortening the development cycle. In addition, in the field of electronic design automation (EDA), both parties are working with EDA partners to develop a new generation of GPU-accelerated EDA tools and design technologies.

Samsung plans to expand its AI Factory infrastructure to global manufacturing centers, including Taylor in the United States, to bring greater intelligence and flexibility to its global semiconductor operations. At the same time, Samsung is also cooperating with a number of NVIDIA AI platforms to connect virtual simulations with real-world robot data, so that robots can understand the environment, make decisions, and demonstrate intelligent operation capabilities in real-life scenarios. Through the NVIDIA Jetson Thor module, Samsung accelerates the capabilities of smart robots in real-time AI reasoning, task execution, and safety control.

Connecting the consumer side and the industrial side through AI

Samsung is collaborating with NVIDIA, Korean telecom operators, academia and research institutions to promote AI Radio Access Network (AI-RAN) research and development. AI-RAN is a key next-generation communication technology that integrates AI computing capabilities into mobile network architecture. This technology allows various AI application devices, such as robots, drones and industrial automation equipment, to achieve real-time operation, sensing, data processing and inference at edge nodes closer to the network. This AI-driven mobile network will become an important "neural network" foundation for promoting the popularization of physical AI.

This new project continues the results of the cooperation between Samsung and NVIDIA last year, when the two parties successfully completed the Proof of Concept of AI-RAN, combining Samsung's software-based network and NVIDIA's GPU technology. In the future, the two companies will continue to deepen cooperation in the field of AI-RAN and jointly promote the development and application of related technologies.

Further reading: AI’s money-burning war begins! Google, Microsoft, and Meta invested a total of US$80 billion, and the market reacted in two ways The three major memory manufacturers all reported good results! WSJ: AI detonates memory and enters "super boom period"