Samsung Electronics' foundry division is rolling out tailored solutions aimed at the three priorities fabless customers weigh most when choosing a chipmaker — process competitiveness, development speed and infrastructure readiness — in a bid to secure leadership in the artificial intelligence (AI) semiconductor ecosystem. On September 15, Samsung Electronics Foundry's SAFE-IP group head Kim Nok-shin (금낙신) outlined the company's tailored strategy and key customer success stories at "IP-SoC Day Korea 2026".
IP-SoC Day is a conference hosted by Design&Reuse (D&R), which operates a marketplace for semiconductor design assets (IP), and is held annually in Korea, the United States, China and Europe. Kim noted that with the AI paradigm shift accelerating product launches, customers face situations requiring them to fabricate new chips within a year. To meet that need, Samsung Foundry is fully supporting these three core customer requirements with tailored solutions to secure leadership of its ecosystem.
Samsung Foundry rolls out tailored solutions to meet customer needs
Samsung Foundry said its tailored solutions have translated into real market outcomes and shared three key customer success stories. The first involves a representative domestic AI accelerator company that used Samsung's 4nm process to integrate multiple dies — including high-bandwidth memory (HBM) and a neural processing unit (NPU) — on a silicon interposer to create a multi-chip product. With support from a holistic power/signal integrity (PI/SI) solution, power efficiency was optimized, and the product is now in mass production. The company is believed to be Rebellions.
The second case centers on Groq's ultra-large LPU accelerator, which features an enormous die area. The product integrates more than 4GB of custom SRAM on a die larger than 700㎟ while securing stable yields and overcoming manufacturing challenges. Through high-speed interface IP bring-up and logic redundancy techniques, both yield and target performance (PPA) were achieved. The company in this case appears to be Groq, which was acquired by Nvidia.
The third case involves a 2nm automotive chip that is leading the next-generation autonomous driving market. Its development cycle was shortened from the typical 2 to 3 years to just 9 months. By fully mobilizing design migration solutions, early-stage PPA analysis and an AI- and machine-learning-based automation platform, ultra-fast tape-out was achieved. The company is believed to be U.S.-based Tesla.
Item Specification Process 4nm SRAM capacity More than 4GB Development cycle 9 months
Disclosure: This article contains affiliate links to partner products. If you purchase through these links, TechRitual may earn a commission. This does not affect product reviews or recommendations. For details, please see our Privacy Policy.