Samsung Electronics to use 2nm process for HBM5 base die, boosting speed by 50%

Samsung Electronics has officially confirmed it will apply its most advanced 2-nanometer foundry process to the base die of HBM5, the next-next-generation high bandwidth memory chip. The company also said orders from major global technology companies using its leading-edge processes have been flowing in steadily since its foundry division won the contract for Tesla's next-generation AI chip, the AI5, last year.

Koo Ja-heum, executive vice president and head of technology development at Samsung Electronics' foundry division, made the announcement Monday through Samsung Newsroom. Koo cited the foundry division's base die technology as the single biggest factor behind Samsung's achievement of being the first in the world to begin mass production of HBM4 and to ship HBM4E samples. He said the foundry division's technological capabilities, combined with its memory chip expertise, allowed Samsung to secure market leadership — and that global technology giants have consistently turned to Samsung Foundry on that basis.

Koo said the importance of the base die has grown as demand for high-bandwidth, ultra-fast HBM operation has intensified. The base die is a logic die positioned at the bottom of an HBM stack, serving as a high-speed interface that connects external processors such as GPUs and CPUs to the HBM. Samsung used a foundry-process-based base die in HBM4 for the first time, allowing it to get ahead of competitors in capturing the market.

"As the demand for high-bandwidth and ultra-fast HBM operation continues to grow, the base dies for HBM4 and HBM4E have adopted Samsung Foundry's fourth-generation 4nm process instead of conventional memory processes," Koo said. "Because HBM5 must improve operating speed by approximately 50% or more over the previous generation, HBM4E, the base die for HBM5 will use a 2nm process with a gate-all-around structure and will implement through-silicon vias at a higher density."

Demand for leading-edge foundry processes is expanding alongside growth in the AI semiconductor market. Samsung began mass production using its first-generation 2nm process in the second half of last year and plans to start mass-producing new mobile products based on an improved second-generation 2nm process in the second half of this year.

"We proved our process competitiveness by winning the Tesla 2nm automotive product — the AI5 — last year, and since then orders have continued to come in from a large number of major customers in AI, high-performance computing and cloud services," Koo said.

He added that the 4nm process has also secured stable yields and performance competitiveness through years of mass production experience and refinement. "In particular, its application to the HBM4 base die and a new LPU product from a US AI company — Nvidia's Groq 3 LPU — has earned recognition for its outstanding performance and process maturity, driving demand," he said.

Samsung Electronics' turnkey solution has played a major role in securing market leadership. "Samsung Electronics' HBM is a turnkey solution product in which the core die is manufactured in memory, the base die in the foundry, and the final product is assembled through packaging at TSP," Koo said.

"Because Samsung handles design, manufacturing and packaging entirely in-house, each division collaborates from the design stage, allowing us to optimize designs for speed, power, heat and yield — and to shorten the time needed to complete mass-production-ready products," he said. "It is also advantageous from a cost perspective."

Samsung Foundry is expanding its application areas beyond mobile to AI, high-performance computing, automotive and robotics. In the AI and high-performance computing space, it offers a one-stop solution linking advanced packaging with HBM, and is also developing silicon photonics, pluggable optics and co-packaged optics technologies for data centers.