Samsung Electronics Co. has shifted a majority of its advanced 4-nanometer foundry capacity toward producing base dies for its sixth-generation high-bandwidth memory, a move aimed at securing supply for AI accelerator makers including Nvidia, Broadcom and AMD in the second half of the year.
The Korean chipmaker has allocated more than 50% of its 4nm wafer output to HBM4 base die manufacturing as of last month, according to ZDNet Korea, which cited sources inside and outside the company. The fabrication lines are running at full capacity, with HBM4 base dies accounting for 50% to 60% of total 4nm production.
Samsung began mass shipping HBM4 in February, claiming an industry first, though initial volumes were modest. The company now plans to expand supply meaningfully starting in the third quarter. During its second-quarter earnings call, management said HBM4 revenue would more than triple quarter over quarter in Q3 and comfortably exceed 60% of total HBM sales in the second half.
At Samsung’s Pyeongtaek Campus 2 and Campus 3, the S4 and S6 foundry lines handle processes ranging from 4nm to 7nm. Total 4nm capacity at those facilities is estimated at roughly 30,000 wafers per month, with about 15,000 wafers now dedicated to HBM4 base dies, according to the report.
A person familiar with the matter told ZDNet Korea that the 4nm process is operating at full production and that the elevated allocation for HBM4 base dies will persist through the remainder of the year.
Why the base die matters
The base die sits beneath the vertically stacked DRAM core dies in an HBM package and manages data transfer between the memory and the system semiconductor. As HBM generations advance with higher data rates and pin speeds, the base die has become a critical performance bottleneck, pushing manufacturers toward logic foundry processes rather than traditional memory fabrication.
Samsung’s integrated structure gives it a distinctive position: it designs both the DRAM and the base die in-house, then fabricates the base die on its own logic process. The HBM4 base die uses the company’s SF4 4nm technology, which Samsung has described as more advanced than what competitors are using.
The shift toward logic-node base dies began with HBM3 and HBM3E. Since then, the competitive landscape has diverged. SK hynix has outsourced base die production to Taiwan Semiconductor Manufacturing Co., while Micron Technology is believed to rely on a custom process.
Samsung is also evaluating more advanced nodes for future HBM base dies. A separate report in August said the company was considering repurposing a production line in its Giheung industrial district to build 2nm base dies, potentially reserved for chips designed for Nvidia.
Supply positioning for AI demand
The capacity allocation signals Samsung’s intent to compete aggressively in the HBM4 market as AI accelerator vendors ramp next-generation products. Nvidia’s upcoming Vera Rubin platform is expected to incorporate HBM4, and securing base die supply is a prerequisite for volume shipments.
For Samsung’s foundry business, the strategy carries trade-offs. Dedicating more than half of its leading-edge 4nm capacity to memory base dies reduces wafer availability for external logic customers. But it also provides internal demand visibility and strengthens the company’s position in a memory segment where it has trailed SK hynix in recent years.
The HBM4 ramp comes at a pivotal moment for Samsung’s semiconductor division, which has faced pressure to demonstrate competitiveness in both memory and foundry operations. The company’s ability to execute on HBM4 volume could influence its share of AI accelerator contracts through 2026 and beyond.