Samsung Plans to Convert Giheung R&D Line into 2nm HBM Foundry Capacity
Nashnova编辑部
Samsung is considering repurposing an R&D line at its Giheung campus into a 2nm foundry assist line for HBM base dies, targeting production in late 2028 — a move that would convert research infrastructure into the advanced-node capacity most strained by AI memory demand.
What is the plan for this line?
NRD-K Line 2 at Samsung's Giheung campus was originally designated for R&D. Samsung is now weighing a conversion to a send-fab — an assist line that receives wafers from other fabs and handles specific process steps.
The target product: 2nm logic base dies for next-generation HBM, with Nvidia among the intended customers.
This means → Samsung is not building new capacity from scratch but "re-zoning" an existing facility to plug an advanced-node gap faster.
Key caveat: Samsung has not publicly confirmed the plan, and no equipment purchase orders have been placed.
Why are base dies becoming more critical?
Inside an HBM package, the base die sits at the bottom of the DRAM stack and manages high-speed data traffic between memory and GPUs.
In plain terms = think of it as the building's foundation plus wiring — the taller the floors (DRAM layers) and the faster the speed, the more the foundation has to handle.
Samsung's roadmap: HBM4/4E base dies use 4nm; HBM5 moves to 2nm with GAA transistors, boosting speed over 50% versus HBM4E.
This reflects a deepening pattern: each HBM generation consumes more of the same advanced-node capacity shared with mobile and AI chip customers, tightening supply with every step.
How tight is advanced-node capacity?
Samsung has said utilization at 8nm and below is already at ceiling.
The company expects the number of 2nm project wins in 2026 to more than double versus 2025; major cloud and AI/HPC customers have entered the product-design phase.
Expansion timeline: Taylor Fab 1 in Texas is ramping in stages; Taylor Fab 2 targets 2030 production. The Giheung send-fab, if realized, would supply bridging capacity between the two.
This means → Samsung's problem is not whether to expand, but that existing capacity cannot hold until new fabs come online.
What is Samsung's integration advantage?
Samsung houses memory, foundry, and packaging under one corporate roof — DRAM core dies from its memory division, base dies from foundry, final assembly from packaging.
In plain terms = rivals such as SK hynix must outsource base-die fabrication to TSMC and then bring wafers back for packaging. Samsung can run the entire flow internally.
This lets all three divisions co-optimize speed, power, thermals, and yield from the design stage, shortening time to volume production.
Why has this line's plan changed before?
NRD-K Line 2's purpose has already been revised once: before July this year, Samsung's memory division had earmarked it as a DRAM assist line with an investment plan in place.
This month, the direction shifted — new DRAM capacity will concentrate at the Pyeongtaek campus, and Giheung is now leaning toward 2nm foundry work.
NRD-K overall is a ₩20 trillion complex planned to house three lines by 2030; Line 1 was completed in late 2024, and site preparation for Line 2 is underway.
This reflects an internal tug-of-war between memory expansion and foundry expansion — the scale tipping toward foundry signals that AI-driven 2nm demand pressure is already reshaping resource allocation inside Samsung.
What can be confirmed, and what cannot?
Confirmed: the industry trajectory from 4nm to 2nm base dies is set, and Samsung's advanced-node capacity is genuinely strained.
Unconfirmed: the Giheung send-fab plan lacks official sign-off, is roughly two years from production, and Samsung retains room to change course as the market evolves.
This means → whether this line materializes will be a key test of Samsung's ability to close the gap with SK hynix in the AI memory supply chain.
Content is for reference only, not financial advice.