Samsung Hybrid Bonding Line Negotiations Stall, Korean Domestic Suppliers Gain Opportunity
0xBroomberg
Samsung is planning a ~50-tool hybrid bonding line for next-gen HBM, but negotiations with preferred supplier Besi have stalled over customization demands and a ~$4.1 million per-unit price tag. SEMES and Hanwha Semiconductor are racing to fill the gap.
What is Samsung building, and why can't it close the deal?
Samsung is planning a ~50-tool die-to-wafer (D2W) hybrid bonding line at its Pyeongtaek campus for next-gen high-bandwidth memory (HBM) and logic chips.
The talks are stuck on two points: Samsung wants Besi to customize equipment designs, and Besi — already shipping to TSMC and Micron — is unwilling to overhaul its standard platform for one customer. Besi's per-unit quote is ~₩6 billion (~$4.1 million), roughly double the competition.
This means → Besi has pricing leverage from its existing client base, while Samsung wants both customization and a lower price. A near-term resolution looks unlikely.
How far along are Korean domestic suppliers?
Samsung's in-house equipment arm SEMES has completed D2W hybrid bonder development and delivered one system to Samsung for performance validation. Official test results have not been disclosed.
Hanwha Semiconductor Technology announced in February that it finished its second-gen SHB2 Nano system, achieving 0.1 µm alignment accuracy. It plans to ship evaluation units to customers in H1 2026.
In April, Hanwha also delivered an evaluation system to SK Hynix, integrating a Cymechs wafer-transfer module, a Hanwha plasma-activation module, and a Zeus cleaning module.
In plain terms = both domestic players are still at the "sample-and-validate" stage — not yet production-ready. But the longer the Besi talks drag on, the wider their window gets.
Why does hybrid bonding matter?
Hybrid bonding — joining copper pads and dielectric layers directly, face-to-face — replaces the micro-bumps used in thermocompression bonding. It shrinks the gap between stacked chips, theoretically boosting bandwidth, cutting power, and reducing thermal resistance.
At Nvidia's GTC in March, Samsung demonstrated the technique and said it could support 16+ layer next-gen HBM stacks with thermal resistance more than 20% lower than thermocompression.
This means → hybrid bonding is the process leap that takes HBM from "good enough" to "next-gen performance." Whoever reaches volume production first locks in a first-mover position.
When will mass production actually happen?
Samsung internally estimates large-scale hybrid bonding production around 2030, citing process difficulty well above thermocompression.
Other sources point to an earlier target of 2029 — when Nvidia's Feynman-architecture accelerators are expected to launch.
Equipment installation is targeted for late 2026. The line is expected at the Pyeongtaek P5 fab, though Samsung has not officially confirmed the site.
In plain terms = tools go in late 2026, volume production starts 2029 at the earliest. The two-to-three-year gap is tuning and qualification — exactly the period domestic suppliers need to prove they can deliver.
Who has the most at stake in this standoff?
Besi: if Samsung pivots to domestic alternatives, Besi's dominance in advanced HBM packaging equipment weakens. Its moat rests on TSMC and Micron orders, but losing Samsung means losing one of the three major HBM players.
SEMES and Hanwha: this is a rare window to leap from "evaluation prototype" to "production order." Landing Samsung's line would give them a track record to approach other customers.
This reflects a broader shift: advanced packaging equipment is moving from a "buy from whoever" market into a customization standoff, where supplier pricing power and customer willingness to switch are being tested at the same time.
Content is for reference only, not financial advice.