SK hynix adopts hybrid bonding for 2029 HBM5 launch

SK Hynix has introduced the integrated hybrid bonding solution from Applied Materials and BE Semiconductor Industries (BESI) as it targets an HBM5 launch around 2029 to 2030 to meet AI demand for more bandwidth and better power efficiency.

The shift will likely become necessary as chipmakers try to raise performance because thermal compression bonding can support up to 16 layers, given JEDEC’s relaxed standards.

Samsung Electronics, SK Hynix, and Micron are expected to adopt hybrid bonding from HBM4 onward, with HBM5 becoming a substantial turning point around 2029 to 2030 in line with the next-generation AI GPU cycle.

🔗 Source: Chosun Daily

🧠 Food for thought

Implications, context, and why it matters.

A 20 billion won (US$13.2 million) purchase signals a measured bet on a proven but demanding technology

SK hynix is buying one integrated hybrid bonding tool for about 20 billion won (US$13.2 million), with installation planned on a development line 1.
The setup pairs Applied Materials’ chemical mechanical polishing (CMP), which smooths wafer surfaces, with plasma processing gear plus a hybrid die bonder from BE Semiconductor Industries (Besi). Counterpoint says TSMC already runs the same class of tool in mass production for AMD’s 3D V-Cache chip stacking 1.
The aim is to remove solder microbumps. That change can shorten interconnects, cut power use, and raise interconnect density. Those gains matter for future AI accelerators and denser High Bandwidth Memory (HBM) stacks 2.

The shift to hybrid bonding looks like a long rollout, not a quick switch

Looser JEDEC (the industry group that sets memory standards) rules let thermal compression bonding (TCB) support up to 16-layer HBM stacks, which eases near-term pressure to switch to hybrid bonding 3.
Counterpoint expects Samsung Electronics, SK hynix, and Micron to start bringing in hybrid bonding with HBM4. It sees HBM5 as the turning point around 2029 to 2030, matching the next AI GPU cycle 3.
Cost and factory difficulty still lead the list of hurdles. Hybrid bonding needs CMP plus tight particle control and process conditions, which can cost more than microbump-based options 2.
The purchase raises the stakes for equipment suppliers. Applied Materials and Besi now face rivals that keep upgrading microbump-based and related bonding tools, while memory makers weigh when to switch 2.