2026-06-09 — views · Hanmi Semiconductor (KRX 042700) · TC Bonder 4.5 Griffin · Advanced-packaging bonders (HBM thermo-compression)

SK hynix's 44.2 billion won Griffin order: Hanmi's first HBM4 TC bonder win says the ramp is on schedule — and TC bonding isn't done yet

SK hynix has placed a 44.2 billion won order for Hanmi Semiconductor's new TC Bonder 4.5 Griffin — its first disclosed HBM4 bonder purchase from Hanmi, due in Cheongju by September 2. A small contract with a big signal for the HBM4 ramp and the TC-versus-hybrid bonding race.

What was ordered

According to a June 9 report from Korea’s The Elec, echoed the same day by TrendForce, SK hynix has signed a 44.2 billion won contract — roughly US$30 million — with Hanmi Semiconductor for thermo-compression (TC) bonders dedicated to HBM4 production. The Elec identifies the tool as the TC Bonder 4.5 Griffin, a newly developed system built specifically for sixth-generation HBM, and notes this is the first disclosed order for the Griffin model. The machines go to SK hynix’s post-package-and-test facility in Cheongju, and the contract runs through September 2, 2026 — by which point delivery, installation, test operation, and inspection must all be complete.

TrendForce adds two useful framings: at a typical unit price of around 3 billion won, the deal likely covers about 15 bonders, and the contract equals roughly 7.6–7.7% of Hanmi’s 2025 revenue. For context, Hanmi and SK hynix had already signed a smaller 9.6 billion won TC bonder agreement in January 2026 — but this is the first time SK hynix’s HBM4 line has shown up explicitly in Hanmi’s order book.

Item	Detail
Buyer	SK hynix
Tool	TC Bonder 4.5 Griffin (HBM4)
Value	44.2 billion won (about US$30M)
Estimated units	~15 (TrendForce estimate)
Destination	Cheongju post-package and test facility
Completion deadline	September 2, 2026

Why a small order carries a big signal

Two things make a US$30 million contract worth reading closely.

First, the timing nails down the ramp. A hard September 2 completion deadline — covering install and qualification, not just shipment — means SK hynix expects qualified HBM4 stacking capacity standing in Cheongju before Q4. That lines up with NVIDIA CEO Jensen Huang’s comment during his early-June visit to Korea that all three memory makers have passed HBM4 qualification and are moving into mass production, and with The Elec’s note that SK hynix’s HBM4 is slated for NVIDIA’s next-generation Vera Rubin platform. The order is a dated, contractual data point underneath those headlines: the HBM3E-to-HBM4 transition, which the market spent months worrying might slip, is being capitalized on schedule.

Second, it settles a supplier-politics question. In December 2025, SK hynix reportedly placed an HBM4 TC bonder order with ASMPT while Hanmi was locked in a patent clash with Hanwha Semitech — raising the question of whether Hanmi’s dominant position (Korean reports pegged it at about 71% global TC bonder share in 2025) was starting to erode at exactly the wrong generation. The Griffin order answers that: Hanmi is inside the HBM4 ramp, not displaced from it. Dual-sourcing is real, but the incumbent kept the anchor slot.

The TC-versus-hybrid-bonding subtext

HBM4 was once penciled in by parts of the industry as the generation where hybrid bonding — the copper-to-copper direct bonding that BESI and others are building toward — would start displacing thermo-compression. Orders like this say the takeover hasn’t happened. SK hynix’s HBM4 stacking still runs on TC bonding paired with its advanced MR-MUF underfill flow, and Hanmi launched the TC Bonder 4 line for HBM4 back in mid-2025, upgrading it since.

That doesn’t kill the hybrid-bonding story — memory makers continue qualification work for later, taller stacks where TC bonding’s physics get harder. But it does mean the incumbent toolchain just bought itself another full generation of volume. For investors mentally modeling a fast TC-to-hybrid switchover, the Griffin order is evidence the cash flows from the old architecture extend further than the narrative does.

The skeptical read

This is filing-driven reporting, with estimates layered on. The ~15-unit count is TrendForce’s inference from typical pricing, not a disclosed figure. Even the model naming differs between outlets — TrendForce describes the TC Bonder 4 family, The Elec specifies the 4.5 Griffin.
44.2 billion won is not a full ramp. SK Group’s chairman has talked about doubling wafer capacity within five years, and TrendForce notes a path from roughly 550,000 to about 1 million DRAM wafer starts per month by 2030. Against that, 15 bonders is an opening tranche; follow-on orders are the confirmation to watch.
Diversification pressure is real. The ASMPT order exists, Hanwha keeps pushing, and the patent fight is unresolved. Hanmi can win HBM4 anchor orders and still cede share at the margin.

Practitioner note

If you track the AI-memory buildout, treat bonder contracts as a capacity clock: each disclosure carries a value, a tool generation, and — most usefully — a completion deadline. A September 2 deadline implies qualified HBM4 output exiting Cheongju in Q4 2026. Keep a running tally across Hanmi, ASMPT, and Hanwha of customer, generation, and deadline, and watch for the Samsung and Micron counterpart orders that this qualification milestone implies. When the deadlines start landing inside the current quarter rather than the next one, the ramp has gone from plan to throughput.

Under-considered angle

The quiet driver here is litigation as procurement strategy. SK hynix dual-sourcing HBM4 bonders across Hanmi and ASMPT is not only about price or capacity — an unresolved Hanmi–Hanwha patent war makes single-sourcing a legal risk, so qualifying a second toolchain is insurance. Equipment qualification, normally a yield decision, becomes risk management. The second-order effect: patent fights between toolmakers can create market share for third parties that would never win head-to-head on specs, which is worth remembering every time a bonder lawsuit headline looks like noise.