SK hynix placed a 44.2 billion won ($28.7 million) order on June 8 for Hanmi Semiconductor's newly developed TC Bonder 4.5 Griffin, marking the first publicly disclosed contract for that equipment model and confirming that the world's largest high-bandwidth memory maker has cleared Hanmi's HBM4-generation bonding platform. The order, disclosed in a regulatory filing with South Korea's Financial Supervisory Service, covers delivery, installation, and test operation at SK hynix's post-package and test facility in Cheongju through September 2.

The contract's technical significance goes beyond its price tag. The TC Bonder 4.5 Griffin was developed specifically for HBM4 manufacturing — the sixth-generation high-bandwidth memory that SK hynix is now producing in volume for Nvidia's Vera Rubin AI accelerator platform. Qualifying a new piece of bonding equipment for a new HBM generation is not automatic: the precision requirements for HBM4's 16-layer die stacks are substantially tighter than those for earlier generations, and SK hynix had been actively evaluating rival suppliers before placing this order. The filing signals that Hanmi's platform cleared those requirements.

SK hynix HBM4 Equipment Spending Accelerates With June Order

SK hynix had been cautious about new TC bonder investment throughout the second half of 2025 and into early 2026 — a restraint that had raised concern in the semiconductor equipment sector about whether HBM4 capacity spending would accelerate on schedule. The June 8 filing reflects a turning point. At roughly 3 billion won per unit, industry analysts estimate the contract covers approximately 15 bonder systems, all destined for the M15X back-end packaging and test facility in Cheongju, about 110 kilometers south of Seoul. The contract runs through September 2 and requires full delivery, installation, and inspection before that date.

The deal represents about 7.7% of Hanmi Semiconductor's 2025 consolidated revenue of approximately 576.6 billion won. By comparison, a 9.6 billion won TC bonder deal between the two companies — covering HBM3/3E equipment rather than HBM4 — was signed in January 2026. The June filing is the first to involve the TC Bonder 4.5 Griffin model specifically, making it the first HBM4-generation bonder contract disclosed between the two companies.

The backdrop includes a changed competitive environment that the order alone does not resolve. SK hynix had been Hanmi's sole HBM bonder customer from the time Hanmi delivered the industry's first TC bonder for HBM production in 2016 until 2024, when SK hynix began diversifying its supplier base. Singapore-based ASMPT received its first meaningful bonder order from SK hynix in 2024, and Hanwha Semitech — a Korean equipment maker that entered the market more recently — received an order for 12 TC bonders worth approximately 42 billion won in March 2025. That diversification triggered one of the more visible supplier-customer disputes in recent South Korean semiconductor history: Hanmi recalled roughly 60 service engineers it had stationed at SK hynix's Icheon campus in April 2025, widely understood as a protest against SK hynix's decision to source from Hanwha. Hanmi also raised prices on equipment for which it had held the rate steady for eight years.

How Does Thermo-Compression Bonding Stack HBM Dies?

HBM is manufactured by vertically stacking multiple DRAM dies — twelve or sixteen of them in HBM4 — atop a base logic die that handles memory control and the high-speed I/O interface. Each layer connects to the next through thousands of microscopic copper conductors called through-silicon vias, or TSVs. A TC bonder is the machine that physically bonds each die to the stack one layer at a time, pressing the die down with controlled heat and force to fuse copper micro-bumps on one die's surface to their corresponding pads on the die below.

Unlike mass reflow soldering, which heats an entire assembly at once in an oven, TC bonding applies pressure and a controlled temperature ramp directly to the die being placed — allowing sub-micron alignment accuracy that mass reflow cannot achieve at the fine pitches HBM requires. Between each pair of dies, a layer of non-conductive film, or NCF, is pre-applied to the die surface; the TCB process heats and compresses this film simultaneously with the micro-bump bonding step, displacing voids and encapsulating the interconnects in a single operation. The result is a hermetically reinforced stack with tightly controlled electrical and thermal characteristics.

HBM4 Raised Bonding Precision Past What Traditional TCB Handles

HBM4's 16-layer stacks compressed the microbump pitch — the center-to-center spacing between adjacent solder connections — toward 25 micrometers and below, a regime in which flux residue from conventional TCB processes creates voids and contamination that reduce both yield and long-term reliability. The HBM4 specification, published by JEDEC as JESD270-4 in April 2025, also relaxed the maximum allowable stack height from 720 micrometers to 775 micrometers. That 55-micrometer margin was consequential: it allowed 16-layer stacks to continue using micro-bump bonding rather than the more expensive and mechanically demanding copper-to-copper hybrid bonding that had previously been considered mandatory for HBM4.

The relaxed height standard directly shaped the equipment roadmap. Hanmi's Chairman Kwak Dong-shin stated publicly in July 2025 that TC bonders remain sufficient for both HBM4 and HBM5 given JEDEC's decision. A single hybrid bonder costs in excess of 10 billion won — more than three times the approximately 3 billion won price of a TC bonder — so the JEDEC height relaxation preserved the economics of the established TCB supply chain. Hanmi launched its TC Bonder 4 platform in May 2025 in direct response to those conditions. The TC Bonder 4.5 Griffin delivered to SK hynix is an upgraded variant of that platform, optimized for SK hynix's specific HBM4 stacking process, which uses an advanced mass-reflow molded underfill technique for the primary stacking operation and TC bonding for precision placement at finer pitches.

Hanmi is simultaneously developing a fluxless TC bonder — which eliminates flux entirely, relying instead on pre-applied films and direct copper-to-copper bonding chemistries — as part of its roadmap for future HBM generations. That product had not been released as of the June 8 filing. Hanmi is also developing a wide TC bonder intended to support the larger die sizes used in HBM5 and HBM6 stacks of 20 or more layers, with first deliveries planned for the second half of 2026. A hybrid bonder targeting HBM6 production is on the company's roadmap for commercial launch by the end of 2027.

Hanmi Revenue Context: First Griffin Order Follows Severe Q1 Decline

Hanmi Semiconductor's financial results for the first quarter of 2026 illustrated how severely the HBM4 transition had suppressed equipment spending. The company posted consolidated revenue of 50.9 billion won and operating profit of 8.45 billion won for the quarter — representing declines of 65.5% and 87.9%, respectively, compared to the same period a year earlier. Net profit fell 65.2% year over year to 19 billion won.

Sangsangin Securities analysts, citing the June 8 order, forecast that Hanmi's second-quarter 2026 revenue could recover sharply to approximately 227.6 billion won, with operating profit of roughly 110.3 billion won — a dramatic swing driven primarily by the new TC Bonder 4.5 Griffin contracts. The current order is smaller than a comparable SK hynix bonder contract from the prior year, which was valued at approximately 55.2 billion won, but analysts expect additional orders to follow as SK hynix scales its HBM4 output. Hanmi expects those figures to grow if additional contracts are placed as production at Cheongju expands.

TC Bonder Supplier Landscape: Hanmi Holds Ground as ASMPT and Hanwha Semitech Enter

The June 8 order does not restore the exclusive relationship that defined Hanmi and SK hynix's partnership for nearly a decade. ASMPT and Hanwha Semitech remain part of SK hynix's bonder supplier base, and the competitive pressure they represent is structural rather than temporary. ASMPT, in particular, entered HBM4 qualification with a platform that supports fluxless bonding applications — a capability that gives it a technical differentiation for fine-pitch applications at the aggressive end of HBM4's stacking requirements. Lumen Alpha analysts forecast in mid-2025 that Hanmi's share of SK hynix's TC bonder purchases could fall to 20–30% by 2026, down from near-exclusivity in earlier HBM generations.

The Hanmi–Hanwha Semitech patent dispute adds a separate layer of uncertainty to the supply chain. Hanmi filed a patent infringement lawsuit against Hanwha Semitech in December 2024, alleging that Hanwha's TC bonder technology infringed patents Hanmi holds on its bonding platform — which the company describes as the world's first dedicated TC bonder for HBM production, developed in 2016 and delivered in volume from 2017. Hanwha filed to invalidate those patents in May 2025 and launched a countersuit for patent infringement. An invalidation ruling was expected within approximately six months of filing. An industry analyst familiar with Korean patent proceedings noted that if patents were invalidated, the infringement lawsuit would likely be dismissed; if upheld, a manufacturing or sales ban on infringing equipment could ultimately be ordered, though such rulings typically take years to enforce. Equipment already delivered to SK hynix would likely be unaffected by any eventual ruling.

Nvidia Vera Rubin: Demand Signal Driving SK hynix Equipment Spend

The timing of the June 8 order fits a pattern of accelerating demand signals from Nvidia. Jensen Huang visited Seoul on June 5, three days before Hanmi's filing, and stated publicly that all three major memory makers — Samsung, SK hynix, and Micron — had passed HBM4 qualification tests and were actively supplying the sixth-generation chips for Vera Rubin. SK hynix is estimated to hold roughly 60–70% of Nvidia's initial HBM4 allocation for Vera Rubin systems.

Samsung became the first company to commercially ship HBM4 on February 12, 2026, announcing a data rate of 11.7 gigabits per second per pin and per-stack bandwidth of up to 3.3 terabytes per second — a 2.7-fold improvement over HBM3E. SK hynix entered its own large-scale HBM4 production shortly after, using its advanced mass-reflow molded underfill process as the primary stacking method. HBM4's 2,048-bit interface — double the 1,024-bit interface of HBM3E — is what makes the tighter microbump pitch and the more demanding TC bonding requirements structurally necessary: fitting twice as many signal channels into a package of comparable footprint requires proportionally finer inter-die interconnections.

SK hynix's longer-term capacity expansion reinforces why equipment orders are growing. The company aims to increase monthly DRAM wafer output from approximately 550,000 wafers to roughly 1 million by 2030, with the Cheongju M15X fab expected to begin operations in the second half of 2026, initially processing approximately 40,000 wafers per month and rising toward 80,000 by 2027. Each wafer dedicated to HBM consumes roughly three times the fab capacity of a standard DRAM wafer, because through-silicon via etching, wafer thinning, and multi-die stacking add nearly 20 additional process steps while reducing the memory bits that fit on each die.

Frequently Asked Questions

What is a TC bonder and why does HBM4 need one?

A thermo-compression bonder is the specialized machine that physically bonds DRAM dies together in a high-bandwidth memory stack, using precisely controlled heat and pressure to fuse copper micro-bumps between layers while simultaneously curing a non-conductive film that seals the interface. HBM4's 16-layer stacks require a new generation of TC bonder because the reduced microbump pitch — approaching 25 micrometers and below — exceeds the yield limits of older flux-based bonding processes, which leave residue that causes voids and reliability failures at those dimensions.

Why did SK hynix add ASMPT and Hanwha Semitech as TC bonder suppliers alongside Hanmi?

SK hynix began diversifying its TC bonder supply chain in 2024, adding Singapore-based ASMPT and subsequently Hanwha Semitech to reduce sole-source risk and introduce competitive pricing pressure after what the company viewed as an unsustainable period of sole-supplier dependency. ASMPT brought fluxless bonding capability relevant to HBM4's precision demands, while Hanwha entered as a domestic Korean alternative. Hanmi protested the diversification by recalling its field engineers from SK hynix's facilities and raising its equipment prices in April 2025.

What does the Hanmi–Hanwha Semitech patent dispute mean for SK hynix's HBM4 supply chain?

Hanmi filed a patent infringement suit against Hanwha Semitech in December 2024, and Hanwha countersued and sought to invalidate Hanmi's patents in May 2025. If a court ultimately upheld Hanmi's patents and ordered Hanwha to stop producing infringing equipment, SK hynix's ability to receive future bonders from Hanwha could be affected — though equipment already installed would likely remain in service. The litigation is expected to take years to conclude, and analysts note that SK hynix's three-supplier strategy reduces the risk that any single outcome would disrupt its HBM4 production lines.

How does HBM4 differ from HBM3E in ways that affect packaging equipment?

HBM4 doubles the memory bus width from 1,024 bits to 2,048 bits per stack, which means fitting twice as many signal channels into a comparable footprint, forcing the microbump pitch tighter. It also moves to a 12- or 16-layer die stack built on a 4-nanometer base logic die rather than the 5- to 8-nanometer base die used in HBM3E. Both the tighter pitch and the taller, thinner stacks place new demands on bonding equipment: sub-micron alignment accuracy, more precise co-planarity control, and the ability to handle wafers thinned to approximately 30 micrometers for the 16-layer configuration.