With approximately 90 percent of the global market for HBM thermal compression (TC) bonders — the back-end equipment that vertically stacks and bonds the individual DRAM dies inside every high-bandwidth memory chip powering AI data centers worldwide — Hanmi Semiconductor holds a position in the semiconductor supply chain that few equipment companies ever achieve: a near-monopoly on an indispensable process step, as TrendForce has reported. On Day 2 of SEMICON West 2026 in San Francisco, the South Korean equipment maker is showcasing the TC Bonder 4, its HBM4-rated production system now in full-scale customer supply, alongside five 2.5D heterogeneous packaging tools and a preview of the Wide TC Bonder targeting HBM5 and HBM6 production from 2027.

That showcase carries commercial weight. SK hynix signed a 44.2 billion won (approximately $31.8 million) contract for the TC Bonder 4.5 Griffin in June 2026, covering roughly 15 units at approximately 3 billion won (around $2.2 million) each — an upgraded variant of the TC Bonder 4. SK hynix's HBM4 output from those systems is broadly expected for Nvidia's Vera Rubin AI accelerator platform.

TC Bonder 4: HBM4 Production's Workhorse, Now Shipping Globally

The TC Bonder 4 launched in mid-2025 and has since entered full-scale supply to global memory manufacturers as SK hynix, Samsung Electronics, and Micron Technology accelerate their HBM4 capacity buildouts. The system is purpose-built for the thermal compression bonding step that joins individual DRAM dies into the vertical stacks that define HBM's architecture — a step that requires sub-micron alignment precision over as many as 16 dies per stack in HBM4, per the JEDEC HBM4 specification finalized in April 2025.

SEMICON West is at least the third major international trade event at which Hanmi has displayed the TC Bonder 4 this year, following appearances at Computex 2026 in Taipei in June and SEMICON Taiwan 2026 in September — a deliberate campaign to put the equipment in front of every major node in the global AI chip supply chain.

How Thermal Compression Bonding Works: The Process Inside Every HBM Stack

To understand why Hanmi's position matters — and why disrupting it would ripple across the AI industry — it helps to understand what a TC bonder actually does.

High bandwidth memory is built by vertically stacking individual DRAM dies, up to 16 in HBM4, and connecting each layer to the one below it via through-silicon vias (TSVs) — copper columns etched through silicon at diameters of 1 to 10 micrometers. Joining those dies into a reliable, thermally stable stack is the TC bonder's job.

In Hanmi's TC-NCF (thermal compression with non-conductive film) process, a bonding head picks up a single DRAM die and aligns it over the growing stack using optical vision alignment accurate to approximately ±1 micrometer. The head applies controlled heat — typically in the 200–400°C range — and pressure of roughly 5 to 500 newtons, causing the copper microbumps on each die (typically at 5 to 20 µm pitch) to compress and form permanent intermetallic bonds. Before bonding, each die is coated with non-conductive film (NCF), a thermoplastic material that flows under heat and pressure to fill the spaces between dies, blocking moisture and mechanically stabilizing the finished stack. The sequence repeats die by die until all 16 layers are bonded.

The critical advantage over mass reflow bonding — which heats an entire wafer or panel at once — is precision warp control. In HBM4's 16-die stacks, batch heating accumulates thermal stress that warps thin DRAM dies and causes alignment failures. Die-by-die TC bonding applies heat locally per die, keeping each layer stable during the bonding step. That mechanical precision is why TC bonding has dominated HBM manufacturing since HBM3, and why Hanmi, which has filed more than 120 HBM patents, has built its commanding position at this specific process step.

The concentration of global HBM4 production around a single company's bonding equipment is not merely a commercial success story. It represents a structural bottleneck: a disruption to Hanmi's production capacity, component supply, or export access would directly constrain the world's ability to scale HBM4 output for AI accelerators. That dependency is drawing geopolitical scrutiny, with reports from May 2025 indicating Hanmi may be limiting TC bonder exports to Chinese chipmakers under US trade policy pressure — a claim the company has not publicly confirmed, and one with significant implications for China's domestic HBM program.

Wide TC Bonder: Engineered for the Die Sizes HBM5 and HBM6 Will Require

Alongside the TC Bonder 4, Hanmi is previewing the Wide TC Bonder at SEMICON West — a next-generation system designed for the larger DRAM die footprints expected in HBM5 and HBM6, where capacity and bandwidth targets will require physically larger dies than HBM4 uses today. The equipment is slated for customer supply starting in 2027.

A wider die footprint enables more TSVs, more input/output interfaces, and ultimately higher memory capacity and bandwidth per stack—the performance attributes AI accelerator designers will need for model inference and training workloads beyond the current generation. The JEDEC HBM4 specification, finalized in April 2025, relaxed the package stack height standard to 775 micrometers — a change confirmed in the JEDEC HBM4 standard that allowed Hanmi's existing TC Bonder 4 to produce compliant HBM4 without requiring the far more expensive shift to hybrid bonding for this generation. The Wide TC Bonder carries that TC bonding approach forward into the larger geometries that follow.

Hybrid Bonding: Hanmi's $71.9 Million Bet on the HBM6 Era

The eventual successor to TC bonding — at least for the most advanced HBM generations — is hybrid bonding, which achieves direct copper-to-copper bonding without solder microbumps and enables interconnect pitches below 1 micrometer. That sub-micron pitch is required for the 20-plus die stacks targeted in HBM6, where TC bonding's microbump-based approach reaches its fundamental pitch limits.

Hanmi is investing 100 billion won (approximately $71.9 million) in hybrid bonder R&D for HBM6 and has set an end-of-2027 target for a commercial system. Hybrid bonders cost more than 10 billion won per unit (approximately $7.2 million) — roughly three times the cost of a TC bonder, as industry sources confirm — reflecting their greater engineering complexity.

Competition for the hybrid bonder market is forming. LG Electronics unveiled hybrid bonder development at Nano Korea 2025, with commercial availability expected around 2028. Hanwha Semitec entered the bonding equipment market in March 2025. Neither has demonstrated mass-production-qualified equipment as of SEMICON West 2026. Chairman Kwak Dong-shin has stated publicly that TC bonders are sufficient through HBM5, with hybrid bonding necessary only when HBM6 demands it.

Five 2.5D Packaging Tools for the AI Chip Stack

Beyond HBM stacking, Hanmi is displaying five 2.5D packaging tools targeting the heterogeneous integration layer — the architecture in which GPU compute dies, CPU chiplets, and HBM memory stacks are assembled side by side on a silicon interposer in a single package. In this 2.5D configuration, dies communicate across the interposer's fine-pitch redistribution wiring, enabling the extremely short interconnect distances and high data rates that AI chip performance requires.

The five systems at SEMICON West, as confirmed at SEMICON Taiwan 2026, are:

FC Bonder 3.5 — a flip-chip bonder for system semiconductor integration, already in active supply to global foundries and OSATs.

FC Bonder 75 — the larger-capacity companion to the FC Bonder 3.5, also shipping to customers.

2.5D TC Bonder 40 C2S — chip-on-substrate variant for CoWoS-type packaging architectures, now in customer supply.

2.5D TC Bonder 40 C2W — chip-on-wafer variant targeting the same CoWoS-class processes, also now shipping.

2.5D TC Bonder 120 — the high-throughput model in the 2.5D lineup, set to launch imminently.

TSMC's CoWoS platform, the dominant commercial 2.5D integration technology, has faced documented capacity constraints that have restricted AI chip supply. Equipment that enables comparable chip-on-wafer and chip-on-substrate bonding at OSATs and chipmakers' in-house packaging facilities represents a strategic diversification option for companies trying to reduce their dependence on a single advanced packaging supplier.

Micro Saw and Vision Placement: MSVP Delivers Panel-Level Packaging Revenue

Completing the SEMICON West exhibit is Hanmi's Micro Saw and Vision Placement (MSVP) system, a combined dicing and precision die-placement tool. The company noted in its Q2 2026 earnings release that MSVP equipment contributed meaningfully to earnings as panel-level packaging adoption broadened among AI chip manufacturers.

Record Revenue, Eighth Factory, and a Beachhead in California

The SEMICON West showcase arrives at a moment of exceptional financial strength. Hanmi posted record Q2 2026 financial results of 251.1 billion won (approximately $180.6 million) in consolidated sales and 130.3 billion won (approximately $93.7 million) in operating profit — a 51.9 percent operating margin — driven by surging TC bonder shipments as major customers entered full-scale HBM4 production. Full-year 2025 revenue reached 576.7 billion won (approximately $414.9 million) — Hanmi's highest annual sales since its 1980 founding — with operating profit of 251.4 billion won (approximately $180.9 million).

To keep pace with demand, Hanmi announced in August 2026 a 130 billion won (approximately $93.5 million) eighth factory in Incheon — its current seven are all there — specifically oriented toward HBM bonder manufacturing. Chairman Kwak Dong-shin bought 5 billion won personally in company shares (approximately $3.6 million) in July 2026, a public signal of his confidence in the company's growth trajectory.

North American Push, Anchored in the CHIPS Act Buildout

SEMICON West is central to Hanmi's North American market development strategy. The company plans to establish HANMI USA, a US subsidiary in San Jose by the end of 2026, to provide local technical support, customer engagement, and equipment servicing for semiconductor manufacturers expanding AI packaging capacity in the United States.

The timing is deliberate. The CHIPS and Science Act's $52.7 billion in direct semiconductor funding has catalyzed more than $645 billion in total US semiconductor investment across 140 projects in 30 states as of mid-2026. Major grant recipients include Intel ($7.86 billion), TSMC ($6.6 billion), Samsung ($6.4 billion), and Micron ($6.1 billion). Micron is expanding HBM packaging capacity in Singapore, Idaho, and New York; Intel is investing in domestic advanced packaging. That buildout creates a North American customer base for Hanmi's tools that HANMI USA is designed to service locally.

"As the mass production of HBM4 gains momentum this year, we are seeing a concentration of TC Bonder orders in the second quarter, a trend expected to accelerate in the latter half of the year," chairman Kwak Dong-shin told reporters at launch. "HANMI Semiconductor's TC Bonders, which hold the leading global market share, will see even greater benefits as the AI semiconductor market expands."

The TC and hybrid bonder market for HBM production is projected to be among the fastest-growing segments in back-end equipment, with Yole Group projecting $1.3 billion expansion in TC bonding and hybrid bonding by 2030. SEMICON West's Advanced Packaging and Heterogeneous Integration Summit track — a dedicated programming thread at the 2026 show — reflects exactly the urgency of that forecast: advanced packaging has moved from afterthought to performance differentiator for AI systems, and the equipment makers that own the critical process steps in that packaging are gaining strategic importance that their financial results are beginning to reflect.

Currency conversions are approximate based on exchange rates at time of publication.

Frequently Asked Questions

What is a TC bonder and why is it critical for HBM4 production?

A thermal compression (TC) bonder is a semiconductor back-end tool that joins individual DRAM dies into the vertical stacks that define high-bandwidth memory. HBM4 stacks up to 16 individual DRAM dies, and each die must align within approximately one micrometer of the die below it. The TC bonder achieves that precision die by die, applying controlled heat and pressure while a non-conductive film (NCF) encapsulant stabilizes each bonded layer. No other production-proven method handles the combination of stack height, alignment precision, and warpage control that HBM4 demands at manufacturing scale — which is why the company that makes the dominant TC bonder is the de facto gatekeeper for every HBM4 chip currently in production.

How does TC bonding differ from hybrid bonding, and when will the industry transition?

TC bonding uses copper microbumps—typically at a 5- to 20-micrometer pitch—to join dies in a stack. Hybrid bonding eliminates those bumps entirely, achieving direct copper-to-copper bonds at sub-micrometer pitch, which is required for 20-plus die stacks. Hybrid bonders cost more than 10 billion won (approximately $7.2 million) per unit, roughly three times the cost of a TC bonder, and are not yet qualified for mass production in HBM. The current industry consensus — supported by Hanmi, Yole Group, and TrendForce — is that TC bonding remains sufficient for HBM4 and HBM5, and that hybrid bonding becomes mandatory when HBM6 production begins, targeted by the end of 2027.

What does Hanmi's near-monopoly mean for the broader AI chip supply chain?

With approximately 90 percent of global HBM TC bonder production concentrated at a single supplier, the HBM manufacturing supply chain carries meaningful single-source risk at this specific process step. If Hanmi's production capacity, component sourcing, or export access were disrupted — whether by supply chain failure or geopolitical restriction — the ability to scale HBM4 production for AI data centers would be directly constrained. Reports from May 2025 indicated that Hanmi may be facing US pressure to limit TC bonder exports to Chinese chipmakers, which would affect China's domestic HBM program while leaving Hanmi's US, Korean, and Taiwanese customers unaffected. The company has not officially confirmed or denied those reports.

What should engineers and equipment buyers know about Hanmi's US expansion?

Hanmi plans to open HANMI USA in San Jose, California by the end of 2026. The subsidiary will provide local technical support, customer engagement, and equipment servicing for semiconductor companies expanding US manufacturing capacity under CHIPS Act funding. For engineers at US fabs building out HBM or advanced 2.5D packaging lines, HANMI USA means local service access for the TC bonder and 2.5D tools they will likely depend on — and a direct relationship with the company that controls the dominant bonding platform for AI chip production into the next decade.

SEMICON West 2026 runs through October 15 at the Moscone Center, San Francisco.