About this document
An industry report that breaks the advantage held by Korean suppliers into three layers and asks what each rests on, then examines what a shift in the underlying principle of bond-line cooling would mean.
Download
An English edition is not available yet. The Korean document is below.
Summary
The basis of competition in back-end semiconductor manufacturing is moving from which equipment a company owns to how it manages cycle time and thermal history on the same equipment. In March 2026 a Korean equipment maker disclosed plans to supply 2.5D thermocompression bonders to foundries in China and Taiwan, and China's largest DRAM maker raised roughly 12 trillion won through its listing. Air-jet cooling has already reached its ceiling at an interfacial heat transfer coefficient of 500-1,000 W/m²·K, a limit set by the properties of the working fluid, and no supplier has yet built a position in the region beyond it. The report analyses why the timing of that transition decides who holds the initiative.
Conclusion
The advantage held by Korean suppliers breaks into three layers: access to equipment, process know-how, and the choice of underlying principle. Access to equipment was opened by the supply plans disclosed in March 2026. Process know-how transfers through repeated production and the movement of engineers. The advantage rests on two layers with a confirmed path to erosion, while the third layer, the one that does not erode, is still unclaimed. The interfacial heat transfer coefficient of air-jet cooling stands at 500-1,000 W/m²·K and does not improve with nozzle design or higher flow rates. Direct solid-contact conduction reaches 20,000-50,000 W/m²·K at the same interface. In Cools' first validation run, the cooling segment of a TCB cycle fell from 7.0 seconds to 3.4 seconds, a 51% reduction. The figure was measured over ten repeated cycles on an 11mm×11mm HBM3E die using a modified production TCB heater, with no change to the heater structure. Throughput and cumulative thermal exposure improve together because both are governed by the same cooling time constant. The question now is not whether conduction cooling will eventually be needed, but who sets the new productivity benchmark before back-end capability levels out.
References
- 1. 주간한국 (Weekly Hankook). 한미반도체, 세미콘 차이나서 첨단 반도체 장비 선봬, 2026.3.25.
- 2. 녹색경제신문 (Green Economy News). 한미반도체, 세미콘 차이나서 AI 패키징 신장비 3종 공개, 2026.3.25.
- 3. 서울경제 (Seoul Economic Daily). 한미반도체 中 공략, HBM 핵심장비 공급, 2025.6.23.
- 4. Counterpoint Research. Global DRAM revenue surge, Memory Tracker Q1 2026, 2026.5.
- 5. 아이뉴스24 (inews24). 한미반도체, 세미콘 동남아서 2.5D TC본더 공개, 2026.5.4.
- 6. 대한경제 (Daehan Economy). 한미반도체, 컴퓨텍스 2026 첫 참가, 차세대 TC 본더 선봬, 2026.6.4.
- 7. 전자신문 (The Electronic Times). 中 CXMT, 내년 12단 HBM 양산, 韓 턱밑 추격, 2026.4.8.
- 8. 이투데이 (Etoday). HBM 호황에 가려진 중국의 추격, D램 기술 격차 3년 수준, 2026.7.
- 9. 파이낸셜뉴스 (Financial News). D램 3년·HBM 5년 中 CXMT 아직 격차, 문제는 추격 속도, 2026.7.29.
- 10. 위클리포스트 (Weekly Post). CXMT, 중국산 DUV 장비 우선 공급, 2026.7.28.
- 11. 시사저널e (Sisajournal-e). CXMT, 韓엔지니어 200명 품었다, 2026.6.2.
- 12. ChangXin Memory Technologies. Shanghai Stock Exchange listing disclosure, 2026.7.27.
- 13. Cools. Internal validation data, measured TCB cooling segment, 2026.
Topics
- Thermal control
- HBM
