Trend AnalysisOther Engineering
High-NA EUV Lithography: Patterning the Sub-3nm Future of Semiconductors
High-NA EUV lithography (0.55 numerical aperture) is the semiconductor industry's path to sub-3nm technology nodes. Recent work addresses the key challenges: shrinking process windows, thinner resists, anamorphic optics, and mask fabrication precision.
By Sean K.S. Shin
This blog summarizes research trends based on published paper abstracts. Specific numbers or findings may contain inaccuracies. For scholarly rigor, always consult the original papers cited in each post.
The semiconductor industry's relentless drive toward smaller transistorsโfrom 7nm to 5nm to 3nm and beyondโdepends on lithography: the ability to project patterns onto silicon wafers with nanometer precision. Extreme Ultraviolet (EUV) lithography at 13.5nm wavelength replaced deep UV (193nm) for leading-edge patterning. Now, the transition from standard EUV (0.33 numerical aperture) to high-NA EUV (0.55 NA) promises to extend single-patterning capability to sub-3nm nodes, potentially avoiding the costly double-patterning that would otherwise be required.
But high-NA EUV brings new engineering challenges. The higher numerical aperture improves resolution but reduces depth of focus. The optics are anamorphic (different magnification in X and Y), requiring new mask designs. And the thinner photoresists needed for high-NA have less tolerance for defects. The four papers reviewed here address these challenges from different angles.
The Research Landscape
Source-Mask Optimization
Li, Liu, and Li (2025) investigate Source-Mask Optimization (SMO) for high-NA EUV. SMO jointly optimizes the illumination source shape and mask pattern to maximize the process windowโthe range of exposure dose and focus at which the printed pattern meets specifications. As features shrink below 10nm, process windows narrow dramatically, making SMO essential rather than optional.
Their results show that pixelated illumination (custom-shaped light sources tailored to each mask pattern) can expand the process window by 20-40% compared to conventional illuminationโpotentially the difference between a manufacturable and non-manufacturable process at sub-3nm nodes.
Pattern Fidelity Monitoring
Khan and Chew (2025), with 1 citation, address a different challenge: how to monitor whether the printed patterns match the intended design. At sub-3nm dimensions, pattern fidelity is measured in angstromsโa regime where traditional optical inspection cannot distinguish defective from good patterns.
Their novel 2D monitoring approach uses advanced scatterometry (measuring how light diffracts from the patterned surface) to assess pattern quality without requiring expensive scanning electron microscopy (SEM) for every wafer. The approach enables in-line monitoring during manufacturingโdetecting problems in real time rather than discovering them during post-production inspection.
Random Via Patterning
Chowrira and Dusa (2025) demonstrate that high-NA EUV can pattern random vias (vertical connections between metal layers) at sub-10nm dimensionsโa critical capability for next-generation logic chips where via density is a key determinant of routing flexibility and chip performance.
The challenge with vias is that they are random, two-dimensional patterns (unlike the regular, one-dimensional line-space patterns that most lithography research focuses on). Random patterns are harder to print reliably because they present more diverse shapes to the optical system, stress the depth of focus, and are more sensitive to mask defects.
Mask Fabrication
Cho, Noh, and Park (2024), with 2 citations, address the upstream challenge: fabricating the masks that high-NA EUV lithography projects onto wafers. The Multi-Beam Mask Writer (MBMW) has become the standard tool for EUV mask patterning, offering write time independent of pattern complexity. But high-NA masks have tighter specifications: smaller features, lower defect tolerance, and the anamorphic magnification (4ร in one direction, 8ร in the other) that requires asymmetric mask designs.
Critical Analysis: Claims and Evidence
<
| Claim | Evidence | Verdict |
|---|
| SMO expands high-NA EUV process windows by 20-40% | Li et al.'s simulation studies | โ
Supported โ in simulation |
| 2D scatterometry enables in-line pattern fidelity monitoring | Khan et al.'s experimental validation | โ
Supported โ demonstrated on test wafers |
| High-NA EUV can pattern random vias at sub-10nm | Chowrira et al.'s patterning experiments | โ
Supported |
| MBMW technology meets high-NA mask requirements | Cho et al.'s mask fabrication analysis | โ
Supported โ with tighter process control needed |
Open Questions
Cost: High-NA EUV scanners cost over $350 million each. At what production volume does single-patterning high-NA become cheaper than double-patterning low-NA?
Resist performance: Thinner resists for high-NA have lower etch resistance. Can new resist chemistries (metal oxide resists) solve this?
Stochastic effects: At sub-10nm dimensions, the random arrival of individual photons causes measurable pattern variability. Can stochastic effects be managed through higher dose, or do they set a fundamental resolution limit?
Beyond high-NA: If Moore's law continues, what comes after high-NA EUV? Proposals include multi-beam direct write, self-aligned patterning, and novel architectures (3D stacking, chiplets).What This Means for Your Research
For semiconductor engineers, high-NA EUV is the near-term path to sub-3nm nodes. For chiplet and packaging researchers, the economic constraints of EUV are driving interest in alternative approaches for less-critical layers.
Explore related work through ORAA ResearchBrain.
๋ฉด์ฑ
์กฐํญ: ์ด ๊ฒ์๋ฌผ์ ์ ๋ณด ์ ๊ณต ๋ชฉ์ ์ ์ฐ๊ตฌ ๋ํฅ ๊ฐ์์ด๋ค. ํ์ ์ฐ๊ตฌ์์ ์ธ์ฉํ๊ธฐ ์ ์ ๊ตฌ์ฒด์ ์ธ ์ฐ๊ตฌ ๊ฒฐ๊ณผ, ํต๊ณ ๋ฐ ์ฃผ์ฅ์ ์๋ณธ ๋
ผ๋ฌธ๊ณผ ๋์กฐํ์ฌ ๊ฒ์ฆํด์ผ ํ๋ค.
๊ณ ๊ฐ๊ตฌ์(High-NA) EUV ๋ฆฌ์๊ทธ๋ํผ: 3nm ์ดํ ๋ฐ๋์ฒด์ ๋ฏธ๋๋ฅผ ํจํฐ๋ํ๋ค
๋ฐ๋์ฒด ์ฐ์
์ด 7nm, 5nm, 3nm ์ดํ๋ก ํธ๋์ง์คํฐ๋ฅผ ๋์ฑ ์ํํํ๋ ค๋ ๋์์๋ ๋
ธ๋ ฅ์ ๋ฆฌ์๊ทธ๋ํผ, ์ฆ ๋๋
ธ๋ฏธํฐ ์ ๋ฐ๋๋ก ์ค๋ฆฌ์ฝ ์จ์ดํผ์ ํจํด์ ํฌ์ํ๋ ๋ฅ๋ ฅ์ ๋ฌ๋ ค ์๋ค. 13.5nm ํ์ฅ์ ๊ทน์์ธ์ (EUV) ๋ฆฌ์๊ทธ๋ํผ๋ ์ต์ฒจ๋จ ํจํฐ๋์์ ์ฌ์์ธ์ (DUV, 193nm)์ ๋์ฒดํ์๋ค. ์ด์ ํ์ค EUV(๊ฐ๊ตฌ์ 0.33 NA)์์ ๊ณ ๊ฐ๊ตฌ์ EUV(0.55 NA)๋ก์ ์ ํ์ ๋จ์ผ ํจํฐ๋(single-patterning) ๋ฅ๋ ฅ์ 3nm ์ดํ ๋
ธ๋๊น์ง ํ์ฅํ ๊ฒ์ผ๋ก ๊ธฐ๋๋๋ฉฐ, ๊ทธ๋ ์ง ์์ ๊ฒฝ์ฐ ํ์ํ์ ๊ณ ๋น์ฉ์ ์ด์ค ํจํฐ๋(double-patterning)์ ํผํ ์ ์์ ๊ฒ์ผ๋ก ์ ๋ง๋๋ค.
๊ทธ๋ฌ๋ ๊ณ ๊ฐ๊ตฌ์ EUV๋ ์๋ก์ด ๊ณตํ์ ๊ณผ์ ๋ฅผ ์๋ฐํ๋ค. ๋ ๋์ ๊ฐ๊ตฌ์๋ ํด์๋๋ฅผ ํฅ์์ํค์ง๋ง ์ด์ ์ฌ๋(depth of focus)๋ฅผ ๊ฐ์์ํจ๋ค. ๊ดํ๊ณ๋ ๋น๋์นญ(anamorphic) ๊ตฌ์กฐ๋ก, X์ถ๊ณผ Y์ถ์ ๋ฐฐ์จ์ด ์๋ก ๋ฌ๋ผ ์๋ก์ด ๋ง์คํฌ ์ค๊ณ๊ฐ ํ์ํ๋ค. ๋ํ ๊ณ ๊ฐ๊ตฌ์์ ์๊ตฌ๋๋ ๋ ์์ ํฌํ ๋ ์ง์คํธ๋ ๊ฒฐํจ์ ๋ํ ํ์ฉ ๋ฒ์๊ฐ ๋ ์ข๋ค. ์ฌ๊ธฐ์ ๊ฒํ ํ๋ ๋ค ํธ์ ๋
ผ๋ฌธ์ ์ด๋ฌํ ๊ณผ์ ๋ค์ ๊ฐ๊ธฐ ๋ค๋ฅธ ๊ด์ ์์ ๋ค๋ฃจ๊ณ ์๋ค.
์ฐ๊ตฌ ๋ํฅ
๊ด์-๋ง์คํฌ ์ต์ ํ(Source-Mask Optimization)
Li, Liu, Li(2025)๋ ๊ณ ๊ฐ๊ตฌ์ EUV๋ฅผ ์ํ ๊ด์-๋ง์คํฌ ์ต์ ํ(SMO)๋ฅผ ์ฐ๊ตฌํ๋ค. SMO๋ ์กฐ๋ช
๊ด์์ ํํ์ ๋ง์คํฌ ํจํด์ ๋์์ ์ต์ ํํ์ฌ ๊ณต์ ์๋์ฐ(process window), ์ฆ ์ธ์๋ ํจํด์ด ์ฌ์์ ์ถฉ์กฑํ๋ ๋
ธ๊ด ์กฐ๋ ๋ฐ ์ด์ ๋ฒ์๋ฅผ ์ต๋ํํ๋ค. ํผ์ฒ ํฌ๊ธฐ๊ฐ 10nm ์ดํ๋ก ์ถ์๋จ์ ๋ฐ๋ผ ๊ณต์ ์๋์ฐ๊ฐ ๊ธ๊ฒฉํ ์ข์์ง๋ฏ๋ก, SMO๋ ์ ํ ์ฌํญ์ด ์๋ ํ์ ์์๊ฐ ๋๋ค.
์ฐ๊ตฌ ๊ฒฐ๊ณผ, ํฝ์
ํ๋ ์กฐ๋ช
(๊ฐ ๋ง์คํฌ ํจํด์ ๋ง์ถคํ๋ ๊ด์ ํํ)์ ๊ธฐ์กด ์กฐ๋ช
๋๋น ๊ณต์ ์๋์ฐ๋ฅผ 20~40% ํ์ฅํ ์ ์๋ ๊ฒ์ผ๋ก ๋ํ๋ฌ๋ค. ์ด๋ 3nm ์ดํ ๋
ธ๋์์ ์ ์กฐ ๊ฐ๋ฅํ ๊ณต์ ๊ณผ ๊ทธ๋ ์ง ์์ ๊ณต์ ์ ๊ฐ๋ฅด๋ ๊ฒฐ์ ์ ์ธ ์ฐจ์ด๊ฐ ๋ ์ ์๋ค.
ํจํด ์ถฉ์ค๋ ๋ชจ๋ํฐ๋ง(Pattern Fidelity Monitoring)
ํผ์ธ์ฉ ํ์ 1ํ์ Khan๊ณผ Chew(2025)๋ ๋ ๋ค๋ฅธ ๊ณผ์ ๋ฅผ ๋ค๋ฃฌ๋ค. ๋ฐ๋ก ์ธ์๋ ํจํด์ด ์๋ํ ์ค๊ณ์ ์ผ์นํ๋์ง ๋ชจ๋ํฐ๋งํ๋ ๋ฐฉ๋ฒ์ด๋ค. 3nm ์ดํ ์์ค์์ ํจํด ์ถฉ์ค๋๋ ์น์คํธ๋กฌ(angstrom) ๋จ์๋ก ์ธก์ ๋๋ฉฐ, ์ด ์์ญ์์๋ ์ ํต์ ์ธ ๊ดํ ๊ฒ์ฌ๋ก ๊ฒฐํจ ํจํด๊ณผ ์ ์ ํจํด์ ๊ตฌ๋ณํ ์ ์๋ค.
์ด๋ค์ด ์ ์ํ ์๋ก์ด 2D ๋ชจ๋ํฐ๋ง ์ ๊ทผ๋ฒ์ ๊ณ ๊ธ ์ฐ๋๊ณ์ธก๋ฒ(scatterometry, ํจํฐ๋๋ ํ๋ฉด์์ ๋น์ด ํ์ ๋๋ ๋ฐฉ์์ ์ธก์ )์ ํ์ฉํ์ฌ, ๋ชจ๋ ์จ์ดํผ์ ๊ณ ๊ฐ์ ์ฃผ์ฌ์ ์ํ๋ฏธ๊ฒฝ(SEM)์ ์ฌ์ฉํ์ง ์๊ณ ๋ ํจํด ํ์ง์ ํ๊ฐํ๋ค. ์ด ์ ๊ทผ๋ฒ์ ์ ์กฐ ๊ณต์ ์ค ์ธ๋ผ์ธ(in-line) ๋ชจ๋ํฐ๋ง์ ๊ฐ๋ฅํ๊ฒ ํ์ฌ, ์์ฐ ํ ๊ฒ์ฌ ๋จ๊ณ์์ ๋ฐ๊ฒฌํ๋ ๊ฒ์ด ์๋๋ผ ์ค์๊ฐ์ผ๋ก ๋ฌธ์ ๋ฅผ ๊ฐ์งํ ์ ์๋ค.
๋๋ค ๋น์(Random Via) ํจํฐ๋
Chowrira์ Dusa(2025)๋ ๊ณ ๊ฐ๊ตฌ์ EUV๊ฐ 10nm ์ดํ ํฌ๊ธฐ์ ๋๋ค ๋น์(๊ธ์ ์ธต ๊ฐ ์์ง ์ฐ๊ฒฐ๋ถ)๋ฅผ ํจํฐ๋ํ ์ ์์์ ์ค์ฆํ๋ค. ์ด๋ ์ฐจ์ธ๋ ๋ก์ง ์นฉ์์ ๋น์ ๋ฐ๋๊ฐ ๋ฐฐ์ ์ ์ฐ์ฑ๊ณผ ์นฉ ์ฑ๋ฅ์ ๊ฒฐ์ ํ๋ ํต์ฌ ์์ธ์ธ ๋งํผ ๋งค์ฐ ์ค์ํ ๋ฅ๋ ฅ์ด๋ค.
๋น์ ํจํฐ๋์ ๊ณผ์ ๋ ๋น์๊ฐ ๋๋คํ 2์ฐจ์ ํจํด์ด๋ผ๋ ์ ์ด๋ค. ์ด๋ ๋๋ถ๋ถ์ ๋ฆฌ์๊ทธ๋ํผ ์ฐ๊ตฌ๊ฐ ์ง์คํ๋ ๊ท์น์ ์ธ 1์ฐจ์ ์ -๊ณต๊ฐ(line-space) ํจํด๊ณผ ๋ค๋ฅด๋ค. ๋๋ค ํจํด์ ๊ดํ๊ณ์ ๋์ฑ ๋ค์ํ ํํ๋ฅผ ์ ์ํ๊ณ , ์ด์ ์ฌ๋์ ๋ถ๋ด์ ์ฃผ๋ฉฐ, ๋ง์คํฌ ๊ฒฐํจ์ ๋ ๋ฏผ๊ฐํ๊ธฐ ๋๋ฌธ์ ์์ ์ ์ผ๋ก ์ธ์ํ๊ธฐ๊ฐ ๋ ์ด๋ ต๋ค.
๋ง์คํฌ ์ ์กฐ(Mask Fabrication)
ํผ์ธ์ฉ ํ์ 2ํ์ Cho, Noh, Park(2024)์ ์๋ฅ(upstream) ๊ณผ์ ์ธ ๊ณ ๊ฐ๊ตฌ์ EUV ๋ฆฌ์๊ทธ๋ํผ๊ฐ ์จ์ดํผ์ ํฌ์ํ๋ ๋ง์คํฌ ์ ์กฐ๋ฅผ ๋ค๋ฃฌ๋ค. ๋ค์ค ๋น ๋ง์คํฌ ๋ผ์ดํฐ(MBMW, Multi-Beam Mask Writer)๋ ํจํด ๋ณต์ก๋์ ๋ฌด๊ดํ ๊ธฐ๋ก ์๊ฐ์ ์ ๊ณตํ์ฌ EUV ๋ง์คํฌ ํจํฐ๋์ ํ์ค ์ฅ๋น๋ก ์๋ฆฌ์ก์๋ค. ๊ทธ๋ฌ๋ ๊ณ ๊ฐ๊ตฌ์ ๋ง์คํฌ๋ ๋ ์๊ฒฉํ ์ฌ์์ ์๊ตฌํ๋ค. ๋ ๋ฏธ์ธํ ํผ์ฒ, ๋ ๋ฎ์ ๊ฒฐํจ ํ์ฉ ๋ฒ์, ๊ทธ๋ฆฌ๊ณ ๋น๋์นญ ๋ง์คํฌ ์ค๊ณ๋ฅผ ํ์๋ก ํ๋ ๋น๋์นญ ๋ฐฐ์จ(ํ ๋ฐฉํฅ์ผ๋ก 4๋ฐฐ, ๋ค๋ฅธ ๋ฐฉํฅ์ผ๋ก 8๋ฐฐ)์ด ๊ทธ๊ฒ์ด๋ค.
๋นํ์ ๋ถ์: ์ฃผ์ฅ๊ณผ ๊ทผ๊ฑฐ
<
| ์ฃผ์ฅ | ๊ทผ๊ฑฐ | ํ๊ฐ |
|---|
| SMO expands high-NA EUV process windows by 20-40% | Li et al.'s simulation studies | โ
Supported โ in simulation |
| 2D scatterometry enables in-line pattern fidelity monitoring | Khan et al.'s experimental validation | โ
Supported โ demonstrated on test wafers |
| High-NA EUV can pattern random vias at sub-10nm | Chowrira et al.'s patterning experiments | โ
Supported |
| MBMW technology meets high-NA mask requirements | Cho et al.'s mask fabrication analysis | โ
Supported โ with tighter process control needed |
Open Questions
Cost: High-NA EUV scanners cost over $350 million each. At what production volume does single-patterning high-NA become cheaper than double-patterning low-NA?
Resist performance: Thinner resists for high-NA have lower etch resistance. Can new resist chemistries (metal oxide resists) solve this?
Stochastic effects: At sub-10nm dimensions, the random arrival of individual photons causes measurable pattern variability. Can stochastic effects be managed through higher dose, or do they set a fundamental resolution limit?
Beyond high-NA: If Moore's law continues, what comes after high-NA EUV? Proposals include multi-beam direct write, self-aligned patterning, and novel architectures (3D stacking, chiplets).What This Means for Your Research
For semiconductor engineers, high-NA EUV is the near-term path to sub-3nm nodes. For chiplet and packaging researchers, the economic constraints of EUV are driving interest in alternative approaches for less-critical layers.
Explore related work through ORAA ResearchBrain.
References (4)
[1] Li, K., Liu, X., & Li, Y. (2025). Study of SMO for high-NA EUV lithography via Sub-3nm nodes. Proc. SPIE.
[2] Chowrira, B., Blanco Carballo, V.M., & Dusa, M. (2025). Moore's law meets high-NA EUV: random via patterning for next-generation nodes. Proc. SPIE.
[3] Khan, A.H., Danilevsky, A., & Chew, K. (2025). Novel 2D high-NA EUV pattern fidelity monitoring. Proc. SPIE.
[4] Cho, Y., Noh, I., & Park, J. (2024). Requirements and strategies of High-NA EUV mask using MBMW. Proc. SPIE.