Trend AnalysisEngineering

High-NA EUV Lithography: The Stitching Problem at 0.55 NA

ASML's EXE:5000 scanner pushes EUV to 0.55 NAโ€”enabling sub-8nm features but halving the exposure field. For large dies, two images must be 'stitched' with sub-nanometer overlay accuracy. This engineering constraint may reshape chip architecture as much as the resolution gain itself.

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.

ASML's EXE:5000 scannerโ€”the first high-NA (0.55 numerical aperture) EUV lithography systemโ€”began operation in 2024, succeeding the 0.33 NA TWINSCAN NXE series that has driven semiconductor manufacturing to the 3nm node. The 67% increase in numerical aperture enables printing features below 8nm half-pitch, extending Moore's Law scaling by at least two technology nodes. This is, by the standards of semiconductor engineering, a major advance.

It comes with a trade-off that receives less attention outside the lithography community but may prove equally consequential: the anamorphic optics that enable 0.55 NA reduce the exposure field from 26ร—33 mmยฒ to 26ร—16.5 mmยฒโ€”effectively halving the printable area per exposure. For any die larger than this reduced field, the scanner must expose two (or more) overlapping images that are "stitched" together on the wafer with sub-nanometer precision. This stitching requirement introduces new sources of error, new design constraints, and potentially new chip architectures.

The Stitching Challenge: Sub-Nanometer Overlay

Wiaux et al. (2025) provide the first detailed experimental characterization of in-die stitching for high-NA EUV, published in Journal of Micro/Nanopatterning, Materials, and Metrology with . Their study, conducted at imec using early EXE:5000 exposures, quantifies the overlay accuracy at the stitch boundaryโ€”the line where two separately exposed images meet.

The key findings:

  • Overlay accuracy at the stitch: Mean overlay of 1.2 nm, with 3ฯƒ variation of ยฑ0.8 nm. For context, the minimum feature size targeted by high-NA EUV is ~8 nm. An overlay error of 2 nm (3ฯƒ envelope) represents 25% of the feature sizeโ€”tight but within the tolerance required for metal interconnect layers.
  • Stitching artifacts: At the stitch boundary, dose variations from the overlap region produce ~5% CD (critical dimension) variation. This can be mitigated by optimizing the overlap width and dose ramping profile, but cannot be eliminated entirely.
  • Design rule impact: Metal lines crossing the stitch boundary require minimum spacing rules 15โ€“20% wider than lines within a single field, creating design asymmetries that complicate standard-cell layout.
Wiaux et al. note that stitching is not newโ€”it was used in older lithography generations for exceptionally large dies. What is new is that high-NA EUV makes stitching necessary for die sizes that were previously covered by a single field. Apple's M-series processors, AMD's Zen architectures, and NVIDIA's flagship GPUs all have die areas that exceed the high-NA fieldโ€”meaning these chips will require stitching if manufactured with high-NA EUV.

Beyond High-NA: Hyper-NA on the Horizon

Bottiglieri et al. (2025) look further ahead, exploring the opportunities and challenges of "hyper-NA" EUV systems with numerical apertures beyond 0.55. Published at SPIE their analysis from ASML engineers maps the physics of even higher resolutionโ€”and the escalating challenges that accompany it.

Increasing NA beyond 0.55 would further reduce the exposure field, exacerbating the stitching problem. It would also require immersion in a high-refractive-index mediumโ€”analogous to the water immersion that extended 193nm lithography by a full generation. But no suitable immersion fluid for 13.5 nm EUV light currently exists. Bottiglieri et al. survey candidate materials and conclude that hyper-NA EUV without immersion is likely limited to ~0.7 NA, beyond which optical aberrations become prohibitive.

The paper also discusses an alternative to higher NA: anamorphic reduction ratio optimization. By adjusting the magnification asymmetry of the projection optics, it may be possible to partially recover field size without sacrificing resolutionโ€”a trade-off that would benefit large-die manufacturers at the cost of some patterning flexibility.

Source-Mask Optimization: Computational Rescue

Two papers address the depth-of-focus challenge that accompanies high NA through computational techniques:

Libeert et al. (2025), also at SPIE, demonstrate that source and mask optimization (SMO) can enhance depth of focus for high-NA EUV by 30โ€“40% compared to conventional illumination. The approach computationally co-optimizes the illumination source shape (the angular distribution of light hitting the mask) and the mask pattern (including sub-resolution assist features) to maximize the process windowโ€”the range of dose and focus conditions over which the printed features meet specifications.

The 30โ€“40% DOF enhancement is significant because high-NA EUV's reduced depth of focus (~30 nm or less, compared to ~100 nm for 0.33 NA) leaves little margin for wafer flatness variations, resist thickness non-uniformity, and scanner focus drift. SMO recovers much of this margin without hardware changes.

Li et al. (2024) publish an influential contribution in Opto-Electronic Advances introducing a fast SMO algorithm for high-NA EUV that reduces computation time by 10ร— compared to conventional methods. The speed improvement matters because SMO must be performed for every unique mask layoutโ€”and a modern logic chip contains billions of features distributed across hundreds of unique mask layers. Slow SMO is a bottleneck that delays time-to-market.

Critical Analysis: Claims and Evidence

<
ClaimEvidenceVerdict
High-NA EUV achieves sub-8nm patterningDemonstrated at imec on EXE:5000 (Wiaux et al.)โœ… Supported
Stitching overlay <2nm is achievable1.2 nm mean, ยฑ0.8 nm 3ฯƒ demonstratedโœ… Supported
SMO enhances DOF by 30-40%Simulation + experimental verification (Libeert et al.)โœ… Supported
Hyper-NA (>0.55) is feasible without immersionLimited to ~0.7 NA by aberrations (Bottiglieri et al.)โš ๏ธ Uncertain
High-NA EUV is ready for volume manufacturingEXE:5000 operational; production qualification ongoingโš ๏ธ Uncertain (timeline)

The Photoresist Bottleneck

A challenge that these optics-focused papers do not addressโ€”but that may determine high-NA EUV's practical timelineโ€”is photoresist performance. At sub-8nm features, the number of photons absorbed per pixel during exposure drops below 100, entering a regime where statistical fluctuations in photon absorption (shot noise) produce random variations in feature size and placement.

This stochastic patterning failure cannot be solved by optics or computationโ€”it is a fundamental consequence of the quantum nature of light. The solution must come from photoresist chemistry: materials that achieve the required chemical change (solubility switch) with fewer photons while maintaining low line-edge roughness. Current chemically amplified resists are reaching their fundamental limits, and the leading alternativesโ€”metal oxide nanoparticle resistsโ€”are still maturing.

Open Questions and Future Directions

  • Will chiplet architectures reduce the need for stitching? If large processors are disaggregated into smaller chiplets (each fitting within the 26ร—16.5 mmยฒ field), stitching becomes unnecessary. The chiplet trend in semiconductor design may be accelerated by high-NA EUV field limitations.
  • Can stitching overlay reach <1nm? Sub-nanometer stitching would eliminate the need for relaxed design rules at stitch boundaries. Whether this is achievable with current scanner stage control technology is an active research question.
  • What is the cost per wafer for high-NA EUV? The EXE:5000 costs approximately $350 million. Whether the resolution gain justifies the capital cost depends on the number of patterning steps that can be consolidated from multi-patterning (two exposures at 0.33 NA) to single-patterning (one exposure at 0.55 NA).
  • How will China's chipmaking sector respond? Export controls prevent Chinese companies from acquiring EUV scanners. High-NA EUV widens the technology gap further. China's domestic lithography efforts (using DUV with multi-patterning) face increasingly severe resolution limits.
  • Is the stitching penalty acceptable for memory vs. logic? DRAM and NAND have relatively regular, repeating patterns that may be more tolerant of stitching artifacts than complex logic layouts. High-NA EUV may find earlier adoption in memory manufacturing.
  • Implications for the Semiconductor Industry

    High-NA EUV lithography represents both a technical triumph and a design constraint. The resolution gain is genuine and extends Moore's Law scaling by at least two nodes. But the field-size reduction and stitching requirement impose costsโ€”in overlay control, design rule complexity, and scanner throughputโ€”that partially offset the resolution benefit.

    The semiconductor industry's response will likely be pragmatic: use high-NA EUV for the most critical patterning layers (metal interconnects at the tightest pitches) while continuing to use 0.33 NA EUV for layers where the resolution gain is not needed and the larger field is preferred. This layer-selective adoption strategy maximizes the value of the $350 million investment while minimizing the disruption of switching the entire fab to a new tool.

    The chips that emerge from this technology will be smaller, faster, and more power-efficient. Whether they will also be affordable depends on factors far beyond lithographyโ€”including packaging costs, yield rates, and the geopolitical landscape that determines who can buy the machines in the first place.

    References (4)

    [1] Wiaux, V., Davydova, N., Van Look, L. et al. (2025). Stitching insights towards high numerical aperture extreme ultraviolet lithography: An experimental study. Journal of Micro/Nanopatterning, Materials, and Metrology, 24(1), 011012.
    [2] Bottiglieri, G., Woltgens, P., van der Meer, R. et al. (2025). Advancing semiconductor patterning with EUV hyper NA: Opportunities and challenges. Proc. SPIE, 13426, 3050509.
    [3] Libeert, G., Franke, J., Leitao, S. et al. (2025). Depth-of-focus enhancement in high-numerical aperture EUV lithography by source and mask optimization. Proc. SPIE, 13426, 3072478.
    [4] Li, Z., Dong, L., Ma, X. et al. (2024). Fast source mask co-optimization method for high-NA EUV lithography. Opto-Electronic Advances, 7, 230235.

    Explore this topic deeper

    Search 290M+ papers, detect research gaps, and find what hasn't been studied yet.

    Click to remove unwanted keywords

    Search 8 keywords โ†’