EUV Lithography Scanner

1. Resolution and wavelength. A common first-order relation is resolution ≈ k₁·λ/NA. EUV uses λ = 13.5 nm, compared with 193 nm for ArF deep-ultraviolet lithography. This 0.33-NA model displays a 13 nm half-pitch pattern; manufacturable resolution also depends on numerical aperture, process factor k₁, mask design, resist, and process control.
2. EUV requires reflective optics. Air and conventional refractive materials strongly absorb 13.5 nm radiation, so the optical path operates in vacuum and uses multilayer mirrors. A typical coating contains roughly 40–50 molybdenum/silicon bilayers—about 80–100 individual layers—with a period near 6.9 nm that produces constructive interference.
3. Laser-produced plasma source. The model represents 50,000 molten-tin droplet events per second. A pre-pulse expands each droplet, and a later high-power CO₂-laser pulse produces a hot plasma that emits near 13.5 nm. A collector directs part of that radiation toward the intermediate focus. EUV and CO₂ radiation are invisible; violet and red are visualization colors only.
4. Photon budget. Twelve reflections are modeled: collector, four illumination mirrors, mask and six projection mirrors. Each reflectance is calculated using Si/Mo optical constants and a graded multilayer period. The displayed survival excludes collection solid angle, pellicle and mask-pattern losses.
5. Projection surfaces. The folded topology follows published 0.33-NA projection-optics diagrams. Surface normals satisfy reflection at each chief-ray hit. The final teaching relay uses an ellipsoid to focus real marginal rays at NA 0.33. The proprietary optical prescription and exact mechanical CAD are not public and are not claimed here.
6. Reflective mask. The reticle carries an absorber pattern over a multilayer reflector and is imaged onto the wafer at 4× reduction. Oblique incidence separates incoming and reflected paths. A thin pellicle can reduce the probability that particles print as defects.
7. Step-and-scan exposure. A narrow illuminated field is scanned across each die. In this model, the reticle moves opposite the wafer at four times its speed to preserve 4× reduction, and a 2.6 mm wafer-side slit at 0.5 m/s receives approximately 260 source pulses per point.
8. Dual-stage operation. While one modeled wafer is exposed, the other stage performs simplified measurement and exchange steps before the stage roles are swapped. Actual scanners coordinate exposure, alignment, focus, and wafer handling with configuration-dependent speed and overlay specifications.
9. Image and resist. A scalar Fourier model sums incoherent source points through a circular pupil, including defocus phase. Off-axis illumination resolves the 26 nm pitch; on-axis illumination does not. Poisson absorption, a 1.8 nm diffusion kernel and positive-tone development create a fixed-seed sample. CD and 3σ LER are measured separately.
10. Scale and timing. The architecture targets a 2021 production scanner: 13.5 nm, NA 0.33, 300 mm wafers and 26 × 33 mm fields. A separate micrometer view shows tin targets; nanometer views show coatings and resist. Scan profiles, pulse widths and source conversion are teaching parameters. No rated production throughput is inferred from the animated cycle.

Try: Replay the droplet cycle, then follow a photon through the full optical path. Turn off the laser during a scan. Compare off-axis and on-axis illumination, change focus, and sweep wavelength through the multilayer reflection peak.

EUV and infrared are invisible; colored paths are overlays. Ray brightness is compressed for visibility. The resist view previews development at the selected exposure settings; it is not a live camera inside the scanner.

Scanner manufacturer’s product page
Projection-optics workshop paper · 2019
LBNL · Mo / Si ↗