In short: Immersion lithography replaces the air gap between the final lens and the wafer with ultrapure water (n≈1.44 at 193 nm), lifting the numerical aperture past 1.0 so 193 nm light can print features it otherwise could not — the workhorse that carried the industry from the 65 nm node to roughly 10 nm and still underpins EUV multi-patterning today.
Why the air gap became the bottleneck
Optical lithography resolution follows the Rayleigh equation, CD = k₁·λ / NA, where CD is the smallest half-pitch you can resolve, λ is the exposure wavelength, k₁ is a process factor (physically floored near 0.25 for a single exposure), and NA is the numerical aperture of the projection lens. By the mid-2000s, λ had bottomed out: 193 nm ArF excimer lasers were the shortest practical wavelength, because 157 nm (F₂) optics demanded exotic CaF₂ lenses and the whole program collapsed under cost and birefringence problems. With λ stuck and k₁ pinned near its floor, the only lever left was NA.
NA is defined as NA = n·sin(θ), where n is the refractive index of the medium the light travels through just before it hits the resist, and θ is the half-angle of the focused cone. In a conventional "dry" scanner that medium is air (n≈1.0), so NA can never reach 1.0 — sin(θ) saturates at 1. Dry 193 nm tools topped out around NA 0.93. That is the wall immersion lithography was invented to climb over.
The physics: bending the resolution curve with water
Insert ultrapure, degassed water between the final lens element and the wafer and n jumps from 1.0 to about 1.44 at 193 nm. Because NA scales directly with n, a lens that captured the same angular cone now reports an NA of up to ~1.35 in production ASML tools (the TWINSCAN XT and NXT lines). Crucially, the exposure wavelength inside the high-index medium shrinks: the effective wavelength becomes λ/n ≈ 193/1.44 ≈ 134 nm. The light arriving at the resist behaves as if it were a shorter-wavelength beam, which is exactly what you need to shrink CD.
Numbers that matter: Dry 193 nm at NA 0.93, k₁≈0.3 → CD ≈ 62 nm half-pitch. Immersion at NA 1.35, same k₁ → CD ≈ 43 nm. Push k₁ to its 0.25 floor with aggressive RET and you reach ~36 nm half-pitch in a single exposure — the practical limit before multi-patterning takes over.
A subtle but important point: immersion also improves depth of focus relative to a dry tool at the same NA, because DOF scales with λ/(n·sin²θ). The higher index partially offsets the DOF penalty of going to very high angles, which is why hyper-NA immersion was manufacturable at all.
Engineering a moving puddle of water
The elegant part is that the water is not a static bath. ASML's design is a localized immersion hood that carries a small puddle — only the region under the lens is wetted — while the wafer scans beneath it at speeds over 500 mm/s. The system continuously supplies and recovers water through a showerhead-and-vacuum manifold, so the meniscus travels with the lens. This keeps the stage and metrology dry and lets the tool keep the high throughput (150+ wafers/hour) that fabs demand.
Three defect mechanisms dominated the early learning curve, and each drove a specific fix:
- Microbubbles: dissolved gas nucleating in the beam path scatters light and prints as defects. Solution: aggressively degassed water and careful fluid handling.
- Watermarks and leaching: water interacting with the resist leaves residues, and resist components leach into the water and fog the lens. Solution: topcoat layers over the resist and, later, topcoat-free resists engineered for immersion.
- Film pulling and thermal drift: the receding meniscus can drag a thin film; evaporative cooling shifts the local refractive index. Solution: tuned scan/recovery dynamics and tight thermal control of the water (millikelvin stability).
Where immersion sits across nodes
| Era / node | Technique | Half-pitch reach |
|---|---|---|
| 90–65 nm | Dry 193 nm, NA ≤ 0.93 | ~65 nm |
| 45–32 nm | Immersion single exposure, NA 1.35 | ~38–45 nm |
| 28–10 nm | Immersion + double/quad patterning (LELE, SADP, SAQP) | ~15–20 nm |
| 7 nm and below | EUV (13.5 nm) for critical layers; immersion still runs most other layers | <15 nm |
Still indispensable: even in a leading-edge EUV fab, the majority of mask layers on a modern SoC are still printed on 193 nm immersion scanners. EUV is reserved for the few most critical layers because of its cost and throughput. Immersion did not retire — it became the backbone beneath EUV.
Why 1.35 is the ceiling (and why no one went higher)
If water gives NA 1.35, why not a higher-index fluid for NA > 1.5? The industry seriously pursued "third-generation" immersion with high-index fluids and lens materials (sapphire, LuAG) targeting NA ~1.55–1.7. It failed on materials: no manufacturable high-index final-lens material was transparent and birefringence-free at 193 nm, and the candidate fluids had absorption and defect problems. The economics also shifted — double patterning plus the eventual arrival of EUV made the extra NA unnecessary. So water at NA 1.35 remains the enduring high-water mark of optical immersion.
What this means for a design and flow engineer
Immersion-era resolution limits are baked directly into the rules you hit in physical design. Minimum pitch, coloring constraints for double patterning (two or four masks per layer), and pattern-decomposition checks all trace back to what a 193i scanner can print. Tools such as Synopsys IC Validator and Siemens Calibre run multi-patterning decomposition and coloring DRC; Cadence Innovus and Virtuoso enforce the color-aware routing and placement rules. If you have ever debugged an "odd-cycle" coloring violation, you were, in effect, fighting the Rayleigh limit of a water-immersed lens. For the upstream optics and mask side, computational lithography and OPC (Synopsys Proteus, Siemens Calibre OPC) model the immersion imaging to pre-distort the mask so the printed shape matches intent.
To go deeper on how the exposure step fits the larger patterning sequence, see our companion pieces on the photolithography process and on multi-patterning techniques, and browse the full CourseTron course catalog. If you want hands-on industry training or a design-services partner, the team at ChipXpert's VLSI training programs covers these flows end to end.
Ready to connect lithography physics to real DRC decks and PnR constraints? Our Nano Fabrication & Patterning track on CourseTron walks you from the Rayleigh equation to coloring-aware routing with worked examples. Explore the course →
Frequently asked questions
Why water and not some other liquid?
Water is cheap, non-toxic, easy to purify and recover, and — critically — highly transparent at 193 nm with a favorable refractive index of about 1.44. Higher-index candidate fluids existed but suffered from absorption, defects, and lack of matching lens materials, so water won on manufacturability.
Does immersion change the exposure wavelength?
The laser is still 193 nm in vacuum, but inside the water the effective wavelength is λ/n ≈ 134 nm. The light interacts with the resist as if it were shorter, which is what enables finer resolution.
Is immersion lithography obsolete now that EUV exists?
No. EUV handles only the most critical layers in advanced fabs; the majority of layers on modern chips are still printed on 193 nm immersion scanners. Immersion also enabled the multi-patterning that bridged the industry into the EUV era.
What is the practical resolution limit of immersion?
About 36–40 nm half-pitch in a single exposure at NA 1.35 with k₁ near 0.25. Beyond that, double and quadruple patterning (or EUV) are required.