In short: Immersion lithography replaces the air gap under the projection lens with ultrapure water, raising the effective numerical aperture from ~0.93 to ~1.35 so a 193 nm scanner can print features that would otherwise demand a shorter wavelength.
Why the air gap became the bottleneck
Optical projection lithography resolves features according to the Rayleigh criterion, CD = k1 × λ / NA. For a 193 nm ArF excimer source, engineers had already pushed k1 toward its hard physical floor of 0.25 using phase-shift masks, off-axis illumination and optical proximity correction. Wavelength was fixed by the laser chemistry. That left numerical aperture, NA = n · sinθ, as the only lever with headroom — and in a dry scanner the medium index n is that of air, essentially 1.0. Even a perfect lens that captured light at grazing 90° could reach NA just under 1.0, and in practice dry tools topped out near 0.93.
Around 2004–2007 the industry was stuck: 193 nm dry scanners could not comfortably print the 45 nm half-pitch node, and the hoped-for 157 nm F2 laser program collapsed over catastrophic birefringence in CaF2 optics. Immersion rescued 193 nm and kept it alive for another decade.
The physics: a liquid lifts the ceiling
Snell's law tells us the spatial frequency a lens can collect scales with the refractive index of the medium the light travels through just before it reaches the resist. Put ultrapure water (n ≈ 1.44 at 193 nm) in the final gap instead of air (n ≈ 1.0), and the maximum collectable angle — and therefore NA — rises by that index ratio. The same lens that managed NA 0.93 in air can reach NA 1.35 through water. Critically, the wavelength inside the water shrinks to 193/1.44 ≈ 134 nm, which is the finer effective wavelength that actually exposes the photoresist.
Wavelength and NA across the nodes
| Approach | Source λ | Medium (n) | Practical NA | Served nodes |
|---|---|---|---|---|
| 193 nm dry | 193 nm | air (1.0) | ~0.93 | 65 nm |
| 193 nm immersion | 193 nm | water (1.44) | 1.35 | 45 / 32 / 22 nm (+ multipatterning down to ~7 nm) |
| EUV | 13.5 nm | vacuum (1.0) | 0.33 → 0.55 (High-NA) | 7 nm and below |
What engineering the water film actually takes
Dropping water under a lens that scans a 300 mm wafer at >500 mm/s is harder than it sounds. The scanner (ASML's TWINSCAN XT/NXT families dominate here) uses a showerhead-and-confinement nozzle that injects degassed ultrapure water ahead of the lens and vacuum-recovers it behind, holding a stable ~1 mm meniscus that travels with the exposure field.
- Bubbles and particles scatter the beam and print as random defects, so the water is degassed to a few parts-per-billion of dissolved gas and recirculated through sub-10 nm filtration.
- Watermarks and leaching appear where the receding meniscus leaves residue, driving the development of topcoat films and, later, topcoat-free immersion resists.
- Thermal stability matters because the water's index drifts with temperature; the loop is held to millikelvin control to avoid focus and overlay drift.
Researchers chased "second-generation" high-index fluids (n > 1.65) and matching high-index lens glasses to push NA past 1.5, but no manufacturable fluid/lens pair emerged. 1.35 NA water immersion became — and remains — the practical limit of optical projection on 193 nm.
The handoff to multipatterning
At NA 1.35 and k1 near 0.28, a single immersion exposure resolves roughly a 38–40 nm half-pitch. To print 22 nm, 14 nm and 10 nm logic, fabs split one layer into several immersion exposures — litho-etch-litho-etch (LELE), or self-aligned double/quadruple patterning (SADP/SAQP) where sidewall spacers define the final pitch. This is where design-side EDA becomes inseparable from the scanner: Synopsys Proteus and Cadence/Siemens OPC+ engines compute the mask corrections, Mentor/Siemens Calibre runs the decomposition and coloring checks, and litho-aware DRC flags any geometry the immersion tool cannot resolve. A single immersion layer may now need four masks.
Where immersion sits today
Immersion did not disappear when EUV arrived — it complements it. EUV (see our deep dive on EUV lithography explained) handles the most critical fine layers below 7 nm, while 193 nm immersion still prints the vast majority of a modern chip's less-aggressive layers far more cheaply and with far higher throughput. A typical advanced node mixes both. If you want the mask-side mechanics that make either work, read our companion piece on optical proximity correction (OPC), and browse the full CourseTron course catalog for the hands-on labs.
For teams moving from coursework into real tapeouts, the physical-design and DFM flows taught alongside this material map directly onto industry roles — the kind of silicon-design services practiced at Avecas's VLSI design services rely on exactly this litho-aware mindset.
Want to see immersion, OPC and multipatterning as a single simulated flow? Our CourseTron Nano Fab track walks you through resolution math, scanner mechanics and litho-aware DRC with real EDA tools. Explore the course →
FAQ
Why water and not some other liquid?
At 193 nm water is highly transparent, has a usefully high index (~1.44), is cheap, non-toxic, and easy to purify and recirculate. Higher-index fluids existed in the lab but none combined transparency, stability and manufacturability well enough to deploy.
Does the water touch the resist directly?
Yes — the film contacts the wafer surface. That is why topcoat films or specially formulated immersion resists are used, to stop component leaching into the water and to prevent watermark defects as the meniscus recedes.
Why stop at NA 1.35 instead of going higher?
NA in immersion is capped by the index of the fluid: sinθ cannot exceed 1, so NA cannot exceed n. Water gives a hard ceiling around 1.44, and defect and depth-of-focus trade-offs make ~1.35 the practical maximum. Beating it required new high-index fluids and glasses that never became manufacturable.
Is immersion obsolete now that EUV exists?
No. EUV prints the finest layers; 193 nm immersion still prints most layers of an advanced chip at higher throughput and lower cost. Modern fabs run both in the same process flow.