Lithography as a choke point: DUV, EUV, and the R&D stack

Photolithography prints a mask pattern onto a wafer using light. The smallest feature you can resolve is bounded by the Rayleigh criterion:

$\text{CD} = k_1 \cdot \frac{\lambda}{\text{NA}}$

where CD is the critical dimension (the smallest feature), $\lambda$ is the light wavelength, NA is the lens system's numerical aperture, and $k_1$ is a process constant set by mask, resist, and pattern enhancement tricks. With all the enhancements layered in, $k_1$ can reach about 0.25.

The equation explains the industry's trajectory: to print smaller features, you must reduce $\lambda$, increase NA, or improve $k_1$. The first two are physics problems; the third is a 30-year process engineering campaign. The transitions from 248 nm KrF to 193 nm ArF to 13.5 nm EUV each represent a forced wavelength move when the other two levers ran out.

From the early 2000s to the late 2010s, the industry pushed deep ultraviolet (DUV) at 193 nm well past what the Rayleigh equation suggested was possible.

The enabling tricks:

• Immersion lithography — fill the gap between lens and wafer with purified water (refractive index ~1.44), raising the effective NA above 1.0.
• Multi-patterning — print a feature with two, three, or four overlapping exposures (LELE, SADP, SAQP). Each exposure relaxes the resolution requirement but doubles, triples, or quadruples the steps, masks, defect rate, and cost.
• Resolution enhancement — optical proximity correction, source–mask optimization, phase-shift masks. Each squeezes more out of $k_1$.

This combination reached the 7 nm node. SMIC uses 193 nm immersion with multi-patterning to produce 7 nm devices today, demonstrating that DUV multi-patterning is not a hard physical wall — it's a cost and yield wall. Each additional mask roughly doubles the exposure count for that layer.

Below 7 nm, multi-patterning becomes uneconomic — too many exposure passes, too many defects, too many masks per layer. The industry moved to extreme ultraviolet (EUV) at 13.5 nm — a 14× reduction in wavelength from DUV.

The trouble: at 13.5 nm, ordinary lithography becomes impossible.

• No source. No conventional laser or lamp emits at 13.5 nm. The EUV source is a plasma: a 30 μm tin droplet is hit by a CO₂ laser pulse 50,000 times per second, vaporizing into a plasma at ~220,000 °C that emits a flash of 13.5 nm light.
• No transmissive optics. Glass absorbs EUV. The entire optical path is reflective — 10+ multilayer mirrors (Mo/Si alternating layers, ~50 pairs each) with ~70% reflectivity per mirror. The mask itself is reflective rather than transmissive.
• No air. Air absorbs EUV. The entire chamber holds high vacuum.

Each of those problems took a decade to solve.

An EUV exposure involves four distinct subsystems, each from a different specialty supplier.

EUV did not emerge from one firm. It emerged from a multi-decade public-private R&D consortium (LLNL, Sandia, Intel, IBM, AMD, Motorola, ASML, Carl Zeiss, beginning in the late 1990s) backed by tens of billions of dollars in cumulative R&D spending across all participants. ASML acquired Cymer (the source manufacturer) and took stakes in Carl Zeiss SMT (the optics partner) to integrate the stack.

The consequence for catchup: a national or corporate effort to build EUV from scratch would need to reproduce, at minimum, three independently hard things — the plasma source, the multilayer mirror coatings and figuring, and the system integration that holds nanometer-scale alignment across a 180-ton machine. Each is a decades-deep R&D track in its own right.

This is the meaning of "lithography is the choke point." The bottleneck is not just one machine vendor; it is the depth of compounding R&D embedded in that vendor's supply chain.

Returning to the Rayleigh criterion: with $\lambda$ fixed at 13.5 nm, the remaining levers are NA and $k_1$.

High-NA EUV raises NA from the current 0.33 to 0.55. From the equation, a 0.55/0.33 ratio gives a roughly 1.7× resolution improvement. The physical consequences: larger and more demanding mirrors, a smaller exposure field (about half the area per shot, requiring more shots per wafer), tighter focus tolerance, and a per-machine cost roughly twice the current EUV scanners.

The industry's roadmap places High-NA EUV at the 2 nm and below nodes. The structural point is not which firm ships first — it is that the same equation has now been pushed against for thirty years, and each step costs more than the last because $k_1$ is already close to its theoretical 0.25 floor.

EUV scanners are produced by one vendor: ASML, in the Netherlands. Roughly 200 units have been delivered globally. Each costs about $200 million and weighs about 180 tons.

The concentration has structural causes that compound at three levels:

• Within ASML: integrating the source, optics, stage, vacuum, and metrology required acquiring or partnering with the suppliers of each subsystem.
• Within ASML's suppliers: Carl Zeiss SMT is the sole supplier of the multilayer reflective optics, Trumpf the sole supplier of the CO₂ drive laser, Cymer (now an ASML subsidiary) the source integrator. Each is itself a single point.
• Within the customer base: the buyers are three firms (TSMC, Samsung, Intel) operating at the relevant node, plus a small group operating at the prior node. The supplier–customer relationship is bilaterally concentrated.

A new entrant would face the choice of either reproducing the entire stack (a decades-long capex commitment) or accepting that it sits one or two nodes behind. Neither path can be shortened by buying machines that are restricted from export.

Lithography illustrates a general pattern in advanced manufacturing: when a tool requires a decade of integrated R&D across a deep supply chain, the eventual outcome is not many competing vendors. It is one vendor, embedded in a network of single-supplier subsystems, selling to a small number of single-source-buyer customers.

This is the structural reason lithography functions as a choke point in policy. The policy lever exists not because anyone designed it that way, but because the underlying economics of compounding R&D collapsed competition over decades. The next lesson examines the equipment and materials chokepoints — etch, deposition, photoresist, wafers — most of which show the same compounding-R&D pattern at smaller scale.