EUV masks and pellicles

In short: An EUV mask is not a transmissive quartz plate but a reflective stack — 40-odd Mo/Si bilayers topped by a patterned tantalum-based absorber on an ultra-flat, near-defect-free low-thermal-expansion blank — and the single hardest unsolved accessory is a thin, >90%-transparent pellicle that can survive hundreds of watts of 13.5 nm power without melting or drooping into the beam.

Extreme-ultraviolet lithography prints the smallest features in modern logic and DRAM, but the photomask that carries the pattern works on a completely different physical principle from the glass reticles used in 193 nm immersion. Because almost every material absorbs 13.5 nm light, you cannot shine EUV through a mask — you have to bounce it off one. That one fact reshapes the entire reticle: the blank, the multilayer, the absorber, defect strategy, and the notorious pellicle. If you have already read our primer on EUV lithography, this article drills into the mask itself.

Why EUV masks are mirrors, not windows

At 13.5 nm the refractive index of every solid is close to 1 and the absorption coefficient is large, so no usable transmissive optic exists. The scanner's illumination and projection optics are therefore all-reflective, and so is the reticle. EUV light arrives at the mask at a non-telecentric chief-ray angle of about 6° (the "CRAO"), reflects off a engineered mirror, and the patterned absorber on top casts the image by removing reflectivity where you want dark features. This oblique incidence creates real pattern-placement and shadowing effects that OPC tools from Synopsys (Proteus) and Siemens EDA (Calibre) must model explicitly — a vertical and a horizontal line of the same drawn width do not print identically.

The reflective multilayer: 40 Mo/Si bilayers tuned to 13.5 nm

The mirror is a distributed Bragg reflector built from alternating molybdenum and silicon layers. Each bilayer is roughly 7 nm thick (about 2.8 nm Mo + 4.2 nm Si), and the stack repeats ~40 times for a total thickness near 280 nm. Constructive interference of the weak reflections from each interface yields a peak reflectivity of only about 65-70% at 13.5 nm — physically the ceiling for this material pair. That number matters enormously: the scanner has roughly ten reflective surfaces between source and wafer, and 0.68 raised to the tenth power is why EUV source power (hundreds of watts of in-band photons) is such a brutal engineering constraint.

Key fact: Molybdenum/silicon is chosen because 13.5 nm sits just below silicon's L-absorption edge, where Si is relatively transparent and Mo provides the index contrast. That absorption edge is exactly why the industry standardized on 13.5 nm rather than some other EUV wavelength.

The absorber and the substrate blank

On top of a thin ruthenium capping layer (which protects the stack from oxidation and etch damage) sits the patterned absorber, classically ~60-70 nm of tantalum boron nitride (TaBN) with a thin anti-reflection oxide. Its job is to suppress reflectivity to a few percent in the dark regions. Because the absorber has finite thickness, the 6° oblique light casts a shadow — the "mask 3D" effect — producing best-focus shifts and HV asymmetry that drive interest in thinner high-refractive-index absorbers (Ni, Co, or alloy materials) and phase-shift designs.

All of this is built on a blank of low-thermal-expansion material (LTEM) such as ULE glass or Zerodur, with a coefficient of thermal expansion near zero (<10 ppb/K) so the pattern does not walk as the reticle heats under beam load. Flatness is specified to tens of nanometers across a 152 mm × 152 mm plate.

Defect-free blanks: the number that keeps fabs awake

A single bump or pit buried in the multilayer — even a few nanometers tall — locally disturbs the Bragg phase and prints as a killer defect on wafer. The industry target is often quoted as essentially zero printable defects larger than ~20-30 nm across the whole blank, a specification far tighter than anything in 193 nm masks. Because perfect blanks are not yet economic, fabs rely on defect mitigation: actinic (at-wavelength) blank inspection locates each buried defect, and pattern-shift or dithering algorithms nudge the absorber layout so the drawn features sit over the defects, hiding them under dark regions.

The pellicle challenge: the hardest accessory in the fab

A pellicle is a thin membrane suspended a few millimeters above the pattern so that any stray particle lands out of the focal plane and does not print. On optical masks this is a cheap polymer film. For EUV it is a nightmare because the membrane sits in the beam twice (incoming and reflected) and must be >90% transparent at 13.5 nm while surviving the full source power — 250 W and climbing toward 600 W for high-NA tools.

Early polysilicon pellicles managed only ~78-83% double-pass transmission and could not tolerate high power; thermal load would heat the membrane to hundreds of degrees, risking rupture or sag into the depth of focus. The current front-runners are ultra-thin (tens of nanometers) free-standing membranes of polycrystalline silicon or, increasingly, carbon nanotube (CNT) sheets, which combine high transmission with mechanical strength and radiative cooling. Even so, several leading-edge fabs have run EUV layers pelliclized or bare depending on defectivity data, because every percent of transmission lost is throughput and source power you cannot get back.

Key fact: At two passes through the membrane, a pellicle with 90% single-pass transmission still costs you ~19% of your light (0.90 × 0.90 ≈ 0.81), directly reducing wafer throughput — which is why 90%+ double-pass is the real target and why CNT films are so attractive.

Watching the image form: an animated view

Mask shops and fabs lean on EDA signoff tools to make these masks manufacturable: Siemens Calibre and Synopsys Proteus handle EUV-aware OPC and mask-3D correction, while actinic inspection and e-beam mask writers (from vendors like IMS and NuFlare) close the loop. Teams that design chips at these nodes, such as the engineers at Avecas VLSI design services, must understand mask effects to write DRC-clean, litho-friendly layouts. For a broader tooling view, see our companion piece on mask data prep and OPC.

Want to go from "I've heard of EUV" to explaining Bragg stacks, mask-3D, and pellicle power budgets in an interview? Our Nano Fab track walks through every layer with worked examples — browse CourseTron courses.

Frequently asked questions

Why can't an EUV mask be transmissive like a normal photomask?

Because essentially all materials strongly absorb 13.5 nm light, no practical transparent substrate exists at that wavelength. The mask must reflect using an interference mirror instead of transmitting through glass.

What limits EUV multilayer reflectivity to about 68%?

The Mo/Si material pair provides only modest index contrast at 13.5 nm, so each interface reflects weakly and even ~40 bilayers summed coherently reach only ~65-70%. This is a fundamental materials ceiling, which compounds painfully across the scanner's ~10 mirrors.

Do all EUV layers run with a pellicle today?

Not always. Pellicle use depends on the layer's defectivity risk versus the transmission loss and power budget. As membranes improve (CNT films, better cooling), pelliclization is becoming standard, especially for critical layers and high-NA tools.

How are buried multilayer defects handled if blanks aren't perfect?

Actinic inspection maps each buried defect, then pattern-shift algorithms reposition the absorber so defects hide under dark (non-printing) regions, effectively mitigating defects that cannot yet be eliminated.