Masks - reticles and projection systems — Hardware notes

WHAT are we even talking about?

WHY 4× reduction? Two reasons:

Making the reticle is easier at larger scale — a defect of size $d$ on the reticle becomes $d/4$ on the wafer, so tolerances on the mask are relaxed by 4×.
But you pay for it: the printed field on the wafer is $4\times$ smaller in each dimension (so $16\times$ smaller in area), so you must step across the wafer many more times.

HOW small can a projection system print? (Derive the resolution limit)

We derive the Rayleigh resolution from first principles.

Step 1 — Diffraction sets a minimum angle. A grating of pitch $p$ (line + space) diffracts light. The first diffraction order comes out at angle $\theta$ given by: $p \sin\theta = \lambda$

Why this step? This is the grating equation from wave interference: constructive interference of first order requires a path difference of one wavelength $\lambda$ .

Step 2 — The lens must capture that order. To reconstruct the pattern the lens must gather at least the 0th and 1st diffraction orders. The lens can accept rays up to a maximum half-angle $\theta_{max}$ . Define the numerical aperture: $\mathrm{NA} = n \sin\theta_{max}$ where $n$ is the refractive index of the medium between the last lens and the wafer ( $n=1$ in air; $n\approx1.44$ in water for immersion lithography).

Why this step? NA quantifies "how wide a cone of diffracted light the lens can collect." Bigger cone = captures higher orders = sharper image.

Step 3 — Combine. The smallest resolvable pitch requires $\sin\theta \le \sin\theta_{max}$ , i.e. $\lambda/p \le \mathrm{NA}$ (with $n=1$ ), giving minimum pitch $p_{min}=\lambda/\mathrm{NA}$ . The smallest feature (half-pitch / critical dimension) is half of this in the ideal case, and a process-dependent factor $k_1$ rolls in the real-world losses (partial coherence, resist, imperfect capture):

WHY the tension: To shrink CD you raise NA — but DOF falls as $1/\mathrm{NA}^2$ , so high-NA systems have razor-thin focus tolerance. Smaller $\lambda$ shrinks CD and is gentler on DOF, which is why the industry marched $436\,\text{nm} \to 365 \to 248 \to 193\,\text{nm} \to$ EUV $13.5\,\text{nm}$ .

Worked examples

Worked example Field stepping count

A standard reticle field is $26\times33\,\text{mm}$ at reticle scale. Because of the $4\times$ demagnification, the printed field on the wafer is $\frac{26}{4}\times\frac{33}{4}=6.5\times8.25\,\text{mm}$ , an area of $6.5\times8.25\approx 53.6\,\text{mm}^2$ . On a 300 mm wafer (area $\approx 70{,}686\,\text{mm}^2$ ): $N \approx \frac{70686}{53.6}\approx 1300\ \text{fields}$ Why this step? This is the crux: the field you see on the reticle ( $26\times33$ mm) is not the field printed on the wafer — divide each dimension by 4 first. Because the projected field is only ~54 mm², the scanner must repeat the exposure ~1300 times to cover the wafer — this is throughput, the price of demagnification. (Common trap: forgetting the ÷4 gives a wrong ~82.)

Common mistakes

Resolution Enhancement on the mask (why masks are not "just stencils")

Flashcards

What is the difference between a mask and a reticle?

A mask is any patterned quartz/chrome plate that stencils light; a reticle contains one field/die and is stepped-and-repeated across the wafer with demagnification.

Why do projection systems use 4× demagnification?

Larger reticle features are easier to write/inspect and mask defects shrink 4× on the wafer; cost is smaller printed field → more stepping.

State the Rayleigh resolution formula and define each term.

\mathrm{CD}=k_1\lambda/\mathrm{NA}

;

k_1

=process factor,

\lambda

=wavelength, NA=numerical aperture.

Define numerical aperture.

\mathrm{NA}=n\sin\theta_{max}

, where

n

is the medium's refractive index and

\theta_{max}

the max ray half-angle the lens captures.

Why can immersion lithography have NA > 1?

Water (

n\approx1.44

) between lens and wafer raises

n

\mathrm{NA}=n\sin\theta

How does depth of focus scale with NA and why is that a problem?

\mathrm{DOF}=k_2\lambda/\mathrm{NA}^2

; raising NA to shrink features shrinks focus budget quadratically.

What is the physical floor of

k_1

for single exposure and why?

≈0.25, because the lens must capture at least the 0th and 1st diffraction orders.

Why does EUV use reflective reticles?

All materials strongly absorb 13.5 nm light, so no transmissive quartz works; Mo/Si multilayer mirrors reflect it.

Name three resolution enhancement techniques.

OPC (optical proximity correction), PSM (phase-shift masks), off-axis illumination.

Derive minimum pitch from the grating equation.

p\sin\theta=\lambda

and lens captures up to

\sin\theta_{max}=\mathrm{NA}/n

; so

p_{min}=\lambda/\mathrm{NA}

, half-pitch = CD.

A 26×33 mm reticle field with 4× demag prints what wafer field, and how many fit on a 300 mm wafer?

Wafer field = 6.5×8.25 mm ≈ 53.6 mm²; ~1300 exposures per 300 mm wafer (area ~70 686 mm²).

Recall Feynman: explain to a 12-year-old

Imagine you have a tiny drawing you want to stamp onto a chip, way smaller than a hair. You can't draw that small, so you draw it big on a glass plate, then shine a light through it and use a magnifying lens backwards to shrink the shadow down onto the chip. The chip is coated in special goo that hardens where light hits. Problem: light is wobbly (it spreads out — that's diffraction), so if two lines are too close the shadows smear together. To fix the smear, you either use "bluer" light with a shorter wave, or use a fatter lens that catches more of the spreading light. That's the whole game: shrink the drawing, catch the light, beat the blur.

Connections

Photolithography overview
Photoresist and exposure
Diffraction and the grating equation
Numerical aperture and lens design
Immersion lithography
EUV lithography
Multiple patterning (double/quad)
Optical Proximity Correction (OPC)
Step-and-scan tools (steppers/scanners)

Concept Map

Hinglish (regional understanding)

Intuition Hinglish mein samjho

Dekho, mask/reticle ek tarah ka "light stencil" hai — quartz plate par chromium ka pattern hota hai, jahan chrome hai wahan light block, jahan clear hai wahan light pass. Lekin chip ke features itne chhote hote hain ki hum unhe seedha draw nahi kar sakte. Isliye pattern ko reticle par 4 guna bada banate hain, phir projection lens se usko shrink karke (÷4) wafer ke resist par project karte hain — bilkul slide projector ki tarah, bas ulta (bada picture → chhota image). Yaad rakho: reticle par jo field dikhta hai (jaise 26×33 mm), wafer par woh ÷4 ho ke chhapta hai (6.5×8.25 mm), yani area 16 guna chhota — isliye ek wafer par ~1300 baar step karna padta hai.

Ab sabse important cheez: kitna chhota print kar sakte ho? Iska jawab Rayleigh formula deta hai — $\mathrm{CD}=k_1\lambda/\mathrm{NA}$ . Yaad rakho light diffract karti hai (phailti hai), aur lens sirf ek limited cone ke andar ki rays pakad sakta hai. Jitna chhota wavelength $\lambda$ aur jitna bada NA (numerical aperture), utne fine features. Isliye industry ne wavelength ghatai: 365 → 248 → 193 nm → EUV 13.5 nm. Aur NA badhane ke liye immersion trick aayi — lens aur wafer ke beech paani ( $n=1.44$ ) daal do, taaki NA 1 se upar ja sake.

Par yahan ek trade-off hai jo exam me pucha jaata hai: DOF (depth of focus) $=k_2\lambda/\mathrm{NA}^2$ . Matlab NA badhaoge toh resolution to accha hoga, par focus ka budget NA-square se girta hai — bahut tight ho jaata hai. Aur $k_1$ ka physical floor ~0.25 hai (kyunki lens ko kam se kam 0th aur 1st diffraction order pakadne padte hain). Isse neeche jaana ho toh multiple patterning ya OPC/phase-shift mask jaisi tricks lagti hain. Yeh saara "kaise chhota chhapein" ka khel hai.

Test yourself — Semiconductor Fabrication

Connections