Multi-patterning techniques — Hardware notes

WHY does multi-patterning exist?

WHY this formula? Diffraction. A grating of pitch $p$ diffracts light into orders at angles $\sin\theta_m = m\lambda/p$ . To form an image you must capture at least the 0th and 1st orders in the lens. The lens accepts up to $\sin\theta = \text{NA}$ , so you need $\lambda/p \le \text{NA}$ , i.e. minimum pitch $p_{\min}=\lambda/\text{NA}$ and half-pitch $\text{HP}=p_{\min}/2 = 0.5\,\lambda/\text{NA}$ . With clever illumination (off-axis, dipole) you squeeze $k_1$ down toward $0.25$ .

Plug numbers: single exposure best case $\text{HP}_{\min}=0.25\times\frac{193\text{ nm}}{1.35}\approx 36\text{ nm}.$

So if the design rules demand a 20 nm half-pitch and EUV isn't available, you cannot do it in one shot → you must split the pattern. That is the entire motivation.

HOW: the main techniques

1. LELE — Litho-Etch-Litho-Etch (Double Patterning, DPT)

WHY it works: each individual exposure sees a relaxed pitch $2p$ (well within the tool's limit), but the final composite pitch is $p$ .

Effective $k_1$ : since two exposures interleave, the achievable half-pitch halves → $k_1^{\text{eff}}\approx 0.125.$

Weakness — overlay: mask B must land exactly between mask A's lines. Any misalignment (overlay error) causes pitch walking (alternating wide/narrow spaces). This is LELE's Achilles heel and it does not average out.

2. SADP — Self-Aligned Double Patterning (Spacer)

WHY self-aligned & superior: the spacer width is set by deposited-film thickness, not by a second exposure — so both edges of the final pitch come from one lithography step. This removes the overlay-driven pitch walking of LELE. The line width is controlled by film deposition thickness, which is far more uniform than a second mask alignment.

Cost: you now get lines everywhere; the layout must be a regular grating and you need extra cut/block masks to remove unwanted spacer loops (spacers form closed loops around mandrel ends).

3. SAQP — Self-Aligned Quadruple Patterning

Do the spacer trick twice: spacers-of-spacers. Final pitch = mandrel pitch / 4. $p_{\text{final}}=\frac{p_{\text{mandrel}}}{2^{N}}\quad(N=\text{spacer generations})$ SAQP: $N=2 \Rightarrow$ pitch/4. Used for ~10 nm-class fins/metal.

4. LELELE / triple, and SAOP (octuple)

Extend LELE to 3+ colors (LELELE) → pitch/3, needing layout coloring (graph coloring: adjacent features on same mask must be far enough apart). SAOP = spacer trick ×3 → pitch/8.

Worked examples

Common mistakes

Flashcards

What sets the single-exposure resolution floor?

Rayleigh limit

\text{HP}=k_1\lambda/\text{NA}

with

k_1\!\to\!0.25

\lambda=193

nm,

\text{NA}\le1.35

⇒ ~36 nm.

Why use multi-patterning instead of a shorter wavelength?

When a shorter-

\lambda

tool (EUV) is unavailable/too costly, split the layout into sparse sub-patterns each printable at 193i.

LELE stands for?

Litho-Etch-Litho-Etch: two exposure+etch passes interleave to halve pitch.

Main weakness of LELE?

Overlay error between the two masks → pitch walking (uneven alternate spaces).

Why is SADP "self-aligned"?

Final pitch is set by spacer film thickness from ONE exposure, not a second mask alignment → no pitch-walking overlay term.

Steps of SADP?

Print mandrels → conformal deposit → etch-back to leave sidewall spacers → remove mandrels (2 spacers/mandrel).

Pitch reduction of SAQP?

Pitch/4 (two spacer generations,

p_{\text{mandrel}}/2^2

General pitch formula for N spacer generations?

p_{\text{final}}=p_{\text{mandrel}}/2^N

Why do spacer processes need cut/block masks?

Spacers form closed loops around mandrel ends and print everywhere; cut masks trim unwanted loops/lines.

Effective

k_1

of double patterning vs single?

~0.125 vs 0.25 (halved half-pitch).

Recall Feynman: explain to a 12-year-old

Imagine you have a stamp that can only print stripes that are so-and-so close together — press harder and they smear into a blur. You want stripes twice as close. Trick: print the odd stripes first with the stamp (easy, they're far apart), then print the even stripes in the gaps with a second press. Now the paper has twice as many stripes even though your stamp never got better! Even smarter (spacers): print fat blocks, spray paint a thin coat on their sides, then dissolve the blocks — the leftover paint edges are your super-thin stripes, and their width is just how thick you sprayed. No second aiming needed = no crooked lines.

Connections

Photolithography — sets the $\lambda$ /NA baseline that forces multi-patterning.
EUV Lithography — 13.5 nm light that reduces the need for multi-patterning.
Overlay and Alignment — the error source that kills LELE.
Etch and Deposition (CVD/ALD) — conformal ALD enables uniform spacers in SADP.
Design Rule Check and Layout Coloring — graph-coloring mask decomposition for LELELE.
Rayleigh Criterion — first-principles resolution physics.

Concept Map

Hinglish (regional understanding)

Intuition Hinglish mein samjho

Dekho, problem simple hai: lithography machine ke paas ek fixed wavelength light hai (193 nm ArF), aur diffraction ki wajah se woh ek limit se zyada chhoti features print nahi kar sakti — Rayleigh formula $\text{HP}=k_1\lambda/\text{NA}$ , jisme $k_1$ ka floor ~0.25 hai, matlab best case ~36 nm half-pitch. Agar design ko isse bhi chhota chahiye aur EUV available nahi, to multi-patterning use karte hain. Idea yeh hai ki ek dense layout ko do-teen sparse layouts me tod do, alag-alag print karo, wafer pe combine ho jaayega — final pitch aadha ya chauthai ho jaata hai, jabki har single exposure aaram se print ho jaata hai.

Do main flavours hain. LELE (Litho-Etch-Litho-Etch): pehle mask A se aadhe lines print+etch, phir mask B se beech ke lines. Problem — mask B ka alignment (overlay) thoda bhi hila to spaces uneven ho jaate hain, isko pitch walking kehte hain. SADP (spacer wala) isse smart hai: pehle motte mandrels print karo, unke sides pe thin film deposit karke etch-back karo — sirf sidewall spacers bachte hain, phir mandrel dissolve kar do. Ab final line ki width sirf film thickness se decide hoti hai, doosre mask ke alignment se nahi — isliye "self-aligned" aur pitch walking nahi hota.

Aur zyada chhota chahiye? Spacer trick do baar karo = SAQP = pitch/4 (formula $p_{\text{mandrel}}/2^N$ ). Yaad rakhna: multi-patterning ne wavelength kam nahi kiya, sirf ek exposure se dense kaam maangna band kiya — extra process steps de kar resolution kharida. Yeh matter isliye karta hai kyunki 10 nm/7 nm nodes mostly isi jugaad pe chale, aur cost/yield ka bada portion yahin decide hota hai.

Test yourself — Semiconductor Fabrication

Connections