4.3.14 · Hardware › Semiconductor Fabrication
Intuition Ek-sentence ka idea
ALD ek film ko ek atomic layer at a time grow karta hai ek chemical reaction ko do self-limiting half-reactions mein split karke, jiske beech purges hote hain — isliye film ki thickness number of cycles se set hoti hai, na ki time, flux, ya precisely tuned temperature se.
Definition Atomic Layer Deposition
ALD ek vapor-phase thin-film deposition technique hai jisme do (ya zyada) gaseous precursors ko ek chamber mein sequentially, kabhi ek saath nahi pulse kiya jaata hai. Har pulse sirf surface sites ke saath react karta hai aur ruk jaata hai jab wo sites consume ho jaate hain — yahi self-limiting behavior hai. Ek poora ABAB… cycle ek fixed sub-monolayer amount deposit karta hai jise growth per cycle (GPC) kehte hain.
Canonical example hai Al₂O₃ ko trimethylaluminum (TMA) aur water se deposit karna:
Half A: Al-OH ∗ + Al(CH 3 ) 3 → Al-O-Al(CH 3 ) 2 ∗ + CH 4 ↑
Half B: Al-CH 3 ∗ + H 2 O → Al-OH ∗ + CH 4 ↑
(asterisk * ek surface species ko mark karta hai).
Intuition Do half-reactions kyun?
Agar tum TMA aur water ek saath dose karo (jaise CVD mein), toh wo gas phase mein / har jagah react karenge, jisse ek rough, uncontrolled film milegi. Unhe alag karke aur beech mein purge karke, har precursor sirf us surface ke saath react kar sakta hai jo doosre ne chodi hai. Surface woh "referee" hai jo reaction ko rokti hai — yahi ALD ko uski magical conformality aur thickness control deta hai.
Setup: Ek surface par reactive sites ki ek finite density N 0 hai (e.g. -OH groups per cm²). Jab precursor A flow karta hai, toh wo kisi available site ke saath kuch probability se react karta hai. Maano θ ( t ) = already react ho chuke sites ka fraction.
Naye reactions ki rate proportional hai (precursor arriving) × (sites abhi bhi free):
d t d θ = k P ( 1 − θ )
jahan P precursor partial pressure hai aur k ek rate constant. ( 1 − θ ) kyun? Kyunki jo site already occupied hai woh dobara react nahi kar sakti — reaction ka fuel khatam ho jaata hai .
Is first-order ODE ko solve karo (separable):
∫ 0 θ 1 − θ ′ d θ ′ = ∫ 0 t k P d t ′ ⇒ − ln ( 1 − θ ) = k P t
θ ( t ) = 1 − e − k P t
Growth per cycle phir aata hai:
GPC = N atoms/vol θ ma x N 0 ( thickness added per full cycle )
Typical GPC ≈ 0.9–1.2 Å/cycle Al₂O₃ ke liye. Total thickness:
d = GPC × ( # cycles )
Pulse A (e.g. TMA) → surface -OH ke saath react karta hai, saturate hota hai, CH₄ release karta hai.
Purge (inert gas, N₂/Ar) → excess A aur byproducts ko remove karta hai. KYUN: B ke saath gas-phase mixing prevent karta hai.
Pulse B (e.g. H₂O) → surface -CH₃ ke saath react karta hai, -OH restore karta hai.
Purge → next cycle ke liye clean.
N baar repeat karo → thickness = N × GPC.
Ek temperature window hota hai jahan GPC temperature ke against flat rehta hai. Bahut thanda → precursor condense ho jaata hai ya reaction incomplete rehti hai. Bahut garam → precursor decompose ho jaata hai (CVD ki tarah) ya ligands react karne se pehle desorb ho jaate hain. Window ke andar, process purely surface-controlled hota hai → beautifully reproducible.
Worked example Example 3 — Saturation curve math
k P = 0.5 s − 1 ke saath, 2 s pulse ke baad coverage kya hogi?
θ = 1 − e − 0.5 × 2 = 1 − e − 1 = 1 − 0.368 === 0.63 == .
6 s ke baad: 1 − e − 3 = 0.95 . Diminishing returns kyun? Exponential approach — last sites ko hit karna sabse rare hota hai. Isliye real recipes thoda over-dose karte hain saturation guarantee karne ke liye.
Common mistake "Zyada lamba pulse = thicker film"
Kyun sahi lagta hai: CVD/sputtering mein, zyada time = zyada material. Intuitive!
Fix: ALD mein har pulse saturate hoke ruk jaata hai (θ → θ ma x ). Extra time sirf precursor waste karta hai. Thickness cycle count se set hoti hai, pulse length se nahi. Lambe pulses sirf high-aspect-ratio features mein complete saturation ensure karne mein help karte hain.
Common mistake "Ek cycle = ek full monolayer"
Kyun sahi lagta hai: Naam mein hi "Atomic layer" hai.
Fix: GPC usually sub-monolayer (~0.3–0.5 of a monolayer) hota hai kyunki bulky ligands sterically neighboring sites ko block karte hain. "Atomic layer" layer-by-layer control ko refer karta hai, na ki ek perfect monolayer per cycle ko.
Common mistake "Purge skip karne se time bachta hai"
Kyun sahi lagta hai: Purges idle waiting jaisi lagte hain.
Fix: Purging ke bina, leftover precursor A, B se gas phase mein milta hai → CVD-jaisi reaction , roughness, particles, conformality ka loss. Purge hi woh cheez hai jo ALD ko define karti hai.
Recall Ek 12-saal ke bacche ko explain karo (click to reveal)
Socho ek wall paint kar rahe ho lekin paint sirf dry spots par chipkti hai. Tum "paint A" spray karo — woh jahan bhi chipak sakti hai chipak jaati hai aur phir ruk jaati hai (koi dry spot nahi bachi). Tum extra spray ponch dete ho. Phir tum "paint B" spray karo jo wall ko phir se dry kar deti hai. Phir poncho. Har A-phir-B round exactly ek thin coat add karta hai, hamesha same thickness — deep cracks ke andar bhi — kyunki paint khud decide karti hai kab rukna hai. Aur mota chahiye? Bas zyada rounds karo!
Mnemonic Cycle yaad rakho
"Please Purge, Please Purge" → P ulse A, P urge, P ulse B, P urge.
Aur "ALD = A Layer, Deliberately" — deliberate, self-limiting, cycles gino.
Zyada dose karne se thickness kyun nahi badhti?
Final film thickness kya set karta hai?
ALD high-aspect-ratio trenches mein uniquely kyun accha hai?
Agar purge steps hata do toh kya hoga?
Har ALD half-reaction apne aap kyun ruk jaati hai? Yeh self-limiting hai — precursor sirf available surface sites ke saath react karta hai aur ruk jaata hai jab wo consume ho jaate hain.
Final ALD film thickness kya set karta hai? Number of cycles × growth-per-cycle (GPC); na ki pulse duration ya flux.
Self-limiting half-reaction ke liye coverage vs time equation likho. θ ( t ) = 1 − e − k P t , θ ma x par saturating.
ALD high-aspect-ratio structures ke liye kyun pasand kiya jaata hai? Self-limiting reactions har surface ko equally saturate karti hain, ~100% conformal step coverage deti hain, trenches mein deep bhi.
Ek ALD cycle ke 4 steps kya hain? Pulse A → Purge → Pulse B → Purge.
Purge step kyun zaroori hai? Yeh excess precursor aur byproducts remove karta hai taaki precursors kabhi gas phase mein na milein (jo CVD-jaisi uncontrolled growth cause karega).
"ALD window" kya hai? Ek temperature range jahan GPC flat/constant hota hai kyunki growth purely surface-controlled hoti hai (na condensation, na thermal decomposition).
Classic Al2O3 ALD precursor pair batao. Trimethylaluminum (TMA) + water (H2O), CH4 release karte hue.
Kya ek ALD cycle ek full monolayer hai? Nahi — usually sub-monolayer hoti hai bulky ligands ki steric hindrance ki wajah se.
Al2O3 ALD ke liye typical GPC? Lagbhag 0.9–1.2 Å per cycle.
Two self-limiting half-reactions
Sequentially, never together
Steric hindrance of ligands