Noesis
4.3.14 · D5Semiconductor Fabrication

Question bank — Atomic layer deposition (ALD)

1,528 words7 min read↑ Back to topic

Before we start, three words we lean on constantly — make sure each is a picture in your head, not just a term:

True or false — justify

Trap: the name "atomic layer" invites over-claims, and CVD/sputtering intuition sneaks in.

Doubling the TMA pulse length doubles the film thickness.
False. Each pulse is self-limiting — once every surface site has reacted, more precursor just gets purged away. Thickness is set by cycle count, not pulse time.
One ALD cycle deposits exactly one complete monolayer of atoms.
False. GPC is usually sub-monolayer (~0.3–0.5 of a monolayer) because bulky ligands sterically block neighbouring sites. "Atomic layer" means layer-by-layer control, not one perfect layer per cycle.
Raising the temperature inside the ALD window makes the film grow faster per cycle.
False. Inside the window GPC is deliberately flat vs temperature because growth is purely surface-controlled. Flatness is the whole point — it makes the process reproducible.
If you skip the purge steps, ALD still works, just with a little more roughness.
False. Skipping purge lets precursor A meet B in the gas phase, which is CVD-like — you lose self-limitation, conformality, and get particles. The purge is what defines ALD.
ALD and CVD use fundamentally different chemistries.
False (mostly). ALD is often the same chemistry as CVD but split into two self-limiting halves separated in time. Same reaction, different delivery.
The saturation curve literally reaches .
False in practice. Mathematically it approaches 1 only as ; physically it saturates at a sub-monolayer set by steric hindrance. It gets close enough fast, which is why recipes slightly over-dose.
ALD is a good choice when you need to coat the inside of a 40:1 aspect-ratio trench evenly.
True. Because each half-reaction saturates every surface to the same regardless of how slowly precursor arrives, deep and shallow regions end up equally thick — see DRAM Capacitor Fabrication.

Spot the error

Each statement below sounds authoritative but hides a mistake. Name it.

"To coat a very deep trench, use a higher pressure precursor pulse so it pushes to the bottom faster."
The fix isn't higher pressure, it's a longer dose/purge time so molecules diffuse to the bottom before you switch precursors. Self-limitation guarantees uniform thickness once every surface saturates — you only need to give it time, not force.
"ALD gives 100% conformality because the precursor travels in straight lines and reaches every surface."
That describes line-of-sight sputtering (Physical Vapor Deposition (Sputtering)), which is non-conformal. ALD's conformality comes from self-limiting surface reactions, not from how the vapour travels.
"Since coverage after 2 s is 0.63 and we want 0.95, we should raise the temperature."
You should dose longer (the saturation curve is a function of ), or the temperature must stay inside the ALD window. Raising temperature outside the window causes decomposition, not better coverage.
"The term in is just an approximation."
It's the physical heart of self-limitation: an already-reacted site cannot react again, so the number of available sites is . Remove it and the reaction never stops — you'd have CVD, not ALD.
"ALD deposits fast, so it's ideal for growing thick 500 nm layers."
ALD is slow — one thin sub-monolayer per multi-second cycle. It's chosen for precision and conformality on thin films (a few nm), not throughput. For thick layers CVD or sputtering win.
"The asterisk in means the species is excited or ionised."
No — the asterisk marks a surface species, i.e. a group bonded to the growing film, as opposed to a molecule in the gas phase.

Why questions

Why does dosing a precursor for longer eventually stop adding any material?
Because the reaction is self-limiting: once every reactive surface site is consumed, , so the reaction rate drops to zero. Extra precursor has nothing left to react with.
Why is a purge needed between the two precursor pulses and not just at the end?
To remove leftover precursor A before B arrives, so they can never meet in the gas phase. Gas-phase reaction would be uncontrolled CVD-like growth with roughness and particles.
Why is GPC sub-monolayer instead of a full monolayer?
Because the precursor's bulky ligands physically occupy space and sterically block neighbouring sites, so not every site can react in a single pulse — the surface fills only partially.
Why does the coverage equation involve an exponential, , rather than a straight line?
Because the reaction rate is proportional to remaining free sites, which themselves shrink as coverage grows — a quantity whose growth is proportional to how much is left always produces exponential approach to a ceiling.
Why is ALD prized for high-k gate dielectrics that are only a few nm thick?
Because thickness is controlled to ~1 Å by simply counting cycles (), giving atomic-scale precision impossible with sputtering or timed CVD.
Why does GPC stay flat with temperature only inside the "ALD window"?
Inside the window growth is purely surface-controlled — reaction goes to saturation but precursor neither condenses (too cold) nor decomposes (too hot), so each cycle deposits the same amount regardless of exact temperature.
Why do real recipes deliberately over-dose slightly past the calculated saturation time?
Because the saturation curve approaches its ceiling exponentially — the last free sites are the rarest to hit, so a safety margin guarantees complete coverage everywhere, including deep in trenches.

Edge cases

Trap: the interesting behaviour lives at the extremes — zero dose, infinite dose, too cold, too hot.

What is the coverage at , before any precursor has been dosed?
(from ) — an empty, fully-available surface. This is the starting condition of every half-reaction.
As dose time , does the film keep thickening within a single pulse?
No. and stops; the pulse cannot add more than one saturated sub-monolayer no matter how long you wait. To go thicker you must run another cycle.
What happens to the film if the substrate is held below the ALD window (too cold)?
Precursor may condense on the surface or the surface reaction stays incomplete, so GPC becomes unreliable and no longer self-limiting — you lose ALD's reproducibility.
What happens above the ALD window (too hot)?
The precursor thermally decomposes (CVD-like, non-self-limiting) or ligands desorb before reacting, so GPC either shoots up uncontrollably or drops — either way the clean layer-by-layer control is gone.
What if the very first surface has no reactive -OH sites at all (a nucleation-inhibited surface)?
The first cycles deposit little or nothing — there are no sites for the precursor to grab. This "nucleation delay" is a real ALD phenomenon; sometimes a surface pre-treatment is needed to seed sites.
In a deep trench, why does the bottom still reach the same as the top even though precursor arrives there much later?
Because self-limitation means each site reacts once and stops; late-arriving precursor still finds unreacted sites at the bottom and saturates them to exactly the same ceiling — given enough dose time.
If two different trenches have aspect ratios 5:1 and 40:1, will the final film thicknesses differ?
No — both saturate to the same per cycle, so both get the same thickness. The 40:1 trench only needs longer dose/purge times to let precursor diffuse deep before switching.

Connections