Chemical vapor deposition (CVD): Visual walkthrough

Everything here rests on a single physical picture: gas flows over a hot wafer, and a solid film slowly grows out of that gas. Let us draw exactly that first.

Step 1 — The scene: gas above, wafer below

WHAT. A wafer (a flat disc of silicon) lies at the bottom. Above it flows a river of gas carrying the precursor — the reactant molecules that will become our film. Far from the surface the gas is well-stirred and its reactant amount is steady. Right at the surface, molecules are being eaten by the growing film.

WHY. Before any algebra we must name two numbers, because the whole derivation is a story about the difference between them:

PICTURE. Look at the figure: the reactant dots are dense up top (that is $C_{g}$ ) and thinner near the wafer (that is $C_{s}$ ), because the surface keeps consuming them.

Figure — Chemical vapor deposition (CVD)

Because the surface eats molecules, $C_{s}$ is always less than or equal to $C_{g}$ . Hold that thought — the gap $C_{g} - C_{s}$ is what drives everything.

Step 2 — The invisible wall: the boundary layer

WHAT. Right against the wafer the flowing gas is slowed to a near-standstill by friction with the solid — a thin, almost-stationary skin of gas. This is the boundary layer. A molecule cannot ride the fast river down to the surface; it must crawl across this skin by diffusion (random jiggling that, on average, drifts from crowded regions to empty ones).

WHY this concept. Diffusion is the tool we need because inside the stagnant skin there is no bulk flow to carry molecules — only the aimless thermal wander that statistically moves them from high $C$ to low $C$ . That "high-to-low" drift is the only way across. (For the deeper fluid-dynamics of why the skin exists, see Boundary Layer (Fluid Dynamics).)

PICTURE. The shaded band hugging the wafer is the boundary layer of thickness $δ$ . The fast arrows above it are the main flow; inside it the arrows shrink to nothing.

So a reactant molecule's journey is: fall to the top of the skin (easy, carried by flow) → diffuse across the skin (slow) → react at the surface (chemistry). Two slow steps. We now quantify each.

Step 3 — Flux #1: crossing the boundary layer

WHAT. We count how many molecules cross the skin per second per unit area. Call this a flux.

The rule for diffusion across the skin is beautifully simple: the flux is proportional to how big the concentration gap across the skin is.

WHY this shape (linear in the gap). For a thin skin, diffusion gives a straight-line concentration profile, so the gradient is just (gap)/(thickness). That is why flux is simply proportional to the gap — no squares, no exponentials. Think of $h_{g} \approx D / δ$ : diffusivity $D$ over skin thickness $δ$ .

PICTURE. The straight sloped line of concentration inside the skin; its steepness is the flux. Flatten the slope (small gap) → tiny flux.

Step 4 — Flux #2: the surface reaction

WHAT. Once a molecule reaches the surface, chemistry consumes it — it decomposes and its atoms lock into the solid film. How fast? Proportional to how many reactant molecules are present at the surface, i.e. to $C_{s}$ .

WHY proportional to $C_{s}$ (first order). Double the reactant sitting on the surface, double the number of reaction events per second — the simplest and, for these precursors, accurate assumption. This is the "chemistry" leg, and its eagerness $k_{s}$ obeys the Arrhenius Equation (Step 7 uses this).

PICTURE. Molecules snapping onto the lattice; the taller the stack of waiting molecules ( $C_{s}$ ), the faster the snapping.

Step 5 — The balance: what arrives must react

WHAT. In steady operation nothing piles up at the surface — the film grows smoothly, reactant does not accumulate. So the rate molecules arrive across the skin must equal the rate they are consumed by reaction: $F_{1} = F_{2} = F$

WHY. This is pure bookkeeping (conservation). If arrival beat consumption, molecules would stack up forever at the surface (impossible); if consumption beat arrival, the surface would starve. Balance is the only stable option — and it is exactly what lets us solve for the unknown $C_{s}$ .

PICTURE. Two equal arrows meeting at the surface, an "in = out" balance beam.

Now solve. Set the two expressions equal: $h_{g} (C_{g} - C_{s}) = k_{s} C_{s}$ Expand and gather the $C_{s}$ terms: $h_{g} C_{g} = k_{s} C_{s} + h_{g} C_{s} = (k_{s} + h_{g}) C_{s}$ $C_{s} = \frac{h _{g}}{k _{s} + h _{g}} C_{g}$

Read this out loud: the surface concentration is the bulk supply $C_{g}$ shrunk by the fraction $\frac{h _{g}}{k _{s} + h _{g}}$ , which is always between $0$ and $1$ . The surface never gets more than the bulk — comforting sanity check.

Step 6 — Assemble the growth rate

WHAT. Put $C_{s}$ back into either flux (use $F = k_{s} C_{s}$ ): $F = k_{s} \cdot \frac{h _{g}}{k _{s} + h _{g}} C_{g} = \frac{k _{s} h _{g}}{k _{s} + h _{g}} C_{g}$

WHY it looks like resistors. Divide top and bottom by $k_{s} h_{g}$ : $F = \frac{1}{\frac{1}{k _{s}} + \frac{1}{h _{g}}} C_{g}$ The two "difficulties" $\frac{1}{k _{s}}$ (chemistry is sluggish) and $\frac{1}{h _{g}}$ (transport is sluggish) add, exactly like two resistors in series add. The bigger difficulty dominates the sum — the slowest step controls the rate.

From flux to film thickness. Each incorporated molecule adds volume. If the film packs $N$ atoms per unit volume, the surface climbs at

PICTURE. The resistor-ladder cartoon: supply battery $C_{g}$ pushing current $F$ through resistor $1/ h_{g}$ then resistor $1/ k_{s}$ .

Step 7 — The two limits (why heat changes everything)

The whole point of the sum $\frac{1}{k _{s}} + \frac{1}{h _{g}}$ is that one term is usually much bigger, and that term wins.

WHY temperature picks the regime. Heat barely changes how fast gas diffuses ( $h_{g}$ nearly flat with $T$ ), but it makes chemistry explode: $e^{- E_{a} / k T}$ climbs steeply as $T$ rises. So:

Low $T$ → $k_{s}$ tiny → $\frac{1}{k _{s}}$ huge → reaction-limited.
High $T$ → $k_{s}$ enormous → $\frac{1}{h _{g}}$ now the bigger term → transport-limited (rate flattens out).

PICTURE. Growth rate versus $1/ T$ on a log scale: a steep Arrhenius line at low $T$ that bends into a flat shelf at high $T$ . The two regimes are the two pieces of this one curve.

Engineers deliberately sit on the flat, uniform reaction-limited part for even thickness — one reason conformal LPCVD runs at low pressure (low $P$ raises $h_{g}$ , pushing $h_{g} ≫ k_{s}$ ).

Step 8 — Degenerate & edge cases (never left guessing)

Let us stress-test the formula $F = \frac{k _{s} h _{g}}{k _{s} + h _{g}} C_{g}$ at its extremes.

Case	What happens	Formula says	Sensible?
$C_{g} \to 0$ (no reactant)	nothing to deposit	$F \to 0$	✔ film stops
$k_{s} \to 0$ (cold, dead chemistry)	reaction cannot proceed	$F \to 0$	✔ no growth even with plenty of gas
$h_{g} \to 0$ (thick skin, no diffusion)	nothing reaches surface	$F \to 0$	✔ starved surface
$k_{s} \to \infty$ (infinitely eager)	limited only by supply	$F \to h_{g} C_{g}$	✔ pure transport limit
$h_{g} \to \infty$ (no skin at all)	limited only by chemistry	$F \to k_{s} C_{g}$	✔ pure reaction limit
$k_{s} = h_{g}$ (balanced)	both equally slow	$F = \frac{1}{2} k_{s} C_{g}$	✔ half of either single limit

WHY this table matters. Every knob an engineer turns lives somewhere in this table. The formula degrades gracefully to zero when any essential ingredient vanishes, and saturates to the correct single-resistance limit when the other resistance disappears. No physical scenario surprises us.

PICTURE. The flux surface plotted against $k_{s}$ and $h_{g}$ : flat zero along both axes, rising to the two limiting ridges.

The one-picture summary

One frame ties it together: reactant travels down from bulk ( $C_{g}$ ) → crosses the boundary-layer resistor $1/ h_{g}$ → reacts through resistor $1/ k_{s}$ → becomes film growing at $v = F / N$ . The two resistances add; the fatter one rules.

Recall Feynman: tell it to a 12-year-old

Picture a snack bar (the wafer) with a hungry crowd of snack-molecules floating above it. To reach the counter each snack must first squeeze through a slow, stuffy doorway (the boundary layer) — that squeeze has a "difficulty" $1/ h_{g}$ . Once at the counter it still has to be served — the cook's speed is $1/ k_{s}$ . The total wait is the doorway wait plus the serving wait, just like adding two delays in a line. Whichever step is slower decides how fast snacks get eaten, i.e. how fast the film grows. Now heat the kitchen: the cook (chemistry) speeds up wildly with heat, but the doorway width barely changes. So when it's cold, the cook is the bottleneck (reaction-limited, very heat-sensitive); when it's blazing hot, the cook is lightning-fast and the doorway becomes the bottleneck (transport-limited, heat no longer helps). And if there are no snacks at all, or the doorway is bricked shut, or the cook is asleep — nothing gets served. That single sentence, "total wait = doorway wait + serving wait," is the whole formula $F = \frac{k _{s} h _{g}}{k _{s} + h _{g}} C_{g}$ .

Recall Self-test

The two series resistances in CVD ::: transport $1/ h_{g}$ and reaction $1/ k_{s}$ ; they add. Why does low $T$ give reaction-limited growth? ::: $k_{s} = k_{0} e^{- E_{a} / k T}$ shrinks, so $1/ k_{s}$ dominates the sum. What does $F$ approach when $k_{s} \to \infty$ ? ::: $F \to h_{g} C_{g}$ , pure transport limit. Units of $v = F / N$ ::: (atoms/m²·s) ÷ (atoms/m³) = m/s, a surface speed.

Related builds: Epitaxy, Thermal Oxidation, Physical Vapor Deposition (PVD), Thin Film Metrology.