Chemical mechanical planarization (CMP): Visual walkthrough

Before any symbol appears, here is the whole cast of characters, in plain words:

Step 1 — One grain, one scratch

WHY. Everything CMP does is millions of these tiny grooves added up. If we understand one groove, multiplying by "how many grains" and "how often" gives the whole story. Starting from the smallest event is the honest way to build a law — no hand-waving.

PICTURE. In the figure, the amber grain sits under a downward push (the load, a force we'll call $F$ ) and moves right by a sliding distance we'll call $s$ . The pink groove behind it is the removed volume.

This proportionality is exactly the Archard Wear Law: worn volume rises with load times distance. We are not inventing physics — we are borrowing the wear law and pointing it at a wafer.

Step 2 — From "force" to "pressure"

WHY. Two wafers can feel the "same push" but if one is bigger, that push spreads over more area and each grain feels less. What each grain actually experiences is force divided by area. So we rewrite the wear law using the quantity every grain really cares about: pressure.

PICTURE. The figure shows the same total force $F$ spread over the contact area $A$ . The stack of little downward arrows is the pressure $P$ — force shared evenly across the face.

Step 3 — From "distance" to "rate": the slide never stops

WHY do we differentiate? We want a rate — nanometres removed per second — because the machine runs for a set time and we must predict when to stop. To turn a total ( $Δ h$ vs $s$ ) into a per-second flow, we ask how fast each grows. The mathematical tool for "how fast is this changing" is the derivative $\frac{d}{d t}$ — nothing else answers "instantaneous rate of change". We choose it precisely because sliding distance keeps piling up smoothly with time.

PICTURE. The figure tracks removed thickness climbing straight upward as time ticks — a constant slope. That slope is the removal rate. The sliding distance $s$ grows as $s = v t$ , so its per-second growth is the velocity $v$ .

Step 4 — Everything we ignored: the coefficient $k_{p}$

WHY one lumped constant? Because modelling every grain and every chemical reaction is hopeless. Physics-from-zero doesn't mean physics-with-infinite-terms; it means being honest about what's inside the box. That box has a name: the Preston coefficient $k_{p}$ . It converts the proportionality into an equality.

PICTURE. The figure shows the " $\propto$ " gap being filled by an amber box labelled $k_{p}$ , with arrows feeding into it: chemistry, abrasive, pad, temperature. Everything unmodelled flows into that one number.

Recall Why

$k_{p}$ carries the chemistry Question: A new slurry doubles the removal rate at the same pressure and speed. Which symbol changed, and what does that tell us physically? Answer ::: Only $k_{p}$ can change (since $P$ and $v$ are fixed), so $k_{p}$ doubled — the chemistry softened the surface faster, letting grains scrape more per pass with no extra mechanical force.

Step 5 — Why every point on the wafer sees the same speed

WHY the cross product $\overset{z}{^} \times r$ ? When a disc spins, a point at position $r$ from the centre moves sideways to $r$ , at speed proportional to how far out it is. The tool that turns "position + spin axis" into "sideways velocity" is the cross product: $v = ω (\overset{z}{^} \times r)$ . We pick it because no simpler operation gives both the right direction (perpendicular) and the right size ( $ω r$ ) at once.

PICTURE. Two overlapping spinning discs. The wafer's centre is offset from the pad's centre by $d$ . A test point on the wafer has position $r$ from the wafer centre, so position $d + r$ from the pad centre. Watch the $r$ contributions cancel.

Step 6 — The edge cases you must check

WHY. A reader should never meet a situation the derivation didn't cover. So we deliberately dial each input to zero or to a mismatch and confirm the formula predicts the right physical thing.

PICTURE. Four mini-panels: (a) $P = 0$ , (b) $v = 0$ , (c) offset $d = 0$ , (d) mismatched speeds $ω_{p} \neq = ω_{w}$ .

Step 7 — Putting real numbers in

WHY. A formula you can't compute with is decoration. This is the same worked example from the parent note, now grounded in the pictures above.

PICTURE. A thickness-vs-time line falling from $800 nm$ overburden down to zero at a constant slope $R R$ ; the crossing time is the answer.

The one-picture summary

This final blueprint compresses the whole chain: one grain (load $F$ , slide $s$ ) → divide by area to get pressure $P$ → differentiate in time to get velocity $v$ → bundle the unknowns into $k_{p}$ → arrive at $R R = k_{p} P v$ , with the matched-spin inset showing $∣ v ∣ = ω d$ making removal uniform.

Recall Feynman retelling — the walkthrough in plain words

Picture a single grain of sand pressed on soft clay and dragged: it carves a groove. Bigger push or longer drag → bigger groove — that's the ancient wear rule. But a wafer isn't one grain; it's a whole face, so what matters is the squeeze per patch of face, which we call pressure. And nothing slides a fixed distance — it spins forever — so instead of "how far" we ask "how far each second", which is just speed. Multiply squeeze by speed and you've got how fast the surface vanishes. There's a bunch of messy stuff we refused to model — how sharp the grains are, how the slurry chemistry softens the top — so we sweep it all into one fudge number $k_{p}$ . Out pops $R R = k_{p} P v$ . Then the neat trick: spin the pad and the wafer at the same rate, mounted a little off-centre, and every spot on the wafer slides at exactly the same speed $ω d$ — so it flattens evenly instead of wearing thin at the edges. Set the push to zero, or the sliding to zero, or park the wafer dead-centre, and the removal drops to nothing — exactly as common sense demands.