Chemical mechanical planarization (CMP): Question bank

Notation reminder (so nothing here is used before it is meant):

$R R$ = removal rate (thickness removed per second).
$P$ = pressure (force pressing wafer onto pad, per unit area).
$v$ = relative sliding speed between pad and wafer.
$k_{p}$ = Preston coefficient (a single number that lumps in chemistry, abrasive, pad, temperature).
Preston's law: $R R = k_{p} P v$ .

True or false — justify

Slurry chemistry alone (no abrasive, no pad motion) can planarize a wafer.

False. Chemical etching is isotropic — it attacks valleys as fast as hills, so it removes material everywhere equally and never flattens; it can only thin, never planarize.

Grinding harder (more mechanical force, no chemistry) gives a flatter surface faster.

False. Pure mechanics scratches deeply and damages the crystal lattice; without chemistry there is no soft removable layer limited to the high points, so you get damage instead of planarity.

Preston's law says removal rate is proportional to pressure, so doubling

P

is always the best way to speed up.

False as advice.

R R \propto P

is true, but higher

P

makes the pad bend into wide features (dishing/erosion) and causes scratches — faster is not flatter.

On a wafer spinning against a pad, the wafer's outer edge always polishes faster than its center.

False when platen and carrier speeds are matched (

ω_{p} = ω_{w} = ω

). Then the wafer-radius terms cancel and every point sees

v = ω d

, uniform across the wafer.

CMP is a purely subtractive step — it never adds material.

True. It only removes the top layer; the chemistry softens and the abrasive wipes away. Any "added" flatness is the result of selective removal of the peaks, not deposition.

A higher Preston coefficient

k_{p}

means you must press harder to get the same rate.

False.

k_{p}

multiplies

P

and

v

, so a larger

k_{p}

(e.g. a better oxidizer) gives more rate at the same

P

and

v

— that is gentler, cleaner polishing.

Dishing and erosion are the same defect described two ways.

False. Dishing is a single soft/wide filled feature sinking below its surroundings; erosion is the whole oxide level dropping in a dense array of many closely-packed metal lines. Different geometries, different causes.

High selectivity between two materials gives you a natural stopping point.

True. If

R R_{film} ≫ R R_{underlayer}

, removal nearly halts once you break through — the slow-etching layer acts as a built-in brake, easing endpoint control.

Spot the error

"CMP works because the pad presses equally on hills and valleys, so both get polished."

Wrong: the pad contacts the high points first. Valleys are barely touched, keep their unreacted (hard) surface, and survive — that selective contact is exactly what produces flatness.

"To planarize globally, use a very soft, compliant pad so it conforms to the whole wafer."

Wrong: a soft pad conforms to features and bends into wide trenches, giving local planarity and dishing. A stiffer pad bridges over valleys and touches only peaks → global planarity.

"Preston's equation

R R = k_{p} P v

already accounts for the chemistry, so slurry choice doesn't change the rate."

Wrong: the chemistry is precisely what is hidden inside

k_{p}

. Change the oxidizer or pH and

k_{p}

changes — slurry choice moves the rate directly.

"Since etching is isotropic, adding a strong etchant to the slurry is the fastest path to flatness."

Wrong: a stronger etchant attacks valleys too, degrading planarity. The chemistry must only soften the surface so the pad's mechanical contact at the peaks decides where removal happens.

"We convert rpm to rad/s just to make the number look scientific."

Wrong:

v = ω d

needs

ω

in rad/s so that

v

comes out in m/s, matching Preston's SI units. Feeding rpm straight in gives a wrong

R R

"Endpoint detection means measuring the total time and stopping at the calculated value."

Wrong: real removal rate drifts (pad wear, slurry aging). Endpoint uses a live signal — motor current/friction change or optical reflectance — that senses breaking through a layer, not a pre-computed clock.

Why questions

Why must CMP combine chemistry AND mechanics rather than either alone?

Chemistry alone is isotropic (etches everywhere, can't flatten); mechanics alone scratches and damages. Together, chemistry makes a soft layer only where the pad presses = the peaks, so peaks vanish while valleys stay — that combination is what planarizes.

Why do CMP tools rotate the carrier and platen at the same speed?

With

ω_{p} = ω_{w} = ω

the wafer-radius terms cancel, giving

∣ v ∣ = ω d

everywhere. Uniform

v

⇒ uniform

R R

by Preston ⇒ even, flat removal across the whole wafer.

Why is a small depth of focus in Photolithography the reason CMP exists?

The lens can only image sharply within a thin in-focus band. Surface hills and valleys push features outside that band → blur. CMP resets the surface to flat so the entire pattern stays in focus.

Why does the Damascene Process absolutely require planarization?

You deposit copper everywhere, then must remove the overburden so metal remains only inside the trenches. Without flattening, leftover metal bridges trenches and shorts adjacent lines together.

Why does Preston's law resemble the Archard Wear Law?

Both say worn volume scales with load × sliding distance. Dividing by area turns load into pressure

P

, and differentiating in time turns sliding distance into velocity

v

— giving

R R = k_{p} P v

Why is raising

v

often safer than raising

P

to speed up removal?

Higher

P

raises contact stress → more scratches and dishing; higher

v

keeps contact stress the same, only risking slurry starvation. Preston treats them symmetrically, but defect physics does not.

Why is

k_{p}

called a "lumped" coefficient?

It packages every effect not written explicitly — abrasive size, pad texture, temperature, and especially slurry chemistry — into one measurable constant, so the law stays simple:

R R = k_{p} P v

Edge cases

If the relative velocity

v

is exactly zero (pad and wafer momentarily not sliding), what does Preston predict?

R R = k_{p} P \cdot 0 = 0

. No sliding means no abrasive digging events, so no removal — pressure alone with zero motion polishes nothing.

If the offset

d

between wafer center and pad center were zero (concentric), what happens under matched speeds?

v = ω d = 0

, so

R R = 0

everywhere — no net sliding. Real tools keep

d > 0

(and often oscillate) precisely to guarantee sliding and thus removal.

What happens to planarity if a very wide, soft copear feature is polished with a compliant pad?

The pad bends into the wide soft region and over-polishes its center → the feature dishes below the surrounding oxide. This is why wide metal lines need pattern-density control or dummy fill.

At the instant you break through one layer into a much slower-polishing layer, what does the removal rate do?

It drops sharply because the underlayer has a lower effective

k_{p}

. This abrupt change in friction/reflectance is the very signal endpoint detection watches for.

If the slurry runs dry (starvation) at high

v

, why does removal become non-uniform even though

v

is uniform?

With no slurry film, abrasive and chemistry can't reach the surface evenly, so effective

k_{p}

collapses locally. Preston assumes a healthy slurry supply; starve it and the "constant"

k_{p}

is no longer constant across the wafer.

If pressure

P

is raised until the pad fully conforms to every hill and valley, can you still planarize?

No. A fully conforming pad touches valleys and hills alike, so removal becomes uniform everywhere — the height difference is preserved. Planarization needs the pad to touch peaks preferentially, which fails at extreme

P

Recall One-line summary of the traps

Flatness comes from selective peak contact, not from more force or more chemistry alone; $k_{p}$ hides the chemistry; matched speeds make $v$ uniform; and every "faster" knob ( $P$ , $v$ ) buys defects if pushed too far.