In short: CMP removal rate and defectivity are governed by the slurry (abrasive type/size, pH, and oxidizer chemistry) working with a conditioned polyurethane pad — Preston's equation ties rate to pressure and velocity, while pad asperities, abrasive loading, and conditioning discipline decide whether you planarize cleanly or leave scratches and residue.
Chemical Mechanical Planarization (CMP) is where chemistry and mechanics are deliberately balanced to shear material off a wafer one atomic layer at a time. Two consumables dominate the outcome: the slurry and the polishing pad. Get their interaction right and you planarize a copper or oxide surface to sub-nanometer non-uniformity; get it wrong and you produce dishing, erosion, micro-scratches, and particle residue that kill yield at the next lithography step. This article digs into how abrasives and chemistry, pad conditioning, removal rate, and defectivity actually work — the "why" behind the recipes.
The slurry: abrasives plus active chemistry
A CMP slurry is a colloidal suspension carrying two jobs at once. The abrasive provides the mechanical component; the chemistry (oxidizers, complexing agents, pH buffers, corrosion inhibitors, surfactants) softens or reacts with the top of the film so the abrasive can sweep away a weakened layer rather than gouging the bulk. This cooperative action is the whole point of the word "chemical" in CMP.
- Silica (SiO₂) — the workhorse for oxide (ILD/STI) polishing, typically 30–90 nm colloidal particles at 10–30 wt% solids, run at high pH (10–11) where silica dissolution and the Si–O–Si network interplay drive removal.
- Ceria (CeO₂) — used for STI and shallow oxide because it chemically bonds to silica ("chemical tooth"), giving high oxide:nitride selectivity with far lower solids loading (often <1 wt%).
- Alumina (Al₂O₃) — harder, historically used for tungsten and metal CMP, but its angular morphology raises scratch risk.
For copper CMP, the chemistry carries most of the load: an oxidizer such as hydrogen peroxide (H₂O₂) grows a thin Cu-oxide, a complexing agent (glycine, citric acid) and a corrosion inhibitor (benzotriazole, BTA) protect recessed copper, and the abrasive removes only the reacted film from the high points. This is why copper polishes with far gentler mechanics than oxide.
Key fact: Abrasive particle-size distribution matters more than mean size. A slurry with a well-controlled 70 nm mean but a long tail of 0.5 µm agglomerates will scratch; most defect excursions in a fab are traced to slurry filtration failures or large-particle counts (LPC), not to the nominal recipe.
Removal rate and Preston's equation
The first-order model every process engineer memorizes is Preston's equation:
RR = Kp · P · V
where RR is the material removal rate, P is the down-force pressure, V is the relative pad-wafer velocity, and Kp (the Preston coefficient) absorbs everything else — abrasive hardness and concentration, chemistry, pad modulus, and temperature. A typical oxide process might run P ≈ 3–5 psi (20–35 kPa), V ≈ 0.5–1 m/s, delivering removal rates of 2,000–4,000 Å/min. Copper rates can exceed 6,000 Å/min because the oxidizer does so much of the work.
Preston is linear and intentionally simple; real behavior deviates. At very low pressure there is a threshold below which no removal occurs (the Kelvin/adhesion regime), and at high temperature the chemical reaction accelerates, so frictional heating couples pressure and rate non-linearly. Engineers therefore treat Kp as a locally-fit constant and rely on endpoint detection — optical interferometry or motor-current/friction signatures — rather than timing a blind polish.
| Film | Abrasive | pH | Key chemistry | Typical RR |
|---|---|---|---|---|
| ILD oxide | Colloidal silica | 10.5–11 | KOH buffer | 2,500–3,500 Å/min |
| STI oxide | Ceria | 6–8 | Anionic surfactant (selectivity) | 1,500–3,000 Å/min |
| Copper | Low silica | 4–7 | H₂O₂ + glycine + BTA | 4,000–7,000 Å/min |
| Tungsten | Silica/alumina | 2–4 | Fe-based oxidizer | 3,000–6,000 Å/min |
The pad and why conditioning is non-negotiable
The pad is almost always cast or impregnated polyurethane (the IC1000 family being the industry reference), often stacked on a soft subpad for pressure uniformity. Its porous, microtextured surface holds slurry in asperity valleys and transmits down-force through asperity peaks. Two things degrade it during use: asperities plastically flatten and glaze, and pores clog with polish debris and spent slurry. A glazed pad loses removal rate and planarization efficiency within minutes.
Conditioning fixes this. A diamond-grit conditioning disk (diamonds of ~75–150 µm embedded in nickel) is swept across the pad, micro-cutting fresh asperities and re-opening pores. Fabs run it either in-situ (during polish) or ex-situ (between wafers). The conditioner sweep profile directly controls pad wear uniformity, so a worn or improperly-dressed disk shows up as a cross-wafer removal-rate signature. Pads have a finite life (hundreds to a couple thousand wafers) tracked by cumulative conditioning and thickness.
Defectivity: scratches, dishing, erosion, residue
CMP defects are the single largest consumable-driven yield loss in the back end. The major modes:
- Micro-scratches — from oversized abrasive agglomerates, hard foreign particles, or a shedding conditioner. Mitigated by point-of-use (POU) slurry filtration (sub-micron), LPC monitoring, and softer abrasives.
- Dishing — recessing of wide metal features because the soft pad bends into large open areas. Controlled by inhibitor chemistry (BTA) and by dummy-fill design that keeps metal density uniform.
- Erosion — thinning of dense arrays of lines plus their dielectric. Managed by slurry selectivity and reduced over-polish.
- Organic/particle residue — spent slurry and inhibitor films left behind; removed by post-CMP brush/megasonic cleaning, not by the polish itself.
Dishing and erosion are strongly layout-dependent, which is why they are modeled during physical design. Density-aware fill and CMP-hotspot checking in tools such as Siemens Calibre and Synopsys IC Validator flag regions likely to dish or erode before tape-out, and foundry CMP density rules exist precisely to bound these effects.
Key fact: Lowering down-force to cut scratches also cuts removal rate (Preston), which lengthens polish time and can worsen dishing through over-polish. CMP optimization is always a multi-objective trade-off — there is no free "cleaner and faster" knob.
Putting it together on the tool
A production CMP recipe is co-designed: slurry flow (150–250 mL/min) must keep the pad wetted and swept clean of debris; platen and carrier speeds set V; the carrier membrane applies zoned pressure to flatten within-wafer non-uniformity; and conditioning keeps Kp stable wafer-to-wafer. Statistical process control watches removal rate, within-wafer non-uniformity (WIWNU, often targeted <3%), and defect counts, with consumable lots (slurry and pad) qualified before release. The best CMP engineers think of the slurry and pad as a single coupled system, not two parts to be optimized separately. For a broader walk-through of the flow context, see our companion piece on the CMP process, and for the metrology side read wafer defect inspection.
If you are building these skills for a physical-design or process-integration role, structured training accelerates the jump from theory to tool recipes — our CourseTron course catalog covers CMP alongside the full nanofab flow, and design-services teams at Avecas (VLSI design services) apply exactly these CMP density and hotspot rules in real tape-outs.
Ready to master CMP from slurry chemistry to CMP-aware physical design? Enroll in CourseTron's Nano Fab track and work through real removal-rate, defectivity, and density-fill case studies with industry tools. Browse the Nano Fab courses →
Frequently Asked Questions
Why is it called "chemical mechanical" planarization?
Because material is removed by two cooperating actions: chemistry (oxidizers, complexing agents, pH) that reacts with or softens the top of the film, and mechanical abrasion by particles pressed through the pad. Neither alone gives the planarity and rate the combination does.
What limits the CMP removal rate?
To first order, Preston's equation says rate scales with pressure times velocity. In practice it is bounded by scratch/defect limits at high pressure, slurry replenishment at high velocity, chemistry reaction kinetics, and pad condition. You cannot raise rate indefinitely without defectivity or uniformity penalties.
How often is the pad conditioned?
Continuously (in-situ) or between every wafer (ex-situ), depending on the process. Conditioning restores asperities and re-opens clogged pores so the Preston coefficient stays stable; without it, rate and planarization efficiency fall within minutes of polishing.
What causes CMP micro-scratches?
Almost always oversized particles — slurry agglomerates, foreign contamination, or grit shed from a worn conditioning disk — rather than the nominal abrasive size. Point-of-use filtration and large-particle-count monitoring are the primary defenses.