How CMP Machines Work: Process Mechanics, Forces & Planarization Physics

Why CMP Needs Both Chemistry and Mechanics

Material removal processes can be classified along a spectrum from purely mechanical to purely chemical. Pure mechanical abrasion — grinding, lapping — removes material quickly but leaves deep subsurface crystal lattice damage, microscopic cracks, and rough surface topography that propagate defects into the device layers above. Pure chemical etching — wet or dry — removes material with minimal physical damage but cannot achieve planarization on its own, since chemical etch rates are largely insensitive to surface topography and tend to replicate the existing surface profile rather than flatten it.

CMP machines occupy the productive middle ground between these two extremes by combining both mechanisms in a single, continuous process. The chemical component of the slurry reacts with the wafer surface to form a softened reaction layer — far easier to remove than the unreacted bulk material beneath it. The mechanical component, delivered by abrasive particles under controlled pressure and relative velocity, then removes this softened layer preferentially from the raised, high-contact-pressure regions of the wafer topography, while leaving recessed, low-contact-pressure regions comparatively untouched. This pressure-dependent removal selectivity is the physical mechanism that produces planarization rather than uniform etching: high points are removed faster than low points until the surface flattens.

For a foundational overview of what a CMP machine is and where it fits in the broader semiconductor process flow, see our complete guide: CMP Machines: The Complete Guide to Chemical Mechanical Planarization Equipment.

Preston’s Equation: The Governing Removal Rate Model

The foundational mathematical model for CMP removal rate originates from Franklin Preston’s 1927 work on glass polishing, adapted directly to semiconductor CMP in the early 1990s. Preston’s Equation expresses the instantaneous material removal rate as a linear function of two mechanical variables: applied pressure and relative sliding velocity.

The Preston coefficient (K_p) is not a universal constant — it is an empirically determined value specific to a given combination of slurry chemistry, pad material, and wafer surface material. K_p absorbs all of the chemical reactivity, particle hardness, particle concentration, and pad mechanical properties into a single proportionality factor. In practice, process engineers determine K_p experimentally for each consumable set and process condition through blanket wafer removal rate characterization, then use it to predict removal rate response when pressure or velocity is adjusted.

Where Preston’s Equation Breaks Down

While Preston’s Equation provides a useful first-order model, real CMP processes deviate from strict linearity in several well-documented ways that every process engineer must account for:

• Low-pressure threshold effects: Below a critical pressure threshold, removal rate drops faster than linear prediction because asperity contact area — and therefore actual mechanical engagement — falls below the level needed to sustain consistent chemical-mechanical synergy.
• High-velocity saturation: At very high relative velocities, hydrodynamic lubrication effects can partially separate the pad from the wafer surface with a thicker slurry film, reducing direct mechanical contact and causing removal rate to plateau rather than continue increasing linearly.
• Pattern density dependence (the “loading effect”): Preston’s Equation was derived for blanket, unpatterned films. On patterned wafers, the local contact pressure experienced by raised features depends on the fraction of the die area occupied by raised structures (pattern density). Dense feature regions experience lower effective local pressure (the load is distributed across more contact area) and therefore polish more slowly than isolated, low-density features — a phenomenon with no representation in the basic Preston model and a primary cause of within-die non-uniformity.
• Edge effects: Near the wafer edge, the carrier head’s retaining ring and the discontinuity in mechanical boundary conditions create localized pressure distortions that cause removal rate to deviate from the Preston prediction within approximately 3–5mm of the wafer perimeter.

Advanced CMP modeling extends Preston’s basic framework with pattern-density-dependent effective pressure terms, time-dependent K_p evolution (accounting for pad conditioning state and slurry depletion), and multi-asperity contact mechanics models that more accurately predict real production removal rate behavior.

Forces at the Wafer-Pad Interface

The force balance acting on a wafer during CMP determines both the bulk removal rate and the local removal rate distribution that governs uniformity. Three categories of force interact at the polishing interface:

Normal Force (Down-Pressure)

Applied by the carrier head’s pneumatic membrane system through the backing film, normal force presses the wafer against the polishing pad. Production CMP processes typically apply 0.5 to 7 psi of down-pressure depending on the application — tungsten CMP and other high-removal-rate bulk steps operate toward the higher end of this range, while delicate final-polish and low-k dielectric steps use minimal pressure to avoid film damage. Modern multi-zone carrier heads independently control pressure across 5 to 7 concentric radial zones, allowing the normal force profile to be actively shaped to compensate for systematic radial removal rate non-uniformities.

Shear Force and Friction

As the platen and carrier head rotate, relative sliding motion at the wafer-pad interface generates a friction force opposing that motion. This shear force is what mechanically dislodges the chemically softened reaction layer from the wafer surface — it is the direct mechanism of material removal, not merely a byproduct of it. The magnitude of this friction force is sensitive to the lubrication regime at the interface (discussed below), the abrasive particle concentration and hardness, and the real area of asperity contact between the pad surface texture and the wafer.

Hertzian Contact at the Asperity Scale

At the microscopic scale, the polishing pad surface is not perfectly smooth — it consists of a distribution of microscopic asperities (peaks) separated by valleys that channel slurry flow. The wafer does not contact the pad uniformly across its full area; rather, contact occurs at a population of discrete asperity tips, each behaving according to Hertzian elastic contact mechanics. The real contact area — the sum of all individual asperity contact zones — is typically only a small fraction of the nominal (apparent) wafer area, meaning the actual local pressure at each asperity contact point is substantially higher than the nominal pressure calculated by dividing carrier head force by wafer area. This asperity-scale pressure concentration is what makes CMP mechanically effective at relatively modest nominal pressures.

Tribology of the Slurry-Pad-Wafer System

CMP is fundamentally a tribological process — the study of interacting surfaces in relative motion, including friction, wear, and lubrication. The lubrication regime at the wafer-pad interface, governed by slurry film thickness relative to pad asperity height, determines whether the process behaves more like a lubricated bearing (minimal mechanical contact) or a dry abrasive contact (maximal mechanical contact).

The Stribeck Curve and CMP Lubrication Regimes

The Stribeck curve, a classical tribology framework relating friction coefficient to a dimensionless lubrication parameter (combining viscosity, velocity, and pressure), maps directly onto CMP operating conditions:

Understanding which lubrication regime a given recipe operates in is essential for predicting how the process will respond to slurry flow rate changes, pad wear over its service life, and platen temperature drift — all of which alter the effective viscosity and film thickness at the interface.

Surface Chemistry: How Different Materials Are Removed

The specific chemical reaction mechanism by which the slurry softens the wafer surface differs fundamentally by material system. Understanding these mechanisms explains why each CMP application requires a distinctly formulated slurry chemistry.

Silicon Dioxide (Oxide) CMP Chemistry

Oxide CMP proceeds primarily through a hydrolysis mechanism at high pH (typically pH 10–11). Hydroxide ions attack strained Si-O-Si bonds at the silica surface, forming silanol (Si-OH) groups that weaken the surface network and create a softened hydrated layer more easily removed by mechanical abrasion. The silica abrasive particles in the slurry are themselves chemically similar to the film being polished, contributing to a self-sharpening removal mechanism sometimes described as a “chemical tooth” effect.

Copper CMP Chemistry

Copper CMP relies on an oxidation-dissolution-passivation cycle. An oxidizer (commonly hydrogen peroxide) rapidly oxidizes the copper surface to form a thin copper oxide layer. This oxide layer is then mechanically removed by abrasive action, exposing fresh copper for the next oxidation cycle. Critically, a corrosion inhibitor — most commonly benzotriazole (BTA) — forms a passivating complex on the copper surface in non-contact (recessed, low-pressure) regions, protecting them from continued chemical attack while contact regions continue to be removed by the mechanical action. This selective passivation is precisely what enables copper CMP to achieve planarization without excessive dishing or galvanic corrosion in fine interconnect structures.

Tungsten CMP Chemistry

Tungsten CMP also uses an oxidation-based mechanism, typically with strong oxidizers (potassium iodate, hydrogen peroxide, or ferric nitrate-based systems) that form a tungsten oxide surface layer subsequently removed mechanically. Because tungsten is mechanically harder and chemically less reactive than copper, tungsten CMP generally requires more aggressive abrasive content and higher applied pressure than copper CMP, contributing to its position as one of the highest-force CMP applications in a standard process flow.

Ceria-Based STI CMP Chemistry

STI CMP’s requirement for extremely high oxide-to-nitride selectivity is achieved through the unique chemistry of cerium oxide (CeO₂) abrasive particles, which form strong chemical bonds with the silicon dioxide surface through a ligand-exchange mechanism not significantly active with silicon nitride. This chemistry-driven selectivity — rather than purely mechanical hardness difference — is why ceria slurries achieve selectivity ratios that silica-based slurries cannot approach, regardless of mechanical parameter tuning.

Dual Rotation Kinematics and Uniformity

A subtle but critical aspect of CMP machine design addresses a geometric problem inherent to rotational polishing: if a wafer were polished against a pad using pure rotation with no additional motion, points at different radii on the wafer would experience different linear sliding velocities relative to the pad — points near the wafer center would experience near-zero relative velocity, while points near the wafer edge would experience high relative velocity, since linear velocity scales with radius at constant angular velocity (v = ωr).

By Preston’s Equation, this radial velocity gradient would translate directly into a removal rate gradient — heavy removal at the wafer edge and minimal removal at the wafer center — producing severe, uncorrectable bowl-shaped non-uniformity. CMP machines solve this problem through dual independent rotation: both the platen and the carrier head rotate simultaneously, typically at similar but not identical speeds. This dual-rotation kinematic scheme causes every point on the wafer surface to trace a complex, continuously varying relative path across the pad surface over time, averaging out the radial velocity gradient that pure single-axis rotation would produce.

The precise relationship between platen speed and carrier head speed — and any deliberate offset between them — is itself a tunable process parameter that process engineers use to fine-tune the residual removal rate uniformity profile, often in combination with multi-zone carrier head pressure adjustment, to achieve the sub-2% WIWNU specifications required at advanced process nodes.

Process Variable Summary

The table below summarizes the primary mechanical and chemical process variables governing CMP machine behavior, their typical production ranges, and their first-order effect on process outcomes.

Each of these variables is in turn influenced by the physical and chemical specification of the polishing slurry and pad — meaning that consumable selection effectively sets the baseline process window before any recipe-level optimization begins.

This article is part of the JEEZ CMP knowledge base. For the complete equipment overview, components breakdown, manufacturer landscape, and consumable selection guidance, return to our pillar guide: CMP Machines: The Complete Guide to Chemical Mechanical Planarization Equipment.

Häufig gestellte Fragen