CMP Process Steps: How Chemical Mechanical Planarization Works

公開日: 2026年6月24日ビュー501
← Back to Complete Planarization Guide
CMP Process Engineering

Chemical Mechanical Planarization combines chemical reactivity and mechanical abrasion to remove material from wafer surfaces and achieve global planarization. This guide provides a thorough technical breakdown of every CMP process step — from pre-measurement through post-CMP cleaning — along with equipment anatomy, process variables, endpoint detection methods, and common defect modes.

Updated: June 2026 | By JEEZ Technical Team

01What Makes CMP Unique Among Planarization Techniques

Chemical Mechanical Planarization occupies a category of its own among semiconductor process techniques because it achieves two goals simultaneously that no other process can accomplish together: it removes material at a high, controllable rate (typically 100–600 nm/min depending on film and recipe), and it produces a globally planar surface whose uniformity spans the full 300 mm wafer diameter. This global planarization capability — uniquely provided by CMP among all production-ready techniques — is what makes it indispensable for every advanced-node IC manufacturing flow.

The technique exploits two synergistic mechanisms that operate simultaneously at the polishing interface:

  • Chemical action: The slurry chemistry reacts with the surface of the film being removed, forming a chemically modified surface layer that is softer, more reactive, or more easily dissolved than the underlying bulk material. For oxide CMP, alkaline silica slurry (pH 10–11) breaks surface Si–O–Si bonds. For copper CMP, H₂O₂ oxidizes Cu to Cu₂O. For tungsten CMP, KIO₃ or H₂O₂ forms WO₃. This surface modification is what makes CMP dramatically faster than purely mechanical abrasion.
  • Mechanical action: Nano-scale abrasive particles (20–200 nm) in the slurry are continuously renewed at the wafer–pad interface by pad rotation and groove geometry. They mechanically remove the chemically modified surface layer, exposing fresh material for the next chemical reaction cycle. Neither the chemical nor the mechanical action alone is sufficient — together they achieve material removal rates 10–100× faster than either mechanism independently.
Why synergy matters

Colloidal silica particles alone — pressed against SiO₂ at CMP pressures — remove oxide at <5 nm/min. Alkaline chemistry alone dissolves SiO₂ at <2 nm/min at room temperature. Together in a CMP slurry, they achieve oxide removal rates of 200–500 nm/min — demonstrating the multiplicative effect of the chemo-mechanical mechanism.

02Preston’s Equation: The Governing Physics

The quantitative relationship governing CMP material removal was first described by Frank Preston in 1927, in the context of optical glass polishing. Preston’s equation remains the foundational model for CMP process development and equipment design today:

Preston’s Equation

MRR = Kp × P × V

MRR — Material Removal Rate (nm/min or Å/min)
Kp — Preston coefficient: a material- and slurry-specific constant that captures all chemical, surface energy, and abrasive interaction effects
P — Contact pressure between the pad and wafer surface (psi or kPa)
V — Relative velocity between the wafer and pad surfaces (m/s)

The physical implication of Preston’s equation is profound for planarization: at any point on the wafer, the removal rate is proportional to the local contact pressure. A surface feature that protrudes above the mean plane contacts the polishing pad with higher pressure than a recessed area. Higher pressure → higher MRR → the protrusion is removed faster → the surface converges toward planarity. This self-leveling behavior is the fundamental mechanism that makes CMP uniquely capable of global planarization.

Limitations and Extensions of Preston’s Model

While Preston’s equation accurately captures the first-order behavior of CMP, real processes exhibit deviations that advanced models address:

  • Pattern density dependence: The local contact pressure is influenced by the fraction of the pad surface that is in contact with material vs. exposed trenches (the “pattern density”). The Preston-Noh-Philipossian model extends Preston to account for this effect.
  • Slurry transport limitations: At high pad speeds or low slurry flow rates, fresh slurry may not be efficiently replenished at the wafer center, causing center-edge removal rate asymmetry that deviates from pure Preston behavior.
  • Pad viscoelastic effects: Real polyurethane pads are viscoelastic — their compliance changes with polishing time, temperature, and conditioning history. This introduces time-dependent changes in the effective Preston coefficient during a polishing run.

03CMP Equipment Anatomy

A production CMP system integrates multiple precision-engineered subsystems. Understanding each component’s role is essential for diagnosing process performance and optimizing recipes.

Rotary Platen

A large (600–700 mm diameter) rigid disk — stainless steel, granite, or composite — that holds and rotates the polishing pad. The platen drive controls rotational speed (typically 20–150 rpm) and must maintain extremely tight speed stability (±0.1 rpm) to prevent MRR variation. Platen flatness is critical: run-out >5 µm can introduce systematic within-wafer uniformity errors.

Carrier Head (Wafer Chuck)

Holds the wafer face-down against the rotating pad. Modern carrier heads use a multi-zone pneumatic pressure system with 4–7 independently controlled concentric zones across the wafer diameter, allowing the process engineer to apply differential pressure to compensate for radial removal rate non-uniformity. Retaining rings maintain the wafer’s lateral position while applying adjustable downward force on the pad edge to control edge-exclusion behavior.

Polishing Pad Assembly

Typically a two-layer stack: a hard, microporous polyurethane top layer (IC1000 or equivalent, Shore D ~55–65) affixed over a softer compliant sublayer (SUBA or equivalent). The pad is affixed to the platen via pressure-sensitive adhesive. Groove patterns machined into the pad surface (concentric, radial, X-Y, or spiral) control slurry distribution and prevent hydrodynamic lift-off at high rotation speeds.

Slurry Delivery System

Dispenses slurry at controlled temperature (typically 20–25°C ± 0.5°C), flow rate (100–300 mL/min), and, in some systems, point-of-use mixing ratio (for two-part slurries where oxidizer is added just before dispense). In-line particle size monitoring, flow controllers, and point-of-use filtration (0.2–1 µm) are standard in high-volume manufacturing to detect agglomeration events before they reach the wafer.

Conditioning System

A rotating arm holds a diamond conditioning disk (CVD diamond crystals embedded in a metal matrix, 4–6 inch diameter) that is pressed against the rotating pad surface. The conditioning disk abrading action continuously removes the glazed, polishing-debris-filled top layer of the pad, exposing fresh pore structure and surface asperities. Conditioning sweep rate, disk force (1–8 lbs), and conditioning-to-wafer-polishing ratio (conditioning duty cycle) are key process parameters.

Endpoint Detection Module

Integrated optical windows in the platen allow in-situ reflectometry or interferometry through the pad to the wafer surface. Motor current sensors on the platen and carrier head drives respond to friction changes at the wafer–pad interface. Eddy current sensors (for metal films) detect changes in the electromagnetic properties of the film as thickness decreases toward the target.

04The CMP Process: Step-by-Step

A complete CMP process for a single wafer involves the following sequential steps, each contributing critically to the final planarization outcome:

1

Pre-CMP Incoming Measurement

Before the wafer enters the CMP tool, its incoming film thickness profile is measured across a defined grid of points (typically 49–121 points on a 300 mm wafer) by an in-line reflectometry or ellipsometry metrology tool. This data determines the starting thickness distribution, calculates the required removal depth to reach the target post-CMP thickness, and adjusts the polishing time or pressure profile (via run-to-run control) to compensate for process drift from previous lots.

2

Pad Break-In and Conditioning

A new polishing pad must be “broken in” — conditioned for 20–60 minutes before production use — to stabilize its surface texture and achieve consistent material removal rates. A diamond conditioning disk abrades the pad surface, creating a reproducible population of pad asperities (micro-scale surface protrusions 1–10 µm high) that determine the pad’s MRR delivery. In-situ conditioning continues during production polishing to maintain pad state. Ex-situ conditioning is performed between wafers to reset the pad surface between lots.

3

Slurry Pre-Wetting and Stabilization

The pad surface is pre-wetted with slurry or DIW (deionized water) before the wafer contacts the pad. This ensures the pad pores are filled with slurry before polishing begins, avoiding a dry-contact period at wafer-down that can cause excessive friction, wafer slippage, or initial over-polishing. Pre-wetting time (typically 10–30 seconds) and slurry flow rate during pre-wetting are recipe-controlled parameters.

4

Main Polishing (Bulk Removal)

The carrier head lowers the wafer onto the rotating pad. Down-force (1–6 psi, independently adjustable per zone on multi-zone heads), platen rotational speed (50–120 rpm), carrier rotational speed (30–100 rpm, counter-rotating or co-rotating relative to platen), carrier sweep oscillation amplitude and frequency, and slurry flow rate are all simultaneously controlled to the target recipe values. The Preston-governed removal begins immediately. Duration is set by the run-to-run control system’s calculation from the incoming thickness measurement.

5

Chemical Reaction at the Polishing Interface

Throughout polishing, the slurry chemistry reacts with the wafer surface in a continuously cycling process: chemical species in the slurry modify the surface of the film (forming a softer or more soluble surface compound), abrasive particles mechanically remove the modified surface layer, fresh unmodified material is exposed to further chemical attack, and the cycle repeats. The rate of this cycle — and therefore the MRR — is governed by both the slurry chemistry and the mechanical parameters (P × V in Preston’s equation). The contact time is typically milliseconds per asperity contact event, but the continuous rotation ensures thousands of such contact events per second per unit area.

6

Endpoint Detection and Polish Stop

The CMP tool monitors multiple in-situ signals throughout polishing to detect when the target removal depth has been reached. Optical EPD generates a time-domain signal (reflectance or interference fringe pattern) that changes characteristically as the film thins to the stop layer. Motor current EPD detects the friction change as a harder underlying layer (e.g., Si₃N₄ in STI CMP) is reached. When the endpoint signal matches the programmed threshold, the system triggers a timed over-polish (typically 5–15% of the main polish time) and then lifts the wafer off the pad.

7

Post-Polish Rinse (at Tool)

Immediately after the carrier head lifts from the pad, the wafer is rinsed on the platen with a high-flow DIW stream to displace bulk slurry from the wafer surface. This first rinse removes the majority of slurry particles and prevents them from drying on the wafer surface — dried slurry is significantly more difficult to remove in post-CMP cleaning and is a major source of particle defects. The rinse continues for 15–30 seconds before the wafer is transferred to the post-CMP cleaning section.

8

Post-CMP Cleaning (Cleaner Module)

The wafer is transferred to a dedicated cleaning module — on the same CMP tool platform (integrated) or in a separate cleaning tool. PVA brush scrubbing in dilute chemical solution (citric acid, NH₄OH, or DIW depending on the film system) removes residual particles and metallic contamination. Megasonic cleaning in SC1 or DIW dislodges particles from recessed surface features. Final DIW rinse and spin-dry prepare the wafer for post-CMP metrology and the next process step. See our complete Post-CMP Cleaning guide for full details.

05Key Process Variables and Their Effects

CMP recipe development involves optimizing a multi-parameter space where each variable affects both removal rate and uniformity:

Process Variable典型的な範囲MRRへの影響Effect on Uniformity
Down-force (pressure)1–6 psiDirect (linear per Preston)Higher P → center-heavy profile; use zone control to correct
プラテン速度30–120 rpmDirect (V component)Higher speed → more uniform slurry refresh; edge effects at extremes
キャリア速度20–100 rpmMinor independent effectCounter-rotation improves radial uniformity
Slurry flow rate100–300 mL/minLimited above threshold; below threshold MRR dropsLow flow → center starvation → center-low uniformity
Slurry temperature18–28°C+1–3%/°C for chemical-dominated processesTemperature gradient across wafer → MRR variation
Back pressure0–3 psiReduces effective down-force at wafer centerUsed to reduce center-heavy profiles
Conditioning force1–8 lbsHigher → faster pad wear → higher initial MRRAggressive conditioning may create center-edge pad texture differences

06Endpoint Detection Methods

Accurate, repeatable endpoint detection is one of the most technically demanding aspects of CMP process control. The challenge is to stop polishing at exactly the right film thickness — thin enough to achieve planarization, but without over-polishing into the stop layer or generating excessive dishing. Three primary EPD methods are used in production:

Optical Endpoint Detection (OED)

A laser or broadband light source is directed through a transparent window in the rotating platen and through the polishing pad onto the wafer surface. The reflected optical signal — either monochromatic interference fringes (at a single wavelength) or a spectroscopic reflectance spectrum (at multiple wavelengths) — is analyzed in real time. As the film thins, the optical path length changes, causing the interference pattern to cycle. The number of cycles counted provides a direct measurement of film thickness removed. When the cycle count matches the programmed target, or when the spectrum matches the reference for a “endpoint reached” condition, polishing stops. OED is standard for dielectric CMP (oxide, STI) and copper CMP.

Motor Current / Friction Endpoint Detection

The friction coefficient between the polishing pad and the wafer surface changes when the polishing transitions from one film to another — for example, from tungsten to TiN barrier to SiO₂. Each material has a different coefficient of friction with the pad and slurry combination, producing a characteristic change in the motor torque required to maintain constant rotational speed. The motor current sensors on the platen drive and carrier head drive continuously monitor this torque signature. A friction-based EPD step fires when the current crosses a threshold corresponding to the stop layer being reached. Motor current EPD is particularly effective for STI CMP (oxide → nitride transition) and tungsten CMP (W → TiN/TiW barrier transition).

Eddy Current Endpoint Detection (for Metal Films)

An electromagnetic coil embedded in the platen induces eddy currents in conductive films on the wafer surface. The amplitude and phase of the reflected electromagnetic signal are proportional to the metal film’s thickness. As the copper or tungsten film thins during polishing, the eddy current signal changes in a predictable, calibrated manner. Eddy current EPD provides a continuous, in-situ thickness measurement during metal CMP with sub-nm precision, enabling highly accurate polishing stop control without relying on friction changes that may be less sharply defined in metal-to-metal stack transitions.

Full metrology guide: Post-CMP Cleaning & Planarization Metrology

07Within-Wafer Uniformity Control

Achieving WIWNU below 1% across a 300 mm wafer requires active, multi-parameter process control. The primary tool for uniformity correction is the multi-zone carrier head: by independently adjusting the pneumatic pressure in each of the 4–7 concentric zones (center, inner-middle, middle, outer-middle, edge, retaining ring zone), the process engineer can directly shape the radial removal rate profile across the wafer. A center-heavy removal profile is corrected by reducing center zone pressure; an edge-heavy profile is corrected by reducing edge zone pressure. Modern CMP tools perform automatic zone pressure adjustment based on incoming film thickness maps, implementing closed-loop run-to-run control that maintains WIWNU within specification as consumable aging occurs.

Beyond carrier head zone pressure, uniformity is managed by: (1) slurry temperature uniformity (delivered via temperature-controlled supply lines); (2) platen flatness maintenance (periodic platen re-lapping); (3) pad conditioning uniformity (conditioning sweep profile optimization to avoid center-edge differences in pad texture); and (4) slurry flow rate uniformity (multiple dispense nozzles for large pad diameters).

08Common CMP Defects and Root Causes

表面の傷

Large particles (from slurry agglomeration, pad debris, or contamination) dragged across the wafer surface at the CMP interface. Scratches are the most yield-damaging CMP defect, as they can extend across multiple die and sever multiple metal lines or vias simultaneously. Root causes: slurry agglomeration (inadequate storage or temperature excursions), point-of-use filtration failure, and pad contamination from previous lots.

ディッシング

Over-polishing of metal lines or features relative to the surrounding dielectric in damascene CMP. Dishing appears as a concave depression in wide copper or tungsten features and is caused by the continued polishing of the soft metal after the harder surrounding dielectric reaches the endpoint. Controlled by limiting over-polish time, optimizing slurry chemistry (BTA concentration for copper), and reducing down-force during the final clearing step.

浸食

Pattern-density-dependent thinning of the dielectric in dense metal arrays — the cumulative result of the slurry removing dielectric material from between closely spaced metal lines more rapidly than in isolated regions. Erosion creates a “global” topographic variation at the pattern scale. Managed through CMP recipe optimization, the use of dummy fill features (added in the IC layout to equalize pattern density), and slurry selectivity tuning.

デラミネーション

Separation of a thin film from the layer below it during CMP, driven by shear stress at the wafer–pad interface. Particularly problematic for ultra-low-k (ULK) porous dielectric films (k < 2.3), which have very low cohesive and adhesive strength. Prevented by using reduced down-force, soft (compliant) pad stacks, and optimized slurry pH for ULK-compatible surface chemistry.

Metal Corrosion

Chemical attack of exposed copper, cobalt, or ruthenium surfaces by slurry chemical agents during or after polishing, producing pitting, oxide growth, or galvanic corrosion where dissimilar metals are exposed simultaneously. Controlled by matching oxidizer type and concentration to the specific metal stack and by minimizing wafer exposure time between CMP polishing and post-CMP cleaning.

Residues

Incompletely cleared barrier metal (TaN, TiN) or dielectric material remaining on the wafer surface after polishing — a “clearing failure.” Causes electrical shorts between adjacent copper lines if barrier metal bridges the dielectric between lines. Addressed by extending the clearing over-polish time, increasing barrier removal rate (slurry composition), or reducing the dielectric pattern density to lower the clearing endpoint variability.

Related: CMP Slurry for Semiconductor Planarization — Chemistry, Types & Selection

Optimize Your CMP Process with JEEZ Consumables

JEEZ manufactures CMP polishing slurries, polishing pads, and absorption films engineered for consistent, high-uniformity planarization across oxide, STI, tungsten, and copper CMP applications. Contact our technical team to discuss your process requirements.

Contact JEEZ →

よくあるご質問よくある質問

What are the basic steps of the CMP process?
The CMP process follows eight key steps: (1) pre-CMP incoming film thickness measurement; (2) pad break-in and conditioning; (3) slurry pre-wetting; (4) main polishing with controlled pressure and velocity; (5) chemical reaction at the polishing interface; (6) endpoint detection and polish stop; (7) post-polish rinse on the tool; and (8) post-CMP cleaning in a dedicated cleaning module. Each step contributes to the final planarization quality and must be optimized as part of the overall recipe development process.
What is Preston’s equation and why does it matter for CMP?
Preston’s equation states MRR = Kp × P × V, where MRR is the material removal rate, Kp is the Preston coefficient (material- and slurry-specific), P is the contact pressure, and V is the relative velocity between wafer and pad. It matters for CMP because it explains the self-leveling behavior that produces global planarization: elevated surface features experience higher contact pressure (P), which increases their local MRR, causing them to be removed faster than recessed areas until the surface converges toward planarity.
How does endpoint detection work in CMP?
CMP endpoint detection uses three principal methods: (1) Optical EPD — laser light directed through the platen and pad onto the wafer surface; interference fringe changes track film thickness in real time; (2) Motor current EPD — friction changes as the polishing transitions from one film to another are detected as changes in platen or carrier motor torque; and (3) Eddy current EPD — for conductive films, an electromagnetic coil measures metal film thickness by the eddy current response. All three methods can stop polishing automatically when the target film thickness or stop layer is reached.
What causes scratches in CMP and how are they prevented?
CMP scratches are caused by large particles (slurry agglomerates, pad debris, or contamination) dragged across the wafer surface. Prevention strategies include: strict slurry storage temperature control (to prevent agglomeration); point-of-use filtration (0.2–1 µm in-line filters); pad cleaning between wafers; slurry expiry date management; and regular inspection of slurry supply lines for contamination. Post-CMP optical inspection with bright-field and dark-field imaging is used to detect and monitor scratch defect density as a process health indicator.
What is dishing in copper CMP and how is it controlled?
Dishing in copper CMP refers to the concave depression that forms in wide copper features when the soft copper continues to be polished after the surrounding harder dielectric reaches the endpoint level. It is controlled by: limiting the over-polish time after barrier clearance; using slurry chemistry with benzotriazole (BTA) as a corrosion inhibitor that forms a passivation layer on exposed copper, reducing its MRR relative to the dielectric; reducing down-force during the final clearing step; and using fixed-abrasive pad systems for low-dishing applications such as hybrid bonding preparation.

この記事をシェアする

ご相談とお見積もり

ニュースレターを購読して最新情報を入手