Oxide CMP Process Parameters: MRR, Within-Wafer Uniformity & Endpoint Detection Guide
Oxide CMP process performance — material removal rate, within-wafer uniformity, surface finish, and defect density — is determined by the combined effect of slurry chemistry, tool hardware parameters, and polishing consumable condition. This guide provides a comprehensive, quantitative treatment of all key process parameters for oxide CMP, including MRR fundamentals, WIWNU management, endpoint detection methodologies, slurry delivery, and pad conditioning. For a broad introduction to oxide CMP slurry types, see our Oxide CMP Slurry: Complete Technical Guide.
Why Process Parameters Matter in Oxide CMP
Oxide CMP is deceptively complex. On the surface, it appears to be a straightforward mechanical process: apply slurry, press wafer against rotating pad, remove oxide. In reality, oxide CMP is a tightly coupled chemical-mechanical system in which a dozen interdependent parameters — each individually controllable but collectively non-linear in their interactions — determine the process outcome across all relevant quality dimensions simultaneously.
A process engineer optimizing oxide CMP for a new device node cannot simply increase pressure to get more MRR: higher pressure also increases defect risk and degrades uniformity. Cannot increase slurry pH to get faster oxide dissolution: pH drift affects uniformity and can contaminate gate dielectrics. Cannot change the pad without re-qualifying every uniformity-critical parameter. The interactions are real, significant, and must be understood quantitatively to optimize the process. This article provides the foundation for that understanding.
Material Removal Rate (MRR)
Preston’s Equation and Its Practical Limits
The widely-used starting point for oxide CMP MRR modeling is Preston’s equation:
MRR = Kp × P × V
where Kp is the Preston coefficient (material- and process-specific, units of nm/Pa·m), P is the applied pressure at the wafer-pad interface (Pa or psi), and V is the relative velocity between wafer surface and pad surface (m/s or rpm-derived). In practice, the linear Preston relationship holds only within a limited process window — at very low pressures (below ~0.5 psi for oxide CMP), removal becomes sub-linear because insufficient mechanical energy is available to displace the chemically softened surface material. At very high pressures (above ~4 psi), removal becomes super-linear and micro-scratch density increases rapidly as abrasive contact becomes damaging.
Factors Beyond Preston’s Equation
Real oxide CMP MRR depends on many factors that the simple Preston model does not capture:
- Abrasive concentration and PSD: MRR is approximately linear with abrasive concentration up to a saturation point (~12 wt% for colloidal silica), beyond which additional abrasive provides no MRR benefit. Broader PSD (higher D99) generally increases MRR but at a cost of higher scratch risk.
- Slurry pH: For colloidal silica ILD slurry, MRR increases approximately 1.5–2× as pH increases from 9 to 11, due to the pH dependence of SiO2 surface hydrolysis kinetics. A pH drift of –0.5 units (e.g., from pH 10.5 to 10.0) can reduce MRR by 10–15%.
- Temperature: Oxide removal rate increases with temperature (typically 1–2% per °C above 20°C) due to accelerated chemical dissolution. Fabs with poor thermal control in the polishing enclosure may see MRR variation of ±10% across a thermal cycle, manifesting as polish-time drift over the working day.
- Pad texture and age: MRR is highest on a well-conditioned pad with abundant open micro-texture for slurry hydrodynamic transport. As a pad glazes (loses micro-texture) between conditioning intervals, MRR decays — typically 10–25% over a conditioning interval on a production oxide CMP recipe.
- Slurry flow rate up to the saturation point: Below the slurry flow saturation point (typically 150–200 mL/min for a 300 mm ILD process), MRR is flow-limited. Above saturation, additional flow does not increase MRR but reduces thermal effects by better cooling the pad surface.
Within-Wafer Non-Uniformity (WIWNU)
WIWNU is defined as the standard deviation of remaining film thickness at all measurement sites across the wafer, expressed as a percentage of the mean remaining thickness:
WIWNU (%) = σ(thickness) / μ(thickness) × 100
For production ILD oxide CMP at advanced nodes, WIWNU <3% (1σ) at 49-site measurement is the minimum acceptable. Sub-14 nm node applications require WIWNU <1.5%. STI CMP targets WIWNU <1% (1σ) for FinFET fin height uniformity control.
Root Causes of WIWNU
Carrier head pressure non-uniformity: The carrier head presses the wafer against the pad through a flexible membrane partitioned into 3–7 independent pressure zones. Zone-by-zone pressure differences create corresponding MRR differences across the wafer. Multi-zone head optimization — adjusting zone pressures to compensate for the pad’s natural radial uniformity profile — is the primary WIWNU correction tool.
Pad wear profile: Polishing pads wear non-uniformly, developing a radial wear profile that evolves over the pad lifetime. This shifts the MRR radial profile gradually and must be compensated through progressive adjustment of carrier head zone pressures across pad lifetime.
Slurry delivery asymmetry: Non-uniform slurry distribution across the pad surface — typically from a fixed-point delivery nozzle that creates a slurry wedge — contributes to azimuthal MRR asymmetry. Multiple delivery points or a slurry distribution bar reduces this effect.
Pattern density effects: As discussed in the ILD CMP article, circuit pattern density variation across the die drives systematic within-die thickness non-uniformity that WIWNU measurement at 49 sites may undercount. Full-wafer mapping at 200+ sites is required to characterize pattern-density-induced WIWNU at advanced nodes. For ILD-specific WIWNU management, see: ILD Oxide CMP Slurry: TEOS Planarization Process Guide.
Endpoint Detection Methods
In-Situ Optical Interferometry
Optical interferometry is the gold standard for endpoint detection in ILD oxide CMP. A broadband light source (typically 400–800 nm) is directed at the wafer surface through a transparent window in the polishing platen. The reflected signal contains interference fringes corresponding to the optical path length through the remaining oxide film. A spectrometer monitors the reflectance spectrum in real time; a thin-film model fits the spectrum to extract remaining oxide thickness continuously during polishing. When the fitted thickness reaches the target value (e.g., 200 nm remaining for a via-damascene step), the tool terminates polishing.
Modern optical endpoint systems achieve thickness measurement precision of ±3–5 Å (3σ) in steady-state polishing, enabling highly accurate termination of ILD CMP at precise remaining-thickness targets. The primary challenge is signal degradation when the polishing slurry film between the window and the wafer is opaque or uneven — this requires periodic slurry clearing protocols where DI water briefly replaces slurry at the measurement window.
Motor Current Sensing
Motor current (torque) sensing measures the friction force between the wafer and pad by monitoring the electrical current draw of the platen motor or carrier head motor. When polishing transitions from oxide (higher friction coefficient) to nitride or bare silicon (lower friction coefficient), the motor current decreases — providing an endpoint signal based on material transition rather than film thickness. This method is the primary endpoint approach for STI CMP (where polishing stops on arriving at the nitride surface) and is also used as a secondary confirmation signal in ILD CMP. Motor current endpoint is less precise than optical interferometry (typical endpoint detection accuracy ±10–30 Å) but is insensitive to slurry optical opacity, making it robust in high-abrasive-concentration slurry environments.
Combined Endpoint Strategies
Advanced-node oxide CMP tools combine both methods: optical interferometry tracks thickness continuously during bulk removal, while motor current sensing provides a secondary flag as the process approaches the final target. Some fabs add ex-situ metrology verification (measuring post-CMP thickness in the equipment set before the next deposition) as a lot-release gate, with automatic recipe adjustment for the next lot based on measured thickness deviation.
Slurry Delivery & Flow Control
Proper slurry delivery is necessary but frequently underappreciated as a process variable. The following parameters must be controlled for consistent oxide CMP performance:
- Flow rate: Set above the saturation point for the specific slurry/pad combination (typically 180–250 mL/min for 300 mm ILD CMP). Below saturation, MRR is flow-limited and sensitive to minor flow fluctuations. Above saturation, thermal benefits are the primary advantage of higher flow.
- Delivery nozzle position and angle: The slurry delivery nozzle should direct slurry toward the pad center at an angle that promotes uniform spreading outward under centrifugal force during platen rotation. Delivery that is too far from center concentrates slurry at mid-radius; too close to center results in slurry bypassing the wafer-pad interface.
- ポイント・オブ・ユース濾過： A 0.2–0.5 µm absolute filter at the final slurry delivery point removes particle aggregates formed in the supply line before they reach the polishing surface. Filter integrity should be verified at each maintenance interval — a clogged filter that bypasses can pass unfiltered slurry with large particle clusters, causing scratch excursions.
- pH at point of use: Slurry pH should be verified at the delivery point, not just at the bulk supply tank. CO2 absorption from ambient air, dilution from rinse water in supply lines, and temperature effects can all shift pH by 0.3–0.8 units between bulk supply and point of use.
パッドのコンディショニングと寿命管理
The polishing pad is the other critical consumable in oxide CMP — slurry chemistry and pad texture interact synergistically to deliver process performance. Pad conditioning is the process of mechanically refreshing the pad surface texture using a diamond-embedded conditioner disk to restore the micro-texture that enables slurry uptake and abrasive transport to the wafer-pad interface.
原位置コンディショニングと原位置コンディショニング
In-situ conditioning (conditioning disk rotates on the pad surface simultaneously with wafer polishing) maintains near-constant pad texture across the pad lifetime and reduces MRR decay between wafers. Ex-situ conditioning (conditioning between wafers, not during polishing) allows cleaner slurry on the pad during polishing but results in MRR decay within the polishing sequence that must be compensated by recipe time adjustments.
Pad Break-In
A freshly installed pad — even after the specified break-in conditioning sequence — typically delivers MRR 15–25% above the pad’s steady-state MRR for the first 15–25 wafers. Fabs running tight thickness control use a pad break-in lot (engineering wafers processed at the start of each new pad) to characterize the break-in MRR profile and establish the appropriate recipe time adjustment curve for production wafers.
Pad Lifetime
Polishing pads have finite lifetimes determined by mechanical wear (pad thickness reduction from conditioning) and chemical degradation. For standard polyurethane IC1000-type pads used in oxide CMP, lifetime is typically 300–700 wafer passes, depending on conditioning aggressiveness and the specific pad variant. Pads should be changed at a fixed wafer count or when MRR requires more than ±10% recipe time adjustment from the nominal recipe to maintain target thickness.
Run-to-Run Process Control
Run-to-run (R2R) feedback control automatically adjusts process recipe parameters (polish time, downforce, zone pressures) between lots based on post-CMP metrology results, maintaining target film thickness and uniformity despite pad aging, slurry lot variation, and ambient condition drift. R2R control is standard practice for ILD oxide CMP at advanced nodes and significantly reduces the process engineer time required for manual recipe maintenance.
A well-implemented R2R controller for ILD CMP maintains post-CMP oxide thickness within ±20 Å (3σ) of target across a full pad lifetime, versus ±50–100 Å for open-loop timed processes. The feed-forward component (predicting the next recipe based on incoming oxide thickness measurement from pre-CMP metrology) further reduces variation by pre-compensating for lot-to-lot incoming film thickness variation. For defect-related process control, see: Oxide CMP Slurry Defects: Root Causes, Detection Methods & Yield Impact.
Frequently Asked Questions: Oxide CMP Process Parameters
What is the Preston equation and when does it not apply?
Preston’s equation (MRR = Kp × P × V) models CMP removal rate as linear with pressure and velocity. It applies within a standard process window (0.5–4 psi, 60–120 RPM for oxide CMP) but breaks down at very low pressures (sub-linear regime where chemical softening dominates) and high pressures (super-linear regime with increased defect risk). Real oxide CMP also depends on slurry pH, temperature, abrasive concentration, and pad condition — factors not captured in the simple Preston model.
What WIWNU is achievable in production oxide CMP?
For ILD oxide CMP with multi-zone carrier heads and R2R control, WIWNU of 1.5–2.5% (1σ) is achievable in production at 300 mm. For STI CMP where nitride loss uniformity is tightly constrained, target WIWNU is 0.8–1.2% (1σ). These values require regular carrier head zone pressure re-optimization, conditioning disk maintenance, and run-to-run feedback control aligned to post-CMP metrology.
Which endpoint method is more accurate: optical or motor current?
In-situ optical interferometry is significantly more accurate than motor current sensing for ILD CMP, achieving ±3–5 Å (3σ) thickness control vs. ±10–30 Å for motor current. Optical is preferred wherever film thickness measurement is the endpoint criterion. Motor current is preferred where material transition detection (oxide-to-nitride) is the endpoint criterion, as it is insensitive to slurry optical opacity. Advanced-node tools typically combine both methods for maximum robustness.
How does slurry pH affect oxide CMP removal rate?
For colloidal silica ILD slurry, oxide MRR increases approximately 1.5–2× as slurry pH rises from 9 to 11. A pH drop of 0.5 units from the process optimum (e.g., from pH 10.5 to 10.0) typically reduces MRR by 10–15%. pH drift occurs on-tool from rinse water dilution, ambient CO2 absorption, and thermal effects. Point-of-use pH monitoring with alarm thresholds is best practice for advanced-node ILD CMP to catch drift before it affects film thickness control.
How often should the CMP polishing pad be changed?
Polishing pad lifetime for standard polyurethane IC1000-type pads in oxide CMP is typically 300–700 wafer passes, depending on conditioning aggressiveness and the specific application. The most reliable change criterion is MRR performance: when the recipe time required to hit target thickness exceeds ±10% of nominal, the pad should be changed. Fixed pad lifetime (wafer count) is also commonly used as a preventive maintenance trigger in high-volume production.
