Dishing, Erosion, and Defects in Tungsten CMP: Root Causes and Slurry-Level Solutions

1. Dishing: Definition, Measurement, and Root Causes

Dishing is the concave depression that forms in the center of a tungsten feature — a contact plug, via, or gate fill structure — after chemical mechanical planarization. It occurs because the polishing pad is not rigid: under the normal contact pressures of CMP (1–5 psi), the pad deflects slightly into the recess of a wide or isolated tungsten feature, allowing abrasive particles to continue removing tungsten from the center of the feature even after the surrounding oxide has been reached. The result is a metal surface that sits below the level of the surrounding dielectric when measured in cross-section — a concavity that is most severe in isolated large-diameter features and least severe in dense arrays of small features.

Measurement

Dishing is measured in angstroms (Å) or nanometers using atomic force microscopy (AFM) or a stylus-based step height profiler on patterned test wafers. Standard metrology uses test vehicles containing isolated tungsten features of multiple widths — typically 1, 2, 4, 10, and 100 μm — to capture the dependence of dishing on feature size. The measurement is the depth of the concavity at the feature center relative to the surrounding ILD surface. In-line metrology on product wafers uses eddy current sensing for approximate tungsten recess measurement without patterned test structures.

Root Causes at the Slurry Level

High W:Oxide selectivity in the clearing step: A highly selective slurry (W:Oxide ratio >25:1) continues removing tungsten efficiently even after the surrounding oxide is exposed, because the oxide MRR is extremely low. The pad continues to contact the tungsten center of wide features at a rate not matched by oxide removal at the feature perimeter, driving the feature center lower than the surrounding ILD. This is the dominant slurry-level cause of dishing in isolated or wide features (>2 μm).

Extended Step 1 overpolish: Any overpolish time in Step 1 — time spent polishing after the nominal tungsten clearance endpoint — contributes to dishing because the process is now removing tungsten from within the features rather than from the overburden above them. The instantaneous dishing rate during overpolish approximates the tungsten MRR of the slurry divided by the area loading factor of tungsten on the wafer.

Pad deflection geometry: Dishing magnitude scales with feature width (not diameter for contact arrays) because wider features allow more pad deflection depth at the center. Isolated features dish more than equal-width features in a dense array, where neighboring features support the pad at a higher effective average height.

Slurry-Level Solutions

• Transition from Step 1 to Step 2 (lower selectivity buff) as early as endpoint allows — do not extend Step 1 overpolish unnecessarily
• Use a moderate-selectivity clearing slurry (10:1–25:1) for the endpoint phase rather than a hyper-selective formulation (>100:1)
• Reduce downforce during the clearance phase to decrease pad deflection depth into feature recesses
• Tighten endpoint detection window to minimize the overpolish duration that accumulates dishing

2. Erosion: Definition, Measurement, and Root Causes

Erosion is the loss of inter-layer dielectric (ILD) material in high-density tungsten array regions relative to low-density or isolated oxide regions on the same wafer. It occurs because the collective mechanical loading of many densely packed tungsten plugs causes the polishing pad to bear against the array surface at a lower height, with the result that polishing removes more oxide within the high-density array than outside it. Erosion reduces the effective dielectric thickness between interconnect levels and can cause electrical reliability failures if it reaches a magnitude that compromises the isolation between adjacent tungsten contacts in a dense array.

Measurement

Erosion is measured as the difference in oxide height between a high-density tungsten array region (typically 50–70% tungsten fill area) and an adjacent unpatterned oxide region on the same test wafer, expressed in angstroms. Standard measurement uses AFM or stylus profilometry across a step structure at the array boundary, combined with patterned metrology wafers that include a range of density test structures from isolated to high-density tungsten fill.

Root Causes at the Slurry Level

High oxide MRR in the Step 2 buff slurry: Non-selective buff slurries with high oxide MRR cause rapid dielectric removal in high-density array regions during overpolish. Total erosion scales directly with the product of oxide MRR and Step 2 polish time: erosion (Å) ≈ oxide MRR (Å/min) × overpolish time (min).

Extended Step 2 overpolish required by dishing from Step 1: If Step 1 generates significant dishing in isolated features, Step 2 must run longer to correct it, which drives erosion in the array regions. This coupling between dishing and erosion is the fundamental reason why both must be optimized together.

Poor planarization efficiency (PE): Planarization efficiency is the ratio of material removed from high topographic features to material removed from recessed features under the same polishing conditions. A slurry with low PE removes material from high and low spots at nearly equal rates, requiring more total polishing action to achieve clearance and driving proportionally more erosion. High-PE formulations preferentially remove high spots, reducing total time to clearance and minimizing array erosion.

3. The Dishing–Erosion Trade-off: Why You Cannot Solve Both Simultaneously with One Slurry

The relationship between dishing and erosion creates an unavoidable engineering trade-off that is intrinsic to the geometry of patterned surfaces under CMP — it cannot be eliminated by formulation alone.

The two-step process architecture manages this trade-off by assigning each selectivity class to the phase of the process where its benefits dominate: high selectivity for efficient bulk removal with minimal dielectric loss (Step 1), low selectivity for planarization correction and surface quality improvement (Step 2). The process engineer’s job is to allocate time between these steps such that dishing from Step 1 and erosion from Step 2 are both within specification simultaneously.

4. Microscratch Defects: Causes and Slurry-Level Controls

Microscratches are linear surface defects produced when a particle harder than the wafer surface is dragged across it under high contact stress during polishing. In tungsten CMP, they are detected by post-CMP optical brightfield and darkfield scanning inspection (Surfscan or equivalent) and appear as bright linear features in darkfield mode. Microscratch counts are typically specified in defects per cm² at a given size sensitivity threshold (commonly 0.09 or 0.16 μm equivalent sphere diameter).

Root Causes in Tungsten CMP Slurries

Large-particle tail in abrasive PSD: The primary source of high-severity microscratches in most CMP processes. Particles or agglomerates larger than 1 μm in the abrasive PSD that escape point-of-use filtration can generate scratches spanning millimeters on the wafer surface when trapped between pad and wafer under full polishing pressure. Fumed silica slurries are more susceptible than colloidal silica due to their intrinsically broader PSD; both require effective large-particle characterization.

Iron hydroxide precipitate particles: Fe(OH)₃ generated by pH excursions above ~3.5 in the slurry delivery system deposits rust-colored, hard, irregularly shaped particles that produce scratches morphologically similar to abrasive-originated defects. This source is often confused with abrasive quality issues until XRF or EDS analysis of the scratch deposits identifies iron as the primary element.

Pad conditioning debris: Diamond particles from the CMP pad conditioner, pad fibers, or hardened polishing debris (agglomerated pad material and slurry residue) can trap between pad and wafer and generate scratch events that correlate in time with conditioning cycles rather than with slurry lots. Distinguished from slurry-origin scratches by correlation with tool maintenance records and the absence of the scratch defect in lots processed with a newly dressed pad.

Slurry agglomeration from instability: Colloidal or fumed silica formulations near or beyond their shelf life, or subjected to freeze-thaw cycles during transport, may develop particle agglomerates that increase the large-particle tail beyond the original specification. Incoming PSD verification and shelf-life enforcement prevent this source.

Prevention Strategies

• Specify incoming slurry acceptance criteria for particles >1 μm (by AccuSizer single-particle optical sensing): e.g., <100 particles/mL at >1 μm threshold
• Install and maintain point-of-use filtration at 0.2–0.5 μm nominal rating in the slurry delivery system
• Monitor and control slurry pH throughout the delivery loop to prevent Fe(OH)₃ precipitation
• Enforce shelf-life limits on both components and verify by MRR control wafer if any component has exceeded recommended storage time
• Track scratch defect counts vs. slurry lot number to separate material-related from tool-related excursions

5. Corrosion Pitting: Over-Oxidation at the Tungsten Surface

Corrosion pitting appears as submicron-diameter hemispherical depressions in the tungsten surface, distributed preferentially at grain boundaries, crystallographic defect sites, and high surface energy locations where oxidation kinetics are locally faster. Unlike dishing (a large-scale topographic effect) or scratches (linear defects), pits are small (0.05–0.5 μm diameter), numerous, and distributed across the polished surface. In contact features, pits increase electrical contact resistance when they intersect the active contact area.

Slurry Chemistry Causes

• Excess H₂O₂ concentration: Over-oxidation at locally high oxidizer concentrations creates WO₃ faster than the mechanical removal step can clear it. The accumulated WO₃ eventually breaks through the passivation layer at preferential nucleation sites, exposing tungsten to continued aggressive oxidation.
• Excess Fe³⁺ loading: High hydroxyl radical flux from an over-catalyzed Fenton cycle produces locally aggressive oxidation, particularly at grain boundaries where the tungsten crystal structure is disrupted and more reactive.
• Elevated process temperature: H₂O₂ decomposition kinetics approximately double every 10°C. Uncontrolled slurry temperature at the point of dispensing (e.g., in summer conditions or when slurry recirculation loop temperature control fails) can push the chemistry into a corrosion-pitting regime even at nominal oxidizer concentrations.
• pH below 2: Extremely acidic conditions enhance tungsten dissolution through additional non-passivating pathways independent of WO₃ formation.

Prevention requires operating both oxidizer and catalyst concentrations at the lower end of their respective MRR-plateau windows, and monitoring slurry dispense temperature at the tool — particularly in facilities where polishing tool chiller failures are common during summer months.

6. WO₃ Residue and Ionic Surface Contamination

WO₃ Residue

WO₃ micro-residue haze appears on post-CMP brightfield inspection as a diffuse, slightly colored surface film rather than discrete particles. It results from incomplete dissolution of the WO₃ reaction product, either because the complexant (organic acid) concentration in the slurry is insufficient to maintain dissolved tungsten species in solution, because the pH is higher than optimal (reducing WO₃ solubility), or because a very high MRR generates WO₃ faster than dissolution kinetics can process it. Adjusting complexant loading upward or slightly reducing pH within the accepted window typically resolves this without significantly impacting MRR.

Metallic Ionic Contamination

Iron (Fe) contamination at the wafer surface is the most prevalent metallic contaminant introduced by tungsten CMP slurry chemistry. Even at concentrations above 10¹¹ atoms/cm², Fe at the silicon or gate oxide surface creates deep-level traps that increase junction leakage current and reduce minority carrier lifetime — electrical effects that can cause subtle yield loss in logic devices and significantly degrade data retention in DRAM.

In addition to Fe from the catalyst, sodium (Na) and potassium (K) from abrasive manufacturing, and calcium (Ca) from process water used in slurry manufacturing, are secondary metallic contamination sources that should be specified at low-ppb levels in the slurry and verified by ICP-MS on incoming lots. TXRF (Total Reflection X-ray Fluorescence) provides wafer-mapping metallic contamination data; VPD-ICPMS (Vapor Phase Decomposition combined with ICP-MS) provides the highest sensitivity for trace-level metallic surface contamination quantification.

7. The Role of Slurry Selectivity in Dishing and Erosion Outcome

Selectivity — the ratio of tungsten MRR to oxide MRR — is the single slurry formulation parameter with the largest influence on both dishing and erosion simultaneously. The mechanism by which selectivity drives each defect type is distinct and must be understood separately before the two-step process strategy can be designed effectively.

8. Two-Step Process Strategy for Defect Minimization

The practical engineering response to the inherent dishing–erosion trade-off is a two-platen, two-slurry process that separates the bulk removal function from the surface quality and planarization function. Designing the two-step architecture for defect optimization involves setting three key process parameters:

Step 1 overpolish time allocation: The time spent in Step 1 after nominal endpoint detection determines how much dishing accumulates in isolated and wide features before Step 2 begins. The target is the minimum overpolish that achieves statistically reliable tungsten clearance across the entire 300 mm wafer — accounting for the within-wafer non-uniformity (WIWNU) of Step 1 and the endpoint detection tolerance. Tighter Step 1 WIWNU directly reduces the minimum overpolish required, lowering dishing as a secondary benefit.

Step 2 total polish time: Step 2 must run long enough to clear the Ti/TiN liner and reduce surface defects, but not so long that erosion in dense arrays accumulates to specification-violating levels. The erosion budget per unit of Step 2 time is set by the oxide MRR of the selected buff slurry; lower oxide MRR gives more temporal margin for Step 2 without exceeding erosion limits.

Slurry transition management: In some tool configurations, the Step 1 slurry is still present on Platen 1 during the early moments of Step 2 polishing if the pad is not adequately purged between steps. The dishing rate during this transition period differs from both Step 1 and Step 2 steady-state rates. This transient behavior should be characterized during process qualification using time-resolved measurement schemes.

Response surface methodology (RSM) designs — using Step 1 overpolish time and Step 2 total time as the primary factors — with dishing, erosion, WIWNU, and post-clean scratch count as simultaneous response variables — provide the most efficient path to finding the process window that satisfies all defect specifications simultaneously.

9. Abrasive PSD and Its Impact on Defect Performance

The particle size distribution of the abrasive is the slurry property with the strongest direct link to microscratch defect performance. Three PSD metrics are relevant for defect engineering:

Mean particle size (D50): Larger mean particle size generally increases MRR (more contact area per particle at equivalent mass loading) but increases surface roughness and scratch frequency. Most production tungsten CMP slurries use mean particle sizes in the 80–200 nm range for bulk removal and 40–100 nm for buff applications.

PSD width (D90/D10 ratio or equivalent): A narrow PSD (characteristic of colloidal silica) concentrates all particles near the mean size, minimizing the statistical probability of a high-severity scratch event from an outlier particle. A broad PSD (characteristic of fumed silica aggregates) provides higher average MRR but more frequent high-severity scratches from particles at the large end of the distribution.

Large-particle tail (particles >1 μm per mL): This is the single most predictive defect metric in tungsten CMP slurry characterization. Controlled experiments consistently show that slurry lots with elevated counts of particles >1 μm (measured by single-particle optical sensing such as AccuSizer or Brightwell systems) produce significantly higher post-CMP scratch counts than matched lots with low large-particle counts, regardless of mean particle size. Setting an incoming acceptance limit on this metric — typically <100 particles/mL at the >1 μm threshold — is one of the highest-leverage quality control actions a fab can take for W CMP defect management.

10. Post-CMP Inspection Methods

Effective defect control requires matching the right inspection technique to the defect type being characterized. Using only one inspection method provides an incomplete picture of the defect landscape after tungsten CMP.

11. Defect Reduction Checklist: Process and Slurry Controls

The following checklist covers the most impactful defect control actions for tungsten CMP, organized from incoming material verification through in-process monitoring to process design. Use this as a qualification readiness audit before releasing a new slurry lot or a new slurry formulation to production.

• Verify H₂O₂ concentration on every incoming Component B lot by iodometric titration or spectrophotometry — compare to CoA specification (±5% of nominal)
• Verify slurry pH of mixed formulation at 23°C on every incoming lot — confirm within ±0.2 pH units of CoA value
• Measure abrasive large-particle count (>1 μm by AccuSizer or equivalent) on incoming Component A lots — set acceptance limit (e.g., <100 particles/mL)
• Confirm point-of-use filtration elements are within service life (log filter change dates and pressure differential across filter)
• Verify Fe³⁺ catalyst concentration against CoA by ICP-OES or colorimetric method on quarterly basis or upon suspected pitting excursion
• Confirm Component B manufacturing date; enforce use-by limit (typically 3–6 months from manufacture for H₂O₂-based component)
• Run MRR control wafer at start of each run and compare result to the established baseline ±5%
• Confirm Step 1 endpoint detection is active and is catching endpoint within the qualified detection window — verify eddy current baseline and threshold settings
• Confirm Step 1 overpolish time is set to the value established during process qualification — do not extend beyond qualified window without a change control review
• Confirm Step 2 total polish time is within the erosion budget for the target application — calculate erosion estimate from Step 2 oxide MRR × Step 2 time and confirm it is below specification
• After any slurry lot change, run a dishing/erosion metrology lot on the patterned test vehicle before releasing the lot for product wafer production