Colloidal Silica Slurry for Oxide CMP: Optimizing STI and ILD Planarization

1. Oxide CMP in the Semiconductor Process Flow

Oxide CMP steps can be broadly divided into two categories based on their technical requirements:

• Stop-layer oxide CMP: A silicon nitride (Si₃N₄) or other material serves as a polish-stop layer. The SiO₂ must be removed to precisely this stop layer. High oxide/nitride selectivity is essential. STI CMP and some MOL dielectric steps fall into this category.
• Non-selective oxide CMP: There is no stop layer; the polish runs for a fixed time or to a target film thickness measured in-situ. Removal rate uniformity and planarization efficiency are the primary concerns. ILD, PMD, and most BEOL pre-metal dielectric steps fall here.

These two categories require fundamentally different slurry formulations. Stop-layer oxide CMP demands a high-selectivity formulation with carefully engineered polymer additive chemistry; non-selective oxide CMP uses simpler abrasive/carrier systems where the focus is on MRR uniformity and defect control.

2. STI CMP: Technical Requirements and Challenges

Shallow trench isolation (STI) is the standard CMOS device isolation technology from the 130 nm node onward. After the silicon trenches are etched and filled with CVD oxide (typically HDP-CVD SiO₂ or SACVD TEOS), the excess oxide overburden must be removed down to the Si₃N₄ hardmask layer. The key technical requirements for STI CMP are:

The central challenge in STI CMP is achieving high oxide removal rate simultaneously with high selectivity to the Si₃N₄ hardmask. Without adequate selectivity, the nitride acts as a poor polish-stop, resulting in excessive erosion of the STI oxide fill and—critically—polishing into the nitride-protected silicon active area, which alters device threshold voltages.

An additional challenge is the pattern-density loading effect (also called the “micro-loading” effect): die areas with a high density of STI trenches (dense patterns) experience higher local removal rates than die areas with sparse STI patterns (isolated features). This is because the slurry has more SiO₂ surface area to react with in dense regions, and the pad-wafer contact dynamics differ between dense and isolated feature zones. Managing this effect requires slurry formulation optimization alongside pad selection and conditioning recipe tuning.

3. ILD and PMD CMP: Requirements

Interlayer dielectric (ILD) CMP in BEOL processing and pre-metal dielectric (PMD) CMP in MOL share a simpler requirement profile than STI CMP—primarily because most ILD/PMD steps lack a nitride stop layer, eliminating the selectivity constraint. The primary ILD CMP objectives are:

• Global planarization: Reducing the step height from underlying metal patterning to the flatness required for next-level photolithographic overlay—typically WIWNU ≤3% and global step height ≤50 nm after CMP
• Low micro-scratch density: Scratches in the dielectric layer create void traps for subsequent metal fill, causing resistive interconnects or open circuits
• Dielectric integrity: For low-k SiCOH dielectrics (k < 2.5), the slurry pH must be controlled below 11 to prevent hydrolytic attack of the Si-C backbone, which increases the dielectric constant
• Post-CMP surface cleanliness: Slurry particle residues and organic additive residues at the dielectric surface must be removed completely by the post-CMP clean, as they interfere with ALD barrier-metal nucleation at vias

For ILD CMP with colloidal silica at 50–80 nm D50, typical removal rates are 500–1,500 Å/min at standard conditions—adequate for typical ILD film thicknesses of 2,000–6,000 Å with acceptable cycle time.

4. Slurry Formulation for High Selectivity in STI CMP

Achieving STI oxide/nitride selectivity above 20:1 with colloidal silica requires the addition of selectivity-enhancing polymer additives that preferentially suppress Si₃N₄ removal while leaving SiO₂ removal rate largely unaffected. The primary commercially used selectivity additive classes are:

4.1 Polymeric Nitride Inhibitors

Polyacrylic acid (PAA) and its copolymers are the most widely studied colloidal silica selectivity additives. PAA chains adsorb preferentially on Si₃N₄ surfaces (due to specific electrostatic and hydrogen-bonding interactions between the carboxylate groups and the amine-like surface species on Si₃N₄) while showing minimal adsorption on SiO₂. This preferential adsorption creates a protective polymer layer on Si₃N₄ that resists mechanical abrasion and chemical attack by the slurry, reducing Si₃N₄ MRR by 80–95% while SiO₂ MRR is reduced by only 10–20%—yielding selectivities of 20:1 to 100:1 depending on PAA molecular weight and concentration.

Other nitride inhibitor polymers include poly-4-styrenesulfonic acid (PSS), polyvinyl alcohol (PVA), and various commercial proprietary copolymers. The choice of polymer affects not just selectivity but also defectivity (some polymers act as mild surfactants, reducing scratch rates) and post-CMP cleanability.

4.2 pH as a Selectivity Lever

In addition to polymeric inhibitors, pH adjustment is a secondary selectivity lever. The Si₃N₄ etch rate by aqueous hydroxide increases with pH above 11, while the SiO₂ etch rate plateaus at lower pH values. Operating at pH 10.0–10.5 (rather than 11.0–11.5) provides a mild secondary selectivity benefit by keeping Si₃N₄ chemical dissolution lower, and is the standard pH range for high-selectivity STI slurries.

4.3 Additive Concentration Optimization

The optimal inhibitor concentration is not simply “as high as possible.” At excess inhibitor concentration, the polymer begins to adsorb on SiO₂ surfaces as well (reducing oxide MRR below the process target) and can cause excessive particle bridging that increases LPC. The inhibitor loading must be co-optimized with the colloidal silica solids content and pH to achieve the target selectivity at the target oxide MRR.

5. Process Parameter Optimization

Colloidal silica oxide CMP performance is sensitive to all standard Preston equation variables (pressure, velocity) plus slurry-specific variables. Key optimization considerations:

• Down-force (pressure): Higher pressure increases MRR proportionally (per Preston’s equation) but also increases defectivity risk and can cause pad glazing at sustained high pressures. Optimal range for colloidal silica oxide CMP is typically 2–4 psi. Lower pressure (<1.5 psi) may produce non-linear MRR behavior as not all particles are in active contact.
• Table and head rotation speed: MRR increases with linear velocity (pad surface speed × wafer carrier speed). Typical CMP tool settings of 60–120 rpm table / 60–120 rpm head provide linear velocities of 0.5–1.5 m/s at wafer edge. Higher speeds improve planarization efficiency but can cause slurry hydroplaning on the pad at very high speeds.
• Slurry flow rate: Typically 100–300 mL/min for 300 mm CMP tools. Higher flow rates improve temperature control at the polishing interface but reduce slurry residence time on the pad. The minimum flow rate that maintains a continuous slurry film between pad and wafer is the practical lower limit.
• Conditionnement des tampons: In-situ conditioning rate (conditioner down-force, sweep speed, and duty cycle) determines the steady-state pad asperity population. More aggressive conditioning maintains higher asperity density, increasing effective MRR but also increasing defectivity from conditioning debris. Optimize conditioner sweep parameters for the target MRR/defectivity balance.
• Slurry temperature: Most oxide CMP slurries are delivered at 20–25°C. Higher slurry temperature increases the chemical removal rate component, raising MRR; fab-to-fab slurry delivery temperature differences are a common source of MRR variation that is often underappreciated.

6. Endpoint Detection in STI CMP

Accurate endpoint detection is critical in STI CMP, where overpolishing (removing too much Si₃N₄) directly degrades device performance and underpolishing (leaving oxide overburden) causes step height issues for downstream lithography. Two primary endpoint methods are used with colloidal silica STI slurries:

6.1 Motor Current (Friction) Endpoint

The current drawn by the CMP table motor (or the wafer carrier motor) changes as the polishing surface transitions from oxide to nitride, because the friction coefficient between the pad and Si₃N₄ surface differs from pad-and-SiO₂. The oxide-to-nitride transition produces a characteristic change in the motor current signal—typically a slight reduction in current as the smoother, harder nitride surface replaces the softer oxide. This signal change is detected algorithmically and used to trigger endpoint or to initiate a timed overpolish sequence.

6.2 In-Situ Optical (Interferometric) Endpoint

Optical in-situ endpoint systems use a laser beam projected through a transparent window in the polishing pad to measure the reflectance or film thickness of the polished surface in real time. As the oxide film thins, the interferometric signal oscillates and then transitions to the characteristic signal of the underlying nitride or silicon layer. Optical endpoint is more precise than motor current for tight active-area-height control but requires a compatible pad (with optical window) and appropriate slurry transparency at the measurement wavelength—colloidal silica slurries are generally compatible with visible and near-IR optical endpoint systems.

7. Defect Mitigation in Oxide CMP

The most common oxide CMP-specific defects and their mitigation strategies with colloidal silica slurry:

• Micro-scratches: Caused by large particles (LPC >0.5 µm) in the slurry or pad debris. Mitigate with: low-LPC slurry specification (<2,000 counts/mL at >0.5 µm for STI; <800 for ILD); point-of-use filtration (500 nm filter); conditioning recipe optimization to minimize conditioner debris generation.
• SiO₂ residues (slurry residue defects): Slurry particle residues not fully removed by post-CMP clean. Mitigate with: optimize post-CMP brush scrub chemistry (dilute SC1, citric acid); verify slurry particle pH is compatible with post-CMP clean surfactants; evaluate rinse sequence hold time.
• Dishing of STI oxide: Excessive removal of STI oxide in wide trench areas (dishing) relative to narrow trench areas. Caused by pattern-density loading combined with overpolish time. Mitigate with: optimize inhibitor concentration to increase STI oxide resistance to removal after nitride is reached; minimize overpolish time; use nitride-endpoint triggered time-controlled overpolish.
• Erosion of nitride (active area height loss): Removal of Si₃N₄ in dense STI pattern areas during the main oxide removal step. Mitigate with: increase inhibitor concentration to raise oxide/nitride selectivity; verify selectivity lot-to-lot consistency.

8. JEEZ Product Recommendations for Oxide CMP

For STI oxide CMP applications where high MRR and selectivity are required, the JEEZ CS-100 Series (D50 = 90–110 nm, pH 10.0–11.5) is the appropriate platform. JEEZ offers this series with optional selectivity additive packages pre-formulated to achieve SiO₂:Si₃N₄ selectivities of 20:1 to 100:1 depending on the target process window. For ILD oxide CMP where the primary requirement is low defectivity at moderate MRR, the JEEZ CS-60 Series (D50 = 55–70 nm, pH 10.0–11.0) provides an optimal balance.

For the full colloidal silica application landscape and complete technical reference, see: Colloidal Silica Slurry: The Complete Guide to CMP Applications, Properties, and Selection.

9. Frequently Asked Questions