Is There a Definitive Corrosion Control Window for Low-k/Cu Post-CMP Cleaning?

Aqueous alkaline formulations containing alkanolamines and chelating agents are applied after barrier CMP on 300 mm copper dual damascene wafers. The remover is diluted with ultrapure water meeting ASTM D5127-13 Type E-1.2 at 18.2 MΩ·cm resistivity. Point-of-use dilution ranges from 1:10 to 1:20. Process temperature is maintained between 25 °C and 45 °C. Contact time on single-wafer spin tools is 30 s to 90 s. Megasonic energy at 1.7 MHz is applied through a rod-type transducer. Chemical flow rate is 0.8 L/min to 1.2 L/min. The tool chamber is operated under ISO 14644-1:2015 Class 3. Chemical distribution components are qualified per SEMI F57-0214.

The EL-grade specification limits metal contaminants to below 1 ppb for sodium, potassium, iron, copper, zinc, calcium, magnesium, and aluminum by inductively coupled plasma mass spectrometry. Particle content in the undiluted product is controlled below 10 particles/mL at 0.1 μm or larger. This is consistent with front-end semiconductor cleaning chemical requirements.

The primary process conflict is simultaneous benzotriazole film removal and copper dissolution control. Benzotriazole films protect copper lines during CMP but must be removed to avoid contact resistance degradation. Under moderately alkaline conditions, copper oxide dissolves slowly. At pH above 11, organosilicate low-k films exhibit hydroxyl uptake. At pH below 7, copper etch rate rises strongly. The production window is therefore narrow. In-line acceptance on logic lines typically requires copper etch below 1.0 nm/min. k-value shift is held below 0.2 after a 24 h queue time. These values are verified by mercury-probe capacitance-voltage and spectroscopic ellipsometry at 633 nm. Post-clean particle counts are monitored by laser scattering. The larger than 0.12 μm LPD threshold is typically below 100 per wafer. End products include FinFET logic devices at 7 nm and 5 nm design rules for mobile processors and high-performance computing devices.

Cleaning efficiency is governed by electrokinetic particle removal. In 1:10 dilution the working pH settles between 9.2 and 9.8. Silica abrasive surfaces acquire a strongly negative zeta potential. Oxide surfaces acquire a less negative zeta potential. The resulting electrostatic repulsion assists brush and megasonic removal. If the bath pH drifts below 9.0, the zeta potential difference narrows. Particle redeposition at wafer edge regions becomes visible on scan maps. Bath pH is therefore controlled to ±0.2 units. Recirculating chemical baths are monitored by conductivity and titratable alkalinity.

Tungsten plug loss following post-CMP clean is monitored on 300 mm memory lines using contact test structures. The remover encounters a different residue matrix: tungsten oxide, alumina abrasive, ferric nitrate oxidant by-products, and organic surfactant films remaining from W slurry. Brush scrubbing tools with polyvinyl alcohol roller brushes deliver chemistry at 1:15 dilution. Platen pressure is 0.3 psi to 0.8 psi. Spin speed is 200 rpm to 500 rpm. The cleaning sequence runs 45 s to 120 s per wafer. Process temperature is 30 °C to 45 °C.

The constraint is tungsten etch acceleration above pH 10. Alkaline chemistries of this class can produce tungsten oxide dissolution if residence time is not controlled. Ferricyanide residues from tungsten slurry are reduced by organic remover components. Surface metal contamination is measured by vapor phase decomposition inductively coupled plasma mass spectrometry. Typical acceptance for tungsten loss is 0.5 nm per clean. Post-brush rinse is performed with ultrapure water meeting ASTM D5127-13 Type E-1.2. End products include DRAM contact plugs and 3D NAND wordline structures.

On brush scrubber platforms, the cleaning chemistry is dispensed during low-speed scrub and high-speed spin-off. The organic residue remover also suppresses redeposition of alumina particles through surface charge modification. Defect inspection after tungsten CMP cleaning commonly classifies residues as organic smear, alumina particles, and tungsten oxide flakes. The remover targets organic smear and alumina adhesion. Process engineers adjust brush downforce and chemistry flow to manage tungsten oxide flake carryover. End point inspection uses electron beam review on contact chains.

Shallow Trench Isolation CMP Leaves Ceria Particles That Alkaline Chemistry Must Lift Without Oxide Roughening

Oxide CMP on shallow trench isolation structures uses ceria-based slurry. Organic additives include polyacrylic acid and polyvinyl pyrrolidone. After post-CMP drying, these organics form a thin binder that cements ceria particles to the oxide surface. The remover lifts this binder through alkaline hydrolysis and chelation. Typical dilution is 1:10 to 1:30. Process temperature is 20 °C to 45 °C. Soft polyvinyl alcohol brush scrubbing is used on single-wafer cleaners. Brush pressure is maintained at 0.3 psi to 0.5 psi to avoid microscratching.

The critical endpoint is oxide roughening and ceria redeposition. pH above 11 increases oxide etch and roughens the active area. Insufficient chelator concentration leads to ceria particle reattachment on silicon dioxide and silicon nitride. Particle removal is verified by laser scattering. The acceptance level is typically IEST-STD-CC1246E level 25 or lower. End products include CMOS logic isolation structures and memory periphery.

Ceria particle adhesion on silicon dioxide occurs through hydroxyl bridging and organic binder films. Polyacrylic acid binds to ceria via carboxylate groups. Alkaline hydrolysis cleaves these linkages. The chelator component then complexes cerium(III) and cerium(IV) ions. If the chelator concentration is too low, cerium redeposition occurs at wafer edges. This is observed as crescent-shaped defects on laser scattering maps. Defect classification is confirmed by scanning electron microscopy with energy-dispersive X-ray spectroscopy.

When Cobalt and Titanium Interconnects Require Organic Residue Removal Without Galvanic Attack

Cobalt-selective slurries generate organic residues that include benzotriazole derivatives, polyacrylamide dispersants, and silica abrasive. The post-CMP cleaner must remove this matrix from cobalt/titanium nitride interfaces in middle-of-line contacts. Galvanic coupling between cobalt and titanium nitride is the primary failure mode. The remover is buffered at pH 7.5 to 9.0. Point-of-use dilution is 1:10 to 1:15. Temperature is held at 25 °C to 35 °C. Single-wafer spin cleaning uses 1.7 MHz megasonic agitation at 0.9 L/min chemical flow. Contact time is 60 s to 120 s.

Corrosion inhibition is measured by open-circuit potential and electrochemical impedance spectroscopy on wafer coupons. Production fabs require cobalt loss below 0.3 nm per clean and titanium nitride loss below 0.2 nm. These values are confirmed by X-ray fluorescence. Bath life is controlled by pH drift of 0.2 units and total organic carbon loading below 500 mg/L. Exceeding these limits shifts cobalt etch rate nonlinearly. Process fluid components are qualified per SEMI F57-0214. The cleaning chamber operates under ISO 14644-1:2015 Class 2. End products include FinFET local interconnect modules at 7 nm node and below.

The organic residue remover also prevents post-clean oxidation on cobalt surfaces. Cobalt exposed to air after cleaning can form cobalt oxide at ambient humidity above 60% RH. A short nitrogen-purged queue is used to limit oxidation. Process fabs time the post-clean wait to less than 4 h before the next dielectric deposition. Corrosion inhibitors in the formulation provide temporary passivation. This temporary layer is removed by subsequent deposition pretreatments.

In copper redistribution layer CMP on fan-out wafer-level packaging lines, organic residue can be trapped under polyimide edges and along the copper seed layer. The cleaning sequence is shorter than front-end applications. Typical dilution is 1:20. Temperature is 25 °C. Single-wafer spray tools deliver 30 s to 60 s chemical dwell. The remover must not swell polyimide dielectrics. Swelling occurs above pH 12. Copper roughness after clean is kept below 1.0 nm Rq. Polyimide adhesion is tested by cross-cut tape per ASTM D3359. End products include 2.5D interposers and fan-out wafer-level packages. Published data for this specific configuration is limited; fabs rely on in-line defect inspection and adhesion pull tests.

Copper pillar structures on the same wafer are sensitive to sidewall etching when the cleaner contains aggressive alkaline agents. The process engineer adjusts dilution and dwell to keep sidewall etch below 50 nm per pass. This limit is measured by cross-section scanning electron microscopy on test coupons. The post-CMP clean must also remove residual organic additives from copper electroplating baths. These additives include accelerators, suppressors, and levelers. Accelerator compounds such as bis(3-sulfopropyl) disulfide adsorb strongly on copper and can cause localized corrosion if not removed. The remover competes with these adsorbates. X-ray photoelectron spectroscopy can verify carbon and sulfur reduction. In-line fabs typically use contact angle measurements as a rapid proxy for organic residue removal. The target post-clean contact angle is below 10°.

Waveguide Critical Dimension Loss as a Function of Cleaning Chemistry Composition

Waveguide post-CMP residue consists of silica slurry particles, ceria or silica abrasive, and organic surfactant films from oxide and silicon nitride CMP steps. The remover is used at 1:30 dilution at 22 °C to 30 °C. Contact time is 30 s on single-wafer scrubbers. PVA brush pressure is 0.5 psi. RMS roughness is measured by atomic force microscopy. The target is below 0.2 nm on silicon nitride waveguide sidewalls. Critical dimension loss is held below 0.5 nm per clean. If the pH rises above 10, silicon nitride etch increases. If organic residue remains, optical loss at 1550 nm increases. End products include photonic integrated circuits for datacom transceivers.

Residual organic films from CMP act as scattering centers and can increase propagation loss. Loss measurements by optical backscatter reflectometry are used to screen process splits. A shift of 0.1 dB/cm at 1550 nm is considered significant for single-mode waveguides. The cleaning chemistry is therefore validated with patterned photonic test structures. This is more representative than blanket film roughness. The cleaning chamber is operated under ISO 14644-1:2015 Class 3. End products include silicon photonic transceivers for data centers.