Chemical mechanical planarization (CMP) is an essential step to remove overburden oxide surface and form STI (Shallow Trench Isolation) structures during the semiconductor manufacturing. This is usually carried out with ceria slurry due to high selectivity between oxide and nitride surfaces. However, strong ceria adhesion to oxide surface is a major concern, as it cannot be removed easily during the post-CMP cleaning step. Generally, it is performed by a combination of physical and chemical cleaning. Conventional cleaning solutions such as DHF (Dilute Hydrogen Fluoride), SC1(NH₄OH: H₂O_2: H2O mixture), SPM (H₂SO₄:H₂O₂ mixture) solutions are used with optimized concentrations and sequences with the support of physical forces such as PVA brush scrubbing or megasonic. However, the adhesion and removal mechanism of ceria particles from the oxide surface is not clear yet. The ceria particles would attach to the oxide to the surface may be due to electrostatic interaction or valence (Ce³⁺/Ce⁴⁺) effect or chemical bonding linkage of Ce-O-Si. These interactions are strongly affected and changed based on the pH of ceria used for polishing. In this work, it was emphasized to identify the specific individual mechanisms of physical cleaning and chemical cleaning in removing ceria particles from the oxide surface. Two types of physical cleanings, one is the non-contact mode (megasonic) cleaning and another one is contact mode (PVA brush) cleaning, were performed. The effect of different chemicals on removing the ceria particles was also performed. The key mechanism for the ceria particle cleaning was hypothesized.

TEOS coupons (4´4 cm) were polished with ceria slurry (100nm) of pH 4 and 8, containing no additives. The optimization of physical cleaning experiments, performed with only DIW, was carried out with a single wafer megasonic cleaning (Akrion, 0.83MHz) and brush cleaning tool (lab designed pin type brush). Chemical cleaning was performed with various chemistries (SC1, DHF, SPM) to compare and understand the ceria removal performance from the oxide surface. The particles before and after cleaning were observed by FESEM and optical microscopy. To understand chemical interactions, the polished samples were analyzed for XPS, ATR FTIR, and Raman analysis before and after cleaning. The type of chemical interactions before cleaning and their modifications with the type of chemistry used were discussed.

The comparative study of physical cleaning (megasonic vs. brush) were performed for the removal of ceria particles attached to the surface after polishing with slurries of pH 4 and 8. It was observed that the polished oxide surface was more contaminated with ceria in the pH 8 than in pH 4. Megasonic cleaning was able to remove partially, whereas brush scrubbing removed completely for pH 4 samples. However, the ceria particles from pH 8 samples were strongly attached not removed by megasonic and partially removed by brush cleanings. It could be expected that particles would attach electrostatically at pH 4, whereas chemically through Ce-O-Si bond at pH 8. It may be necessary to use chemical forces to break the chemical bonds between ceria-silica. It was suspected that the change of valence of ceria before and after cleaning with cleaning chemistries also would play a key role. Hence different chemistries were used to check the removability of ceria. The samples before and after cleaning were analyzed with different techniques to understand the chemical forces required to break the ceria-silica bonding. Detailed hypothesis for the ceria removal was explained by their chemical interactions with different chemistries.