Eliminating carbon and watermarks during post-CMP cleaning
11/01/2004
The use of organosilicate glass (OSG) as a low-k dielectric in copper interconnects creates challenges in cleaning wafers after they have been through chemical mechanical planarization (CMP). The high methyl content of OSG yields a hydrophobic surface, making it difficult to wet surfaces with aqueous solutions. The use of surfactants to enhance surface wetting raises the risk of organic contaminates. Watermarks can hide killer defects from inspection tools. This article assesses several post-CMP cleaners in addressing these issues.
Carbon residues on dielectric layers can result in poor adhesion of subsequent process films, resulting in pull-outs during wafer polishing, voids during plating, and delamination of buried films after elevated thermal cycles, due to decomposition of the organic contaminate. Quench or staging solutions for polished copper wafers typically contain triazoles to provide corrosion protection for the copper pattern. Copper slurries also contain triazoles and other organic components to form a passivation layer to control the chemical planarization process. The copper oxidation process involves sequential oxidation of Cu and cuprous oxide (Cu2O) to cupric oxide (CuO).
The most commonly used copper corrosion inhibitor is benzotriazole (BTA), which interacts preferentially with Cu2O. The triazoles, and BTA in particular, have been identified as a source of organic contaminates during the cleaning process. Other sources of organic contaminates include pad shedding, tool materials of construction, o-rings and seals, wafer boats, brushes, etc. However, BTA appears to be a common organic contaminate source. In addition to being used to passivate copper films, BTA is a component in most copper CMP slurries for the initial polishing steps.
During the copper CMP process, the wafers are often protected during staging for the next process step by spraying a dilute aqueous solution of BTA on the wafers and keeping them wet. After the CMP process, the wafers are often immersed in a dilute aqueous BTA solution, rinsed with DI water, and spun dry, then staged for the next process level (i.e., copper seed-layer deposition). The use of organic surfactants can result in copper and dielectric surfaces coated with organic materials. Prior to the Cu seed-layer deposition, the surfactant and/or BTA-copper complex must be removed to ensure the plating seed layer adheres to the previous Cu level.
Argon-ion back sputtering, a long used process to remove aluminum oxide from aluminum patterns prior to the next process step, has not been successfully used for copper. Contamination of the plasma tool with organic degradation products has limited throughput significantly. The use of plasma processing, either oxidative or reductive, to remove organic contaminates may be limited by the choice of dielectric. Some low-k dielectrics are more easily damaged by a plasma process than others. Therefore, it may be necessary to shift BTA removal to the post-CMP cleaning process to achieve an improved cost of ownership (CoO) and appropriate wafer throughput, and prevent damage to the dielectric. Post-CMP cleaners have been tested that significantly reduced organic contamination from the polishing process; optimization of both CMP process and cleaning tool can reduce organic contamination [1].
Eliminating watermarks
Watermarks can saturate metrology tools that measure light point defects (LPD). Saturation will mask killer defects. The cause of watermarks on OSG is not completely understood. Data from front-end-of-line (FEOL) processing of silicon indicates that watermarks are caused by etching of the silicon. Silicon and copper are hydrophobic. Silicon dioxide (SiO2) and copper oxides are hydrophilic. In FEOL silicon cleaning, the watermark could be either the pit remaining where the silicon was etched when the aqueous solution beaded up on the surface, or the insoluble oxidation products deposited on the surface of the silicon. The pH of the cleaner could have an impact on the production of watermarks. The generation of watermarks on silicon surfaces during rinsing with DI water is a function of pH and the presence of oxygen [2, 3].
Reducing the oxygen level in DI water has shown an impact on watermark formation [2]. At low pH, H2SiO3 (meta silicic acid) has reduced solubility in water and remains on the surface; silicic acid is soluble at higher pH values and does not precipitate on the wafer surface [3].
The mechanisms for watermark formation during back-end-of-line (BEOL) cleaning, where the exposed silicon is fully oxidized, are not presently understood. Unlike silicon at FEOL cleaning steps, dielectrics exposed at BEOL have the silicon fully oxidized and no further oxidation can occur.
Introduction of a surfactant to reduce surface tension and improve wetting of hydrophobic surfaces may be a solution to eliminate watermarks. However, the surfactant, which may not be easily removed, can become a source of organic contaminates. One can also introduce DI water-free drying methods, such as isopropyl alcohol (IPA) vapor drying. Utilizing IPA vapor drying in the post-CMP cleaning process will increase the CoO due to hazardous waste disposal costs for used IPA. This drying technique also could result in higher cleaning tool cost and necessitate fire suppression systems. Designing a cleaner to eliminate watermarks on hydrophobic OSG may offer the lowest CoO for the post-CMP process.
The experiments
Blanket copper seed wafers were treated with an aqueous 10ppm BTA solution (pH of 5.3) at room temperature (23°C) as follows: 2 hr static immersion; 30 sec DI water rinse; and an N2 dry. The copper seed wafers contained 1000Å of PVD copper. The BTA-treated wafers were cleaned with various commercially available post-CMP cleaners in a single-wafer, spin-spray, noncontact process tool at room temperature as follows: 60 sec chemical treatment at 100rpm; 30 sec DI rinse at 100rpm; and 30 sec N2 dry at 2500rpm.
A Scienta ESCA Model 300, equipped with a monochromatic Al kα x-ray source, was used to acquire XPS spectra. A take-off angle of 15° provided a sampling depth of about 25Å. At a take-off angle of 90°, however, the sampling depth was about 75Å. The ratio of nitrogen to copper was calculated from the N 1s and Cu 3p XPS spectra and used to estimate the thickness of the deposited BTA film and the percentage of BTA removed with the cleaner. The estimated errors were ±2% for the atomic percentages and ±5% for the N/Cu ratios. Patterned uncapped OSG wafers made with Coral dielectric material from Novellus Systems Inc. and a Sematech 854 test pattern were also examined on a KLA-Tencor Corp. AIT1 inspection tool and SEM to classify defects.
Unpolished blanket Coral wafers were used for watermark evaluations. The same noncontact cleaning tool and process parameters were used as those for the BTA removal studies. A KLA-Tencor SP1 system was used for whole wafer scans to look for the typical radial spin-out pattern of watermarks. Contact angles for DI water, pre- and post-cleaning, and for cleaning solutions were measured on a goniometer using hardware and software from Connelly Applied Research. Dilutions of surface preparation products were verified using a Model 330i WTW conductivity meter from WTW Measurement Systems Inc.
BTA and organic contaminate removal
Figure 1 contains contact angle measurements for unpolished Coral wafers for DI water as well as ESC 794 and ESC 797 surface preparation formulations from ATMI. The ESC products were diluted 20:1 with DI water. The diluted ESC products wet the surface of the hydrophobic OSG significantly better than DI water by itself, since the contact angles are about half of that for deionized water.
 Figure 1. Contact angle measured on unpolished Coral wafers. |
Table 1 contains x-ray photoelectron spectroscopy (XPS) data for copper and nitrogen obtained from copper seed wafers with various processes. The data includes untreated controls, samples treated with BTA, and samples treated with BTA followed by noncontact cleaning with various cleaners. The different take-off angles were run to estimate the BTA-Cu film thickness. Using a take-off angle of 15°, the ratio of nitrogen to copper was 2.95 ±0.15. The number of nitrogen atoms/BTA molecule is three, indicating that the BTA-Cu film is thicker than about 25Å. At a take-off angle of 90°, the ratio of nitrogen to copper was 0.56 ±0.03, indicating that the film was <75Å thick since a significant copper signal came from the metal below the BTA-Cu film.
 |
Several ESC cleaners that were tested (e.g., ESC 85A, ESC T794, and ESC 797) achieved complete removal of the BTA-Cu film with a one-minute, noncontact spin-spray process. One post-CMP cleaner, Product E, left a significant amount of BTA on the copper when used at its normal dilution of 50:1. ESC 784 also did not remove 100% of the BTA when used at 30:1 in a noncontact mode. However, using ESC 784 at 30:1 in either a megasonics bath or with brushes may have resulted in 100% BTA removal efficiency.
 Figure 2. Organic defect counts vs. ESC product and dilution. |
Data in Fig. 2 was generated by Rodel Inc. using an AIT1 inspection tool and SEM. This data shows the organic contaminate removal capabilities of several ESC cleaners at two dilutions. At a dilution of 30:1, ESC 794 and ESC 797 removed all organic defects from the Sematech 854 patterned Coral wafers. ESC 784 was slightly less effective in removing organic defects from the copper patterns, but was still better than for their previous process of record. At much higher dilutions (e.g., 120:1), ESC 794 was still quite effective in removing organic defects, indicating a very broad dilution latitude.
Copper passivation challenges
Removing the BTA could result in copper corrosion if the staging time between the CMP process and the next step was more than a few hours long. Device manufacturers prefer to either leave a monolayer of BTA on the wafer or replace the BTA film with another passivation film, which is more easily removed.
 |
Table 2 contains XPS data for passivating film thickness for several ESC cleaners — two of which contain copper passivators. ESC T794B and ESC T794M contain two different copper passivators, and after processing BTA-treated wafers, they left a very thin passivating layer on the exposed copper, but removed the previously deposited BTA film. ESC T794M provided a passivating film that was <<25Å thick, approaching a monolayer, compared to a typical BTA film that was <75Å thick.
Watermark results
Figure 3 shows SP1 whole-wafer defect scans of OSG blanket wafers. The scanned defects are: a) prior to polishing; b) post-polish and post-clean with Product E; c) post-polish and post-clean with Product A; and d) post-polish and clean with ESC 794. The polish and cleaning parameters were the same for all wafers tested.
The SP1 data for ESC 794 indicated an order-of-magnitude lower defect counts, compared to the pre-polish data, and more than two orders-of-magnitude reduction compared to Product E. ESC 794 showed more than an order-of-magnitude lower defect counts compared to Product A. The SP1 defect map for both Product E and Product A showed a typical radial pattern, indicative of watermarks. Several factors contributed to the lower defect counts for ESC 794, which included additives in the product and the detergency of one of the solvents. ESC 794 does not contain a surfactant.
Conclusion
Several surface preparation products have been tested that were capable of removing BTA from the surface of copper in a noncontact cleaning mode. The organic solvents used in the products allow them to remove organic contaminates even at high dilutions with DI water. The efficiency for removing organic contaminates may also be dependent on pH. The same products were also capable of preventing watermarks on hydrophobic OSG, possibly due to their wetting capabilities, since their measured contact angles were significantly lower than that for DI water.
Acknowledgements
The authors would like to thank Rodel for providing the organic defect data; and Ewa Oldak, Jeff Barnes, Chris Watts, Julie Jarrah, Lisa Sassaman, Mike Hughes, and Cuong Tran from the ATMI Surface Preparation Products Group for their contributions to this work.
References
- K. Bartosh, Y. Li, K. Cheemalapti, R. Chowdhury, M. Hughes, D. Peters, Surface Preparation Chemistries for CMP and Post CMP Applications, CAMP Conference, 2003.
- S. MacKinnon, Proc. Microcontamination, Vol. 94, p. 174, 1994.
- M. Wantanabe, et al., Materials Science and Engineering, Vol. B4, p. 401, 1989.
Darryl W. Peters received his BS in chemistry from San Diego State U. and his PhD in physical chemistry from Ohio State U. He is a senior scientist and acting R&D manager for the Surface Preparation Products Group at ATMI Inc., 2125 SW 28th Street, Allentown, PA 18103; ph 610/791-6916, fax 610/861-6932, e-mail [email protected].
Kyle Bartosh received his BS in chemistry from Pennsylvania State U. and is an application engineer in ATMI's SPP group.
Shahri Naghshineh received his BS from Birmingham U., England, and his MS in chemical engineering from Lehigh U., Bethlehem, PA. He is the director of ATMI's SPP group.
Elizabeth Walker received her BS in chemistry from Lebanon Valley College and is an R&D chemist in the SPP group at ATMI.