Wafer Cleaning: Noncontact megasonics for post-Cu CMP cleaning

07/01/2000

Brian Fraser, Sana Rafie, VERTEQ Inc., Santa Ana, California
Mona Eissa, Somit Joshi, Texas Instruments, Dallas, Texas

overview

Post-copper-CMP cleaning presents challenges in addition to slurry removal. It is necessary to remove Cu contamination from some areas of the wafer (front-side dielectric, edge, and backside) while avoiding or limiting Cu removal from other areas (the Cu lines). Consequently, re-evaluation of the usual cleaning tools and chemicals is required. Megasonics offers an effective solution while eliminating problems such as particle loading and brush maintenance associated with brush scrubbers. Additionally, megasonics allows for a much wider chemical pH operating range. Batch and single-wafer megasonic tools allow for both integrated and stand-alone CMP processing.

VERTEQ's Goldfinger CMP system for noncontact single-wafer cleaningClick here to enlarge image

Although many variations to copper (Cu) integration schemes exist, all involve some form of damascene processing. Patterns are etched into the dielectric and filled with barrier and metal layers. Then, chemical mechanical planarization (CMP) is used to planarize and remove the Cu and barrier films over the dielectric. As with all CMP processes, a cleaning process is required afterward to remove the contamination introduced during polishing.

The CMP process uses alumina slurries to remove Cu and barrier metal, which is typically tantalum nitride (TaN). A damascene structure with Cu lines, barrier metal, and exposed dielectric remains. It is necessary to remove the slurry and particles from these areas, as well as from the wafer edge and backside. It also is necessary to remove Cu and other metal contamination from the dielectric on the front side, as well as on the edge and backside. And because Cu diffuses quickly in silicon and silicon dioxide, it is critical to remove it from the wafer to avoid cross contamination. But the cleaning process must not corrode the Cu lines, and it must minimize the amount of each film—dielectric, barrier, and Cu—removed.

Although Cu processing is currently in production, the Cu-CMP process is relatively immature. Post-CMP cleaning depends heavily upon the CMP process, and as a result, the clean process also is immature. Debates and evaluations are ongoing regarding the right method and chemical for post-CMP cleaning.

The post-Cu-CMP clean will take quite some time to mature because so many new cleaning issues are not faced with the other CMP cleans. Many new types of defects associated with metal corrosion and metal line topography have been, and will continue to be, seen Cu contamination will be a constant threat. As the CMP process matures, evaluation of the cleaning process will intensify, and demand for significantly improved cleaning performance will occur.

In addition, integration of more sensitive low-k materials also will be an issue.

Brush scrub, megasonic, or both?

For post-CMP cleans other than Cu, poly-vinylalcohol (PVA) brush scrubbers are the most common method for the clean step immediately following CMP. Megasonic cleaners, however, are becoming more widely used as replacements for brush scrubbers [1-4]. In addition, even when scrubbers are used immediately following CMP, a second clean is often done in a megasonic cleaner (typically a batch tool).

Using a megasonic clean step has three primary advantages:

• More effective removal of recessed and edge defects
• The ability to perform more uniform oxide etches
• Improved wafer-to-wafer defect removal consistency

Although scrubbers are effective cleaners for planar surfaces, residual defects in the recessed areas of the wafer topography exist even after CMP [5].

Figure 1. Impact of various chemical cleaning solutions on PVD Cu roughness (initial wafer roughness indicated as "seed Cu").Click here to enlarge image

Topography-related defects are removed far more effectively with megasonics. A simplistic look at the different cleaning mechanisms demonstrates the reasons for this: for a brush scrubber to remove particles, the brush must be in physical contact with the particle and wafer surface. This is easy if the surface is planar, but increasing topography results in decreased contact between the brush and wafer surface. The PVA material is flexible, and therefore, process parameters can be adjusted to improve brush-to-wafer contact. This method still has inherent limitations in terms of cleaning nonplanar surfaces, however.

The megasonic particle removal mechanism is certainly not completely understood. Many debates occur over the primary mechanism: cavitation, microstreaming, Schlicting streaming, or acoustic pressure gradients [6]. But regardless of which mechanism is most prevalent, particle removal from recessed areas is not inherently limited, as with brush scrubbers.

Cleaning chemistry

Ammonium hydroxide (NH₄OH) is the most common chemical used for post-CMP cleaning. Operating in the basic pH region is optimal to avoid particle reattachment while slightly etching the oxide film. In addition, this solves the problem of brush loading for brush scrubbers. This strategy does not work for Cu, however, because the Cu-NH₃ reaction causes excessive Cu surface roughening, and the Cu oxides and hydroxides do not dissolve, therefore causing brush loading.

The latter is not an issue with noncontact cleaning methods.

The Cu-NH₃ reaction in an oxidizing environment was reported 40 years ago [7]. Aqueous NH₄OH consists primarily of NH₃ gas dissolved in water. With oxygen dissolve in the cleaning solution serving as the oxidizer, excessive Cu dissolution occurs, as shown by:

Cu + O2 + 4NH₃ + H₂O --> Cu(NH₃)₄++ + 2OH-

Figure 2. Zeta potential for alumina and alumina-malonic acid.Click here to enlarge image

The Cu reaction occurs due to the reaction with NH₃, not with NH₄OH. Therefore, an alternative is to use a base similar to NH₄OH that does not contain dissolved NH₃ gas. Tetramethyl-ammoniumhydroxide (TMAH) is a common alternative for this application. Using NH₄OH with an agent that inhibits the Cu-NH₃ reaction is another alternative. Corrosion inhibitors such as benzotriazole (BTA) are well known. In addition, surfactants can act to inhibit the reaction while also providing cleaning benefits.

We evaluated the impact of the various chemical alternatives on Cu by dipping physical vapor deposited (PVD) Cu in several solutions and then evaluating surface roughness with atomic force microscopy (AFM); the results are shown in Fig. 1. The Cu dipped in NH₄OH was roughened significantly, to 14.4nm root mean square (RMS). Both the TMAH and NH₃-reaction inhibitor (surfactant and BTA) solutions had little effect on the initial Cu surface (an increase of <0.2nm RMS).

Figure 3. Cu wafers polished to FSG.Click here to enlarge image

Particle reattachment and brush loading also are problems when operating at an acidic pH. Below pH 8, alumina particles have a positive charge (zeta potential) while the oxide and brush surfaces are negative. The attraction causes dislodged particles to reattach to the wafer or brush surface. Rag-havan and Zhang [8] have demonstrated, however, that organic acids alter the surface of alumina particles and, consequently, their zeta potential (Fig. 2). Without altering the alumina particles, reattachment would occur below pH 8. The altered alumina particles, however, have negative zeta potential values above pH 4. This allows for the use of acidic cleaning solutions because particle reattachment does not occur in the acidic pH range.

It is possible to reduce metal contamination depending on the pH and chelating ability of the organic acid used. Etching is the only way to remove metal contamination within the oxide film, however. Consequently, fluoride is added to the solution if optimal metal contamination levels are required. Time of flight scanning ion mass spectroscopy (TOF-SIMS) evaluation of the Cu contamination level in the oxide between metal lines has been reported for fluorinated and nonfluorinated cleaning solutions [9].

CMP cleaners: integrated or stand-alone?

Currently, CMP and cleaning are done using either an integrated toolset (i.e., dry wafer in, dry wafer out) or individual stand-alone tools (i.e., wet wafers transferred from CMP tool to cleaning tool). The industry certainly is moving toward integrated systems for advanced processing and larger wafer sizes. For Cu processing, it is ideal to reduce the time between CMP and cleaning to avoid potential corrosion issues. However, corrosion issues can be solved otherwise, and because stand-alone tools can offer advantages depending on the fab situation, stand-alone operation will likely continue to be quite common.

Because wafers are processed one at a time in CMP, integrated solutions require that cleaning be performed one wafer at a time. This requires a tool different from the more common batch megasonic tools. Megasonic tools for both single-wafer processing and batch processing are presently available:

Click here to enlarge image

• The core of the batch processing tool [10] consists of two process stations—a tank for one-pass chemical processing and rinsing in the presence of a megasonic transducer, and a chemical recirculation tank when multiple-pass processing is more appropriate. Chemicals and deionized (DI) water are mixed in situ via injection into the megasonic tank. Rinsing is performed by a dump paddle in the tank bottom. The megasonic tool uses a curved piezoelectric transducer coupled to the interior of a chemically inert tube assembly. The diameter of the tube is smaller than the wafer, but the sound is transmitted radially to cover the entire wafer surface simultaneously. A spin-rinse or surface tension gradient dryer follows wet processing.

• The core of the single-wafer processing tool [11] consists of a megasonic assembly and a wafer spinner. The megasonic assembly is a piezoelectric transducer attached to the end plate of a quartz rod which serves as the sound transmitter, operating at ~830kHz. The rod (or transducer lens) is placed in a fixed position slightly above a horizontally positioned wafer, with the rod extending just past the center of the wafer. By applying liquid to the front surface of the wafer, a meniscus is formed between the wafer and transducer lens. This completes the medium path to transmit the sonic energy to the wafer surface. Liquid also is sprayed on the wafer's back surface to allow energy to pass through the wafer and remove particles on the wafer backside. The wafer spinner uses a multifinger chuck to hold and position the wafer so that it can spin about its center. Spin speeds range from <50rpm during cleaning to several hundred rpm during rinsing, and several thousand rpm during drying.

Post-Cu-CMP cleaning performance

We gathered post-Cu-CMP cleaning data for split lots between batch megasonic cleaning and brush scrubbing, and data from single-wafer megasonic cleaning. The batch megasonic cleaning versus the scrub cleaning was evaluated using particle and metal contamination data from polished wafers, as well as electrical data from serpent-comb structures. The same type of cleaning process was used for the two cleaning methods. A two-step clean consisted of a caustic solution followed by an organic acid fluoride mixture. The CMP process used a two-step polish followed by a buff.

Figure 4. Cu-TaN-TEOS wafers polished to the TEOS layer using a two-step polish without buff and a single-step clean using ESC775. Click here to enlarge image

Wafers with blanket Cu-TaN-FSG (fluorinated silica glass) films were polished to FSG and then processed through the clean. Comparisons using the same polish lot were performed. Defects >0.20µm were measured with a Tencor 6420. We found that the blanket wafer defect results were equivalent between the cleaning methods (Fig. 3). Metal contamination was evaluated for those splits by transmission x-ray fluorescence (TXRF). No contamination levels measured above 1010 atoms/cm², and Cu contamination was below detection limit (see table), showing acceptable contamination levels for both cleaning methods.

We evaluated the impact of the clean process on electrical performance using a serpent-comb test structure. The comb yield loss due to shorts is indicative of the effectiveness of the cleanup process in removing the metal residue. We evaluated the megasonic versus scrub clean split described above (same polish lot). The yields were essentially equal for the two cleaning methods — 90% for megasonic and 88% for scrub.

It is important to note that the defect levels are not solely a function of the clean. The dependence of the clean on polish conditions has been widely reported. The cleaning efficiencies listed here are dependent on polish parameters (slurry, pad, pressure, etc.). Although different polish parameters yield different clean performance, the scrub and megasonic cleans yielded equivalent performance.

Cu-CMP results

We evaluated the single-wafer megasonic cleaner using a different polisher and a different cleaning chemical process than those used with the batch megasonic and scrub tests. Split lot tests comparing the single-wafer megasonic to the scrub were not available. Performance was evaluated based on particle and metal contamination data from polished blanket wafers.

Figure 5. Metal contamination levels for Cu-TaN-TEOS wafers from Fig. 4.Click here to enlarge image

We used a two-step polish process without a buff. Both polish steps used alumina particle slurry. Instead of the two-step clean process used in the batch megasonic and scrub evaluation, a single-step caustic solution (ESC775) was used for the single-wafer megasonic clean. The solution contained one of the ammonium hydroxide alternatives and did not contain any fluoride components.

Blanket Cu-TaN-PE-TEOS wafers were polished down to the plasma-enhanced tetramethylorthosilicate (PE-TEOS) layer and then processed through the clean. Defects >0.20µm were measured with a Tencor 6420. These data showed the blanket wafer defect results to be primarily in the 20-40 defect range (Fig. 4). Metal contamination was evaluated for those same wafers using TXRF (Fig. 5). Although fluoride was not used to etch the oxide, the metal contamination levels (including Cu) were only slightly above detection limit.

Conclusion

We have shown an effective post-Cu CMP clean using a noncontact megasonic cleaner with a single step, nonfluorinated chemical process. Depending on metal contamination requirements, a fluorinated chemical can be added as a second step. In addition, the single-wafer cleaner allows for a stronger Cu etchant to be used on the backside while avoiding damage to the Cu lines on the front side. The megasonic method provides a simpler, less-expensive option to traditional brush scrubbing.

Acknowledgments

This article is based on work published in Ref. 12.

References

1. S. Roy, et al., J. Electrochem. Soc., 142,216, 1995.
2. B. Fraser, et al., Electrochem. Soc. Proc., 97-35, 634, 1997.
3. F. Tardif, et al., Solid State Phenom., 65-66, 169, 1999.
4. G. Willits, et al., Semiconductor Fabtech, 10, 301, 1999.
5. M. Olesen, et al., CMP-MIC Conf. Proc., 5, 530, 2000.
6. A. Busnaina, T. Kuehn, private communication.
7. Halpern, J. Electrochem. Soc., 421, 1953.
8. L. Zhang, et al., J. Electrochem. Soc., 146, 1442, 1999.
9. H. Li, et al., Micro, March 1999.
10. U.S Patent 5,656,097.
11. U.S. Patent 6,039,059.
12. M. Eissa, et al., "Post Copper CMP Cleaning: A Noncontact Megasonic Method in Chemical Mechanical Polishing in IC Device Manufacturing III," R. L. Opila et al., eds., Electrochem. Soc. Proc., PV 99-37, p. 499, 2000.

Brian Fraser received a PhD in chemistry from UCLA. He is the business development manager for single-wafer cleaning at VERTEQ Inc., 1241 E. Dyer Rd., Santa Ana, CA 92705; ph 714/445-2269, fax 714/445-2204, e-mail [email protected].

Sana Rafie received MS degrees in materials science and chemical engineering from the National School of Engineering, Lille, France, and the Florida Institute of Technology. She was formerly a process development engineer at VERTEQ, and is now a CMP cleans engineer at Texas Instruments-DMOS4.

Mona Eissa received her PhD in chemistry from UT-Dallas and a BS in chemistry from the University of Cairo. She is working on Cu cleans at Texas Instruments-Kilby Center.

Somit Joshi received his MS in materials engineering from the University of Central Florida. He currently is working on Cu interconnect development at Texas Instruments-Kilby Center.