Paul W. Mertens, Katrien Marent, IMEC, Leuven, Belgium
SPECIAL REPORT: State-of-the-art processing
A major roadblock to single-wafer cleaning has been the lack of a fast, high-performance watermark-free drying technique. Now, IMEC engineers have developed two novel techniques that solve this problem. Both show excellent particle-addition control and high particle removal efficiency. In addition, these techniques have demonstrated excellent post Cu-CMP cleaning results.
Challenges associated with 300mm wafer processing, among other issues, are leading to different manufacturing concepts and cleanroom layout . Farm-layout cleanrooms with dedicated bays are being replaced by flow layouts that cluster different, consecutively performed process steps, including cleans. This often results in a reduction of wafer transport by as much as a factor of five.
Since most processes occur in a single-wafer mode, the introduction of clustered integrated single-wafer cleaning is a straightforward solution. In addition, single-wafer processing opens new possibilities for cleaning steps, such as single-side cleans and cleans requiring improved access to surfaces (e.g., for scrubbers, megasonic transducers, etc.). Integrated single-wafer cleans are already used for post-CMP (chemical mechanical polishing) cleaning.
In general, clustered cleans eliminate queue-time variations that reduce statistical fluctuations. The inspection and feedback loop is drastically shortened, reducing scrap. Most important, for the same throughput, cycle time and work-in-progress is decreased, resulting in a more cost-effective process. It also makes the manufacture of smaller ASIC volumes more flexible.
Clustered cleans have often been associated with dry cleaning. Wet cleaning largely outperforms dry cleaning, however . In addition, the vast know-how on batch cleaning can be transposed easily to single-wafer wet cleaning.
Required tight drying constraints
Some critical processing, such as pre-gate stack formation and CMP, put stringent time constraints on the pre- or post-process cleans.
Figure 2. Comparison of drying methods after 400nm CMP oxide removal.
To obtain matched throughput with state-of-the-art manufacturing, single-wafer cleaning and drying cycle time should be <2 min, preferably ~1 min. This is feasible for a wet clean, but drying becomes problematic. Indeed, a major roadblock for single-wafer wet cleaning is the availability of a fast (i.e., less than a few tens of seconds) and high-performance (i.e., watermark-free) drying technique. Traditional spin-drying often leaves drying spots on mixed hydrophilic-hydrophobic surfaces. Moreover, spin-drying can cause particle contamination from the ambient and also from residue left behind after evaporation of the liquid carry-over layer (i.e., the liquid remaining on the wafer that needs to be removed in the drying process). Possible alternatives are high-performance drying methods using isopropyl alcohol (IPA) vapor or surface tension gradient effects (i.e., Marangoni) . These techniques do not fulfill the tight timing constraints of single-wafer cleaning, however.
IMEC researchers tackled this challenge, developing two efficient drying methods, Rotagoni and Lineagoni, which address this issue.
Marangoni plus rotational forces
With Rotagoni drying (Fig. 1), a wafer held horizontally rotates at a speed typically ranging between 300 and 1000 rpm.
Figure 3. Pt concentrations, measured by total reflectance x-ray fluorescence, before and after two cleaning sequences using three cleaning chemistries.
A liquid, such as ultrapure water (UPW), is delivered to the wafer surface through a narrow dispense tube that slowly moves from the center of the substrate toward the edge at a set speed.
A second nozzle, on the trailing side of the liquid dispense tube, dispenses a tensioactive vapor that reduces the surface tension of the liquid and creates a strong Marangoni force.
The unique interaction of the Marangoni effect and rotational forces results in high-performance liquid removal; the liquid is physically removed by Marangoni force rather than evaporated. To ensure that each part is effectively dried, the dispense-tube speed is scaled to wafer rotation. The process can be integrated in a small-footprint spin cleaner, resulting in efficient liquid transport and contamination control.
Rotagoni dry performance
Tests on our first prototype demonstrated that 200mm wafers could be dried in 20 sec, using a mixture of nitrogen and IPA as the tensioactive vapor.
The performance with respect to particles has been studied by measuring light point defects (LPDs) on a Tencor Surfscan6400 and a Tencor Surfscan SP1-TBI.
In this work, we compared Rotagoni and conventional spin drying for particle addition from the ambient, doing our test in a Class 1000 cleanroom to magnify the effect. Our data showed an addition of ~200 LPDs on the spin-dried wafers vs. a reduction of ~20 LPDs with Rotagoni. We attributed the significant addition of particles from spin-drying to the strong air flow and turbulence created by the high-speed wafer spinning (i.e., typically 3000 rpm). Rotagoni, on the other hand, creates less turbulence above the wafer because it uses a lower rotational speed. In addition, Rotagoni removes any potentially contaminating liquid carry-over layer and, since there is no evaporation as in a spin-drying method, less residue is left.
In another evaluation, we subjected wafers with 1100nm-thick PECVD oxide to an oxide CMP, thinning the oxide to 700nm. The wafers were immediately transferred to a tank of UPW and then subjected to a standard spin-dry, a spin-dry in IPA vapor, and the Rotagoni process. Some wafers also received a spin-rinse spin-dry. We added a standard ~2-min scrubber-clean and dry process as a baseline reference.
Data from these tests showed that the spin processes leave more slurry residues on the wafer than the Rotagoni approach (Fig. 2), owing to the Marangoni force that physically sweeps away water containing slurry residues. The data also suggest that with spin-dry methods, at some point in the drying process, residual liquid is not physically removed, but evaporates and leaves residues. When we added a spin-rinse cycle, it helped to reduce residues because it effectively diluted the concentration of residue in the water layer left behind on the wafer surface prior to drying, but the spin-rinse cycle was still less effective than the shorter Rotagoni process.
We also evaluated the Rotagoni drying method for drying marks using a special cobalt-silicide short-loop process that is very sensitive to such residues. While we observed drying spots with an optical microscope on spin-dried wafers, we did not observe any on Rotagoni dried wafers.
Integrated single-wafer wet cleaning
In collaboration with Verteq Inc., we integrated the Rotagoni technique with the Goldfinger single-wafer megasonic cleaning system, evaluating the combination for the removal of platinum (Pt) particles intentionally deposited in a controlled way. (Pt is used as an electrode in ferroelectric memory applications.) Our tests involved three different proprietary cleaning chemistries with two different cleaning sequences a spin clean followed by a spin dry and a megasonic clean followed by a Rotagoni dry. We discovered that the spin clean combined with a spin-dry process showed almost no significant removal of Pt particles, irrespective of the clean chemistry. The megasonic-Rotagoni combination provided excellent cleaning (Fig. 3).
Figure 4. a) Silica particles after spin-drying and b) the results using a proprietary chemistry, megasonics, and Rotagoni drying.
We also simulated a post-copper (Cu) CMP cleaning application  using a single damascene deposition of Cu in an oxide layer. Cu electroplating was followed by Cu-CMP using silica-based slurry. Subsequently, the wafers were cleaned using a proprietary chemistry in combination with megasonics followed by Rotagoni. The wafers were also just spin-dried immediately after CMP to identify the contaminants present that needed to be removed. On these wafers, we observed silica particles preferentially deposited on Cu during post-CMP rinse (Fig. 4a). In the end, the megasonic process terminated with Rotagoni resulted in good particle removal (Fig. 4b). In addition, we did not observe any Cu corrosion or surface roughening.
Finally, we evaluated cleaning effects on resistance yield of Cu damascene "meander" structures (i.e., 21m x 0.2mm). The post-Cu CMP clean consisted of proprietary chemistry and megasonics followed by different drying methods. Resistance measurements showed that Rotagoni drying resulted in a 100% yield compared to ~94% yield after spin-drying.
Rotagoni is also characterized by a low chemical use. First estimations show a consumption of only one standard liter of nitrogen, <50mliter IPA, and ~100mliter UPW per 200mm wafer. In addition, its low rotation speed reduces forces on the wafer, making it more suitable for larger wafers. The drying time is <20 sec for a 200mm blank wafer, allowing single-wafer clean in 1 min. The method thus meets the criteria for single-wafer wet cleaning with high performance and low cost of ownership.
Lineagoni: Wafers through a slit
In the Lineagoni technique, a wafer is slowly pushed horizontally through narrow entrance and exit slits in a small box-shaped megasonic reactor (Fig. 5). The reactor contains the cleaning fluid, which can be UPW or other chemicals. Within the box reactor, the pressure above the liquid is less than the pressure adjacent to the reactor's slits. Pressure differential and tight slit dimensional tolerances prevent leakage. At the exit slit, the wafer passes through a tensioactive vapor that creates a Marangoni effect and dries the exiting wafer.
We have developed Lineagoni in collaboration with STEAG on its Damasclean platform, which combines a scrubber and a Lineagoni dryer. With the Damasclean platform, wafers are processed horizontally to allow accelerated wafer handling.
We tested the Lineagoni module by measuring metal concentrations on the surface of cleaned wafers. The module showed good metal neutrality with most metals below or close to the detection limit (see table).
We also evaluated 0.11mm particle addition for three different runs of wafers using the Damasclean platform. Results ranged from ~40 particles removed to ~40 particles added. To test particle removal efficiency of Damasclean, we contaminated wafers with ~5000 0.1-0.3mm silicon nitride particles. The removal efficiency was 99%, 98.7%, and 99.4% on three consecutive runs.
Initial tests showed that the Lineagoni drying time is 50 sec for 200mm wafers, which fulfills the timing constraint for single-wafer cleaning that we outlined above. We are nevertheless currently working on further reductions.
Two new drying methods developed by IMEC fulfill the tight timing constraints of single-wafer cleaning. Both show excellent particle-addition control and particle removal efficiency.
These drying techniques allow implementation of single-wafer wet-cleaning processes in semiconductor manufacturing.
Both approaches, each dependent on its own unique features, can be integrated in many applications.
Rotagoni and Lineagoni are trademarks of IMEC. Goldfinger is a trademark of Verteq Inc. Damasclean is a trademark of STEAG Industries AG.
- P. Kücher, "Lessons Learned from 300mm Conversion for Next Generation Manufacturing," Proceedings of European IEEE/Semi Semiconductor Manufacturing Conference, April 2000, Munich, Semi Technical Publications, Mtn. View, CA.
- M. Heyns, et al., "Advanced Wet and Dry Cleaning Coming Together for Next Generation," Solid State Technology, pp. 37-47, March 1999.
- K. Wolke, et al., "Marangoni Wafer Drying Avoids Disadvantages of Spin Drying or Alcohol Rinse," Solid State Technology, August 1996, pp. 8790.
- P.W. Mertens, et al., "A High-Performance Drying Method Enabling Clustered Single-Wafer Wet Cleaning," Digest of Technical Papers, 2000 Symposium on VLSI-Technology, pp. 56-57 (Widerkehr and Assoc., Gaithersburg, MD).
- W. Fyen, et al., accepted for publication in Proceedings of Symposium on Ultra Clean Processing on Silicon Surfaces, Sept 2000, Oostende.
Paul W. Mertens obtained his masters degree in electrical-mechanical engineering and PhD in applied sciences from the KULeuven. He joined IMEC in 1984 to work on silicon-on-insulator technology. Today, Mertens is group leader of the ultraclean processing group at IMEC.
Katrien Marent has an engineering degree in microelectronics. She joined IMEC in 1992 as analog designer and specialized in design of low noise readout electronics for high-energy physics. Marent is scientific editor at IMEC's Business Development Division, Kapeldreef 75, 3001 Leuven, Belgium; ph 32/16-281-880, fax 32/16-281-637, e-mail firstname.lastname@example.org.