Surface Conditioning

Removing post-CMP residue through carbon dioxide snow cleaning

Andy Hakanson and Narendra Garg, Seagate Technology; and Michael R. Borden and Harlan F. Chung, Eco-Snow Systems

Using momentum transfer, CO^₂ snow cleaning removes the submicron particles that are left behind by brush scrubbing as well as other liquid and gas cleaning systems.

Cleaning processes for removing chemical-mechanical polishing (CMP) residues rely on double-sided brush scrubbers and the use of harsh chemistries (typically ammonium hydroxide and hydrogen peroxide). Unfortunately, post-CMP cleaning is expensive to carry out because it involves the frequent replacement of PVA or nylon brushes, the consumption of large quantities of DI water, and the disposal of chemical wastes.

As a result of the cleaning process itself, delicate surfaces can be scratched by particles trapped in the brushes, while residual slurry can cause reliability problems and early failure issues in both semiconductor devices and hard-drive heads. Moreover, the relatively large bristles of nylon brushes are not able to clean etched features (trenches) on the wafer surface.

In hard-drive head manufacturing, CMP is employed to reduce topography and planarize patterned layers, alumina (Al^₂O^₃) films, and films with various concentrations of nickel and iron. In semiconductor manufacturing, CMP is used to reduce topography and to polish tungsten, aluminum, copper, and silicon oxide layers. Apart from many process issues, the main concern with CMP is that the slurry it utilizes to polish wafer surfaces remains behind and is difficult to remove efficiently with double-sided brush scrubbing technology. In addition to potentially causing early device and drive failures, residual slurry can contaminate process equipment further down the line and adversely affect film adherence at later deposition and sputtering steps.

To overcome the limitations of carrying out post-CMP cleaning with double-sided brush scrubbers, Eco-Snow Systems (Livermore, CA) has developed the WaferClean system, a carbon dioxide (CO^₂) snow cleaning process that does not damage wafer surfaces, is environmentally friendly, and eliminates the need to dispose of hazardous wastes. Eco-Snow scientists developed CO^₂ snow cleaning as a remote, on-board cleaning process for removing contamination from space-based infrared surveillance system optics. This work was subsequently refined and extended to other cleaning applications, including post-CMP cleaning, through an extensive R&D process over a five-year period. This article describes the CO^₂ snow cleaning method, which, in addition to its particle-removal ability, can remove ion etch residue (fences) at various stages of magnetoresistive and giant magnetoresistive (MR/GMR) head manufacturing.¹

Generating Carbon Dioxide Snow

Carbon dioxide snow is created by spraying liquid CO^₂ through a small-aperture nozzle. The liquid rapidly expands and cools, resulting in the nucleation of solid CO^₂ particles ("snowflakes") entrained in a CO^₂ gas stream. Depending on the demands of the particular cleaning application, snow can be generated with the desired density, size, and velocity. Although many cleaning mechanisms exist, the primary mechanism by which the carbon dioxide snow method removes particles is momentum transfer, which results from the impact of snowflakes on particles attached to the wafer surface. This process loosens the particles, which are then removed by the accompanying gas stream. Primarily because of momentum transfer, CO^₂ snow cleaning can remove the submicron particles that are left behind by other types of liquid and gas spray cleaning systems.¹

Experiments were conducted on a WaferClean system installed at Seagate Technology's hard-drive head manufacturing facility in Bloomington, MN. The automated CO^₂ cleaning system incorporates two snow nozzles, each of which is customized for removing a specific type of contaminant. Three types of nozzles can be used: an aggressive one that delivers a low-volume, highly concentrated CO^₂ stream of very fast, very hard snow particles to a small surface area; a medium one; and a least-aggressive one that delivers a high-volume, diffuse CO^₂ stream of soft and slow snow particles.

In a series of repetitive movements, the snow spray is scanned across the wafer, the wafer is jogged forward incrementally, and the spray is scanned back across the wafer, resulting in a raster pattern. The wafer can be rotated to provide access to etched or trenched features that otherwise might be inaccessible to the CO^₂ stream. Programmable cleaning recipes enable engineers to choose from a variety of parameters, such as nozzle type, scan rate, step size between scans, the number of sides to be cleaned, raster size, and platen temperature. The interior of the cleaning tool is considered better than Class 1 and contains ionizer bars to minimize (if not eliminate) ESD-related issues. Airflow is carefully controlled and directed so that any debris at the bottom of the process chamber does not contact the wafer surface. The tool can be modified to handle different types of cassettes and wafers with different substrate sizes and can process round or square wafers of varying thicknesses and weights.

The Effects of Snow Cleaning on Various Surfaces

Removing Particles on a Trenched Layer. Initial experimentation on MR head wafers was performed on a patterned, nickel/iron (NiFe)­plated layer deposited over a sputtered seed layer of NiFe. The NiFe was plated over areas where photoresist, which had been coated on the seed layer, was absent. After the unexposed photoresist was removed and the NiFe-plated material was polished to reduce surface topography, the experiment commenced.

The slurry used for polishing this layer was Ultrasol 7A (STI, Monroe, NC), which contains silicon dioxide (SiO^₂) with an average particle size of 0.05 µm. In all applications, the platen temperature of the CO^₂ snow tool was set at 45°C and the nozzle's angle of incidence was 45°. A two-sided raster recipe was chosen for its speed and thorough cleaning ability. The medium-aggressive nozzle was chosen for all experiments, although tests showed that the more-aggressive nozzle is as efficacious as the medium-aggressive one.

Figure 1a shows a scanning electron microscopic (SEM) image of a trenched section of a head wafer that was polished and then processed through a double-sided scrubber. In all applications, the scrubber was optimized to perform the highest degree of cleaning possible. As seen in the image, the double-sided scrubber left behind residual slurry in the trenched areas. While the scrubber was able to effectively remove the slurry contamination on the plated material, it was unable to clean the seed layer in the approximately 3-µm-deep trenches. Figure 1b, an SEM image of the same area after snow cleaning was performed, shows no residual slurry. The CO^₂ cleaning process was able to remove particles <0.1 µm in size. An SEM analysis of multiple wafer sites, including both physical and topographical locations, indicated that no residual slurry was present.

The SEM image in Figure 2a is a close-up view of a trenched section of a head wafer that was polished and then processed through a double-sided scrubber. The layer in question is the same as that in Figure 1. Figure 2b depicts the same area as Figure 2a after the CO^₂ snow cleaning method was implemented, demonstrating that the slurry was completely removed. The image clearly demonstrates that this cleaning method effectively removes slurry particles <0.1 µm in size. In addition, the SEM image indicates that the NiFe-plated layer sustained no damage as a result of cleaning.

 

Figure 1: (a) SEM image of a trenched section of a head wafer that was polished and processed through a double-sided scrubber; (b) image of the same area showing no residual slurry after the CO2 snow cleaning technique.

Figure 2: (a) Close-up view of a trenched section of a head wafer that was polished and processed through a double-sided scrubber; (b) the same area after CO2 snow cleaning, demonstrating that the slurry was completely removed.

Another test of the CO^₂ snow method was performed after the NiFe-plated layer was polished with a slurry containing Al^₂O^₃ particles averaging 0.008 µm in size. Al^₂O^₃ can be much more difficult to remove in some instances than SiO^₂. A comparison of Figure 3a with Figure 1a shows that the double-sided scrubber was less effective at removing Al^₂O^₃ particles on the plated material and in the trenched areas than at removing SiO^₂ particles. On the other hand, comparing Figure 1b and Figure 3b demonstrates that the carbon dioxide snow cleaning procedure is just as effective at removing Al^₂O^₃ slurry as SiO^₂ slurry.

 

Figure 3: (a) SEM image illustrating that the double-sided scrubber was less effective at removing Al2O3 particles on the plated material and in the trenched areas than at removing SiO2 particles; (b) image of the same area showing that CO2 snow cleaning is just as effective at removing Al2O3 slurry as SiO2 slurry.

In addition to undergoing SEM analysis, the NiFe-plated layer was scanned by a Surfscan AIT patterned-wafer defect inspection tool (KLA-Tencor, San Jose) before and after carbon dioxide snow cleaning. The AIT recipe was optimized to inspect this layer and the trenched areas. The resulting analysis demonstrated that the CO^₂ snow cleaning method resulted in an average 55% decrease in total defects and a total absence of residual slurry. The remaining defects were CMP microscratches and other prefilm anomalies. Figure 4 shows the normalized results of the defectivity drop resulting from CO^₂ snow cleaning.

 

Figure 4: Normalized results of the decrease in defectivity after CO2 snow cleaning of a NiFe-plated layer.

Cleaning Surfaces with Sharp Topography. Another test was performed on a patterned NiFe layer that had been polished and scrubbed by a double-sided scrubber using the Ultrasol 7A slurry. However, this layer lacked closed trenches and contained large flat surfaces interrupted by sharp geometries that effectively trapped slurry. Figure 5a, an SEM image of the layer after it was polished and scrubbed, shows that while slurry had been removed from the plated material, it was present in the trenched areas and adjacent to feature walls. The same layer was cleaned with the carbon dioxide snow technique using a four-sided raster because of the layer's unique geometries. Figure 5b shows that this method succeeded in removing all of the residual slurry.

 

Figure 5: (a) SEM image of a layer containing large flat surfaces and sharp geometries with slurry in the trenched areas and adjacent to feature walls; (b) the same layer after CO2 cleaning showing that all of the residual slurry has been removed.

Post-CMP Cleaning of Air-Dried Wafers. To further demonstrate the capacity of the CO^₂ snow cleaning system, wafers were polished and air dried after the CMP process without being processed by a double-sided brush scrubber. The image in Figure 6a shows that slurry remained on both the plated material and in the seed layer trenches after drying. A comparison of Figures 6a and 1a demonstrates that more slurry was present on wafers that had been air dried than on wafers that had undergone brush scrubbing. In the latter case, while slurry remained on the seed layer in the trenches, it was removed from the plated material. In contrast, Figure 6b illustrates that the CO^₂ cleaning technique removes residual slurry from the NiFe-plated layer and the seed layer even on air-dried, unscrubbed wafers.

 

Figure 6: (a) SEM image of an air-dried, unscrubbed wafer showing residual slurry on the plated material and in the seed layer trenches; (b) image illustrating that the CO2 cleaning technique removes residual slurry from the NiFe-plated layer and the seed layer even on air-dried, unscrubbed wafers.

Cleaning Sheet Films on Wafers. In addition to cleaning difficult topographic patterned layers successfully, the carbon dioxide snow cleaning system can clean sheet films. Several NiFe-plated sheet-film wafers polished with Ultrasol 7A slurry were cleaned by a double-sided brush scrubber, scanned by a Surfscan 6420, cleaned with the CO^₂ snow technique, and then rescanned. Figure 7 shows that the application of the cleaning technique resulted in a 65.5% decrease in the defectivity rate.

 

Figure 7: The application of the CO2 cleaning technique resulted in a 65.5% decrease in the defectivity rate on NiFe-plated sheet-film wafers.

Another test was performed on sheet-film wafers with a NiFe seed layer that had been contaminated with a molybdenum disulfide (MbS^₂) antiseize compound. The wafers were scanned by a Surfscan AIT, cleaned with CO^₂ snow, and then rescanned. As illustrated in Figure 8, the defectivity rate decreased by 90.48% after snow cleaning. Auger analysis confirmed the complete absence of MbS^₂. Because the carbon dioxide snow cleaning technique is a dry process that leaves no residue behind, wafers cleaned by this method do not cross-contaminate other wafers because contamination is trapped in the prefilter or exhausted with the effluent gas stream. In contrast, a double-sided scrubber used for cleaning wafers contaminated with MbS^₂ would have to undergo brush or bath changing.

 

Figure 8: The implementation of the CO2 cleaning technique on sheet-film wafers contaminated with an MbS2 antiseize compound resulted in a defectivity rate decrease of 90.48%.

Cleaning Bare Substrate Material. The CO^₂ snow cleaning system was also shown to be an effective means for removing contamination from polished Al^₂O^₃ and bare substrates. Figure 9 illustrates that a bare substrate material composed of ceramic AlTiC, the starting material used to manufacture MR head wafers, experienced a defectivity drop of 48% after it had been snow cleaned. This decrease in defectivity is not as dramatic as that in the other tests in this study because the defect counts on the bare substrate material were low initially.

 

Figure 9: A bare substrate material composed of ceramic AlTiC experienced a decrease in defectivity of 48% after being cleaned with the CO2 snow technique.

Conclusion

Experiments demonstrate that carbon dioxide snow cleaning is more effective than double-sided brush scrubbing in removing post-CMP residual slurry. The performance of the CO^₂ snow method was not significantly affected by the types of nozzles, scan rates, or jog-step size employed. While both types of nozzles used in these experiments were equally able to remove loose contamination, in other tests the aggressive nozzle proved most effective in removing packed deposits of agglomerated slurry.

While the carbon dioxide snow technique not only outperforms the double-sided brush scrubber in cleaning post-CMP wafers, tests with nonscrubbed, air-dried wafers showed that it can eliminate the need for such a scrubber altogether. Moreover, contamination removed from the wafer by the CO^₂ cleaning system does not remain to cause subsequent cross-contamination, as is the case with double-sided brush scrubbers. While the carbon dioxide snow system cannot presently clean wafer backsides effectively, efforts are under way to develop a double-sided CO^₂ cleaning tool to address this issue. Another concern is the high cost, delivery pressure, and purity of the CO^₂ required for processing. To overcome this difficulty, a bulk, high-pressure liquid CO^₂ delivery system, which significantly reduces the tool's cost of ownership, is highly recommended.

Reference

1. T Kosic et al., "Cleaner Wafers with CO^₂ Snow," Solid StateTechnology (May 1998), S7.

Andy Hakanson is a process engineer in the wafer vacuum engineering group at Seagate Technology (Bloomington, MN) and is primarily involved in post-CMP cleaning, defect reduction, and film deposition. Before joining Seagate in 1998, he worked as a process engineer in the defect reduction group at Cypress Semiconductor. He has a BS in chemical engineering from the University of Minnesota in Minneapolis. (Hakanson can be reached at 612/844-8565 or andrew_hakanson@notes.seagate.com.)

Narendra Garg, PhD, is manager of the CMP group at Seagate Technology. Before joining the company 2 years ago, he spent 15 years at Cray Research in various positions. He received his BS and MS in metallurgical engineering at the Indian Institute of Technology in Kanpur and his PhD in materials science and engineering at the Polytechnic Institute of New York in Brooklyn. (Garg can be reached at 612/844-5786 or narendra_garg@notes.seagate.com.)

Michael R. Borden is a product line manager at Eco-Snow Systems (Livermore, CA), where he is responsible for the application, design, and development of semiconductor and MR head CO^₂ snow cleaning systems. Before joining Eco-Snow in 1998, he had more than 12 years of experience in the field of materials engineering, focusing on semiconductor processing, sol-gel, and infrared window technology. Borden is the author or coauthor of five publications and holds seven patents. He received his BS in ceramic engineering from Alfred University in Alfred, NY, and his MS in materials engineering from the University of California, Los Angeles. (Borden can be reached at 925/606-2000, ext. 311, or mborden@ecosnow.com.)

Harlan F. Chung is an applications scientist at Eco-Snow Systems, where he is responsible for the development, design, and application of CO^₂ cleaning processes for the MR head and semiconductor industries. He has 25 years of experience in the fields of semiconductor materials growth and process development. The author of 20 articles on materials growth and process development and the holder of three patents, he received his BA in biochemistry from the University of California, Berkeley. (Chung can be reached at 925/606-2000 or hchung@ecosnow.com.)

MicroHome | Search | Current Issue | MicroArchives
Buyers Guide | Media Kit