PVA brush design advances for Cu/low-k post-CMP cleaning apps
07/01/2008
To meet next-generation challenges, polyvinyl alcohol (PVA) brush designs may need further optimization in design and materials, and many of the cleaning chemistries and approaches will have to change. There are plenty of potential solutions being considered for 32nm and smaller feature devices, and IC manufacturers will need to adopt new methods and new etch chemistries and cleaning regimens to meet the most stringent requirements of advanced applications. This paper presents study results that highlight the importance of PVA brush design, as well as methods of tribological and post-chemical mechanical polishing (PCMP) cleaning evaluations to ensure consistent wafer cleaning performance over brush lifetime. The benefits of using molded-through-the-core (MTTC) design PVA brushes are also demonstrated.
Chemical mechanical planarization (CMP) process performance and yield depends on the effectiveness of the PCMP cleaning. This process step should reduce the roughness of a polished wafer surface and leave it defect-free by consistently removing particles, organic residues, and ionic contamination. Recently developed MTTC polyvinyl alcohol (PVA) roller brushes with an integral/disposable core and adhesion of PVA with the core provide positive anchoring of PVA with the core and dimensional stability, which eliminates any possibility of PVA slippage at the PVA-core interface. Stable behavior of brush-wafer contact pressure, contact area, and skin friction could be useful indicators of the brush cleaning effectiveness and mechanical consistency.
Tribological test data obtained using a 200mm wafer test tribometer (in accelerated 48-hr continuous stress evaluations) are discussed for different brushes. Also presented are the results of PCMP cleaning in wafer defectivity classification (generated in a 90nm production fab, using 200mm blanket and 180nm feature MIT854 patterned wafers on a Mirra Mesa tool set), using MTTC and slip-on-the-core (SOTC) brush designs–during the brush break-in, scrub-only, and PCMP cleaning cycles, and after Cu/low-k barrier step CMP.
Wafer cleaning attributes and trends
It appears that the next-generation PCMP cleaning process will continue to depend on PVA brush-based cleaning together with megasonic cleaning in the foreseeable future. Based on recent literature, the requirements of surface cleaning must be considered while designing future generations of integrated circuits (ICs), since 60% of fab-related (yield) problems are related to cleanings and another 12% to etching steps. In addition, the design dominates how wet processing is done and processing limitations influence the design process.
To meet next-generation challenges, PVA brush designs may need further optimization in design and materials, and many of the cleaning chemistries and approaches will have to change. Plenty of potential solutions are being considered for 32nm and smaller feature devices, and the IC manufacturers will need to adopt new methods and new etch chemistries and cleaning regimens to meet the most stringent requirements of advanced applications. Suggested non-damaging nanoparticle removal technologies include
- Improved performance non-contact megasonic cleaning;
- Shock tube-enhanced laser-induced plasma (LIP) shockwaves for sub-50nm nanoparticle removal;
- Plasma-assisted cleaning by electrostatics (PACE);
- Ionized molecular-activated coherent solution; and
- Parametric nanoscale cleaning by forming nanoscale bubbles to absorb the contaminants.
On photoresist issues, several new or enhanced methods may be used for minimizing silicon and oxide loss during removal, including photoreactive cleaning, CO2 cryogenic press, and non-oxidizing chemistry and methodologies for all-wet chemistries. Current PCMP wafer cleaning processes are contact cleaning techniques, which use chemical and mechanical action to effectively remove the particles from the wafer surface.
Brush cleaning is an effective PCMP cleaning technique, and in an optimum mode, a contact between the particle and the brush is essential to the removal of submicron size particles from the wafer surface. In this operation, Rm>1 for a 0.1µm particle, for typical brush and wafer speed, based on a recent study performed at Northeastern U., where Rm is the ratio of removal moment to adhesion resisting moment, and most of the particles are removed by the drag force instantaneously if Rm>1 [1].
In the above test, 100% particle removal could be achieved, employing intermediate brush pressure, speed, and cleaning time. This study showed that brush cleaning is effective in removing particles down to 0.08µm with different PCMP clean chemistries.
CMP processes use abrasive slurries for planarization. After CMP, the wafers need to be cleaned to remove the slurry abrasive, organic residues, and other particles. This PCMP cleaning is accomplished by employing different tools and clean chemistries. Advanced CMP tools have integrated PCMP modules, enabling the wafer cleaning cycle to be dry-in and dry-out, to prevent contamination.
During the PCMP cleaning using brush, the PVA is compressed when it contacts a particle adsorbed on the surface of the wafer. Pores and asperities on the surface of the PVA brush capture the particle and cause the exposed surface of the particle to adsorb on the surface of the brush (mechanically, chemically, or by capillary suction). The torque created by the rotation of the brush subsequently dislodges the particle from the surface. The fluid present on the wafer surface, and being pumped in and out of brush pores (during compression and elastic recovery of the brush), carries the particle away from the wafer. The PCMP cleaning chemistry is typically sprayed on the brush-wafer interface, with de-ionized (DI) water flowing out through the brush core and PVA layer. In some applications, pre-diluted PCMP chemistries are supplied through the core. PCMP cleans may range from extremely acidic (pH<2) to very basic (pH>11), and may contain surfactants, chelating agents, and other additives.
The cleaning performance of a PVA brush strongly depends on the chemical and mechanical properties and stability of the brush material, magnitude of wafer-brush frictional force, and the adhesion forces between the particle and wafer, as well as the particle and brush. Zeta potentials of the particle and the wafer in various cleaning solutions, and pH of the CMP slurry, are very important in the particle adhesion and removal in a PCMP cleaning process. A combination of chemical action (provided by cleaning chemistry) and mechanical action of the rotating PVA brush removes the wafer surface deposits.
With NH4OH at pH ~10-11, PVA brush, wafer, and the slurry abrasive particles all have similar negative zeta potential. This condition would result in repulsion between PVA and particles, whereby no particles deposit on PVA or on the wafer, and leave no scratches. An efficient copper PCMP process not only removes particles, organic residues, and ionic contamination, but also controls the copper corrosion, prevents water marks on dielectric, and leaves the polished surface free from all defects, providing consistent process throughput and lower cost of ownership. Comparative frictional attributes as determined by wafer-liquid-brush interface coefficient of friction (COF), and mechanical integrity in terms of contact pressure and contact area, can provide valuable insights on the lubricious behavior of wafer-brush contact and brush lifetime.
 Figure 1. Close-molded PVA roller brush working surface with nodules. |
PVA brushes used to be an industrial product before being introduced at IBM and commercialized in the early 1990s. MTTC brush design (Fig. 1; Entegris Planarcore) is a disposable PVA brush that reduces tool downtime and provides dimensional stability over its lifetime. This design provides positive anchoring of the PVA brush with the core and eliminates any slippage at the PVA-core interface. MTTC design also provides efficient core flow equalization, resulting in consistency throughout brush lifetime in the PCMP cleaning performance.
With PVA brushes being part of smaller feature devices’ PCMP cleaning, next-generation brushes would require low particle concentration and minimal extractable PVA and innovative designs for cleaner and less stress-inducing processes. Cleaner PVA should help in reduced particle counts/adders on the wafer, decreased defectivity, and reduced tool downtime. Furthermore, specific applications may require charged, IPA-resistant, Cu/low-k specific, and/or next-generation CMP slurry and PCMP clean chemistry-compatible brush technology. More stringent process consistency requirements over brush lifetime may require innovative products such as MTTC brushes. This design provides ease-of-use and time savings, excellent dimensional stability, and uniform out-of-the-brush flow along the brush PVA length.
Results of PVA brushcharacterization studies
Study #1: Comparative tribological performance evaluation. Factors affecting brush PCMP cleaning efficiency include contact pressure at the brush PVA nodule surface and the wafer, physical and chemical properties of cleaning fluid and its flow rate, overall kinematics of the brush in relation to the tool, cleaning time, mechanical properties of the brush PVA, magnitude of frictional forces between wafer and brush relative to magnitude of adhesion forces between the particle and wafer, and the particle and brush PVA. The above information is critical in predicting brush performance consistency. At the U. of Arizona, an accelerated tribological stress evaluation (48-hr marathon run) of PVA brushes was conducted employing two SOTC design brushes, A and B, and one MTTC design (brush C). This case study addressed how the extent of brush deformation and the magnitude of frictional forces vary as a function of extended use for different brushes.
 Figure 2. Tribological test results of brush pressure versus brush-wafer contact area for brushes a) A, b) B, and c) C. |
The tribological tests showed how the extent of brush deformation, as measured by the brush-pressure versus brush-wafer contact-area curves (Fig. 2a-c), evolve as a result of extended use for various types of brushes. The enveloped area bounded by the curves shows the extent of brush deformation. Those brushes that experienced the least amount of deformation variability during the 48-hr marathon test also exhibited the least amount of variability in frictional attributes.
 Figure 3. COF mean and total fluctuation range during a 48-hr accelerated stress tribological test for brushes A, B, and C. |
Figure 3 presents COF mean and total fluctuation range during a 48-hour accelerated stress tribological test for brushes A, B, and C. The magnitude of frictional forces (as measured by the brush-fluid-wafer COF) significantly vary as a function of extended use of various brushes. The results showed a very different behavior of wafer-liquid-brush contact pressure, contact area, and dynamic COF for different brushes. Brushes A and C showed a more consistent behavior of mean COF, whereas brush B experienced catastrophic failure somewhere between 2 and 8 hours (Fig. 3). Also, the total variation range of COF for MTTC brush C was the smallest as compared to the other two SOTC brushes. The experimental conditions and equipment details for this study are included in the following. All tested PVA roller brushes had similar dimensions, were commercially available, and had cylindrical nodules.
Experimental conditions and equipment
 |
The conditions applicable to case study #1 are listed in Table 1.
 |
Study #2: PCMP cleaning performance in Cu/low-k application. The objective of this study was to generate comparative PVA brush PCMP cleaning data (defect maps/classification) for newly developed MTTC design brushes, and a third-party POR SOTC brushes in a 90nm production fab, using 200mm blanket and 180nm feature MIT854 Cu/low-k patterned wafers on a Mirra Mesa CMP tool and PCMP cleaner set.
Tested brushes and equipment
The equipment/brushes used in case study #2 are listed in Table 2.
 Figure 4. Defectivity classification data for a) POR brush type and b) an Entegris PVA brush type. |
Defectivity classification data for the above two brush types are seen in Fig. 4. The PCMP cleaning defectivity performance of MTTC design brushes was comparable or better than the SOTC design brushes.
Conclusions
The stable behavior of PVA brush-wafer contact-pressure, contact-area, and dynamic-friction could be useful indicators of PCMP cleaning and mechanical consistency of brushes over their lifetime. Accelerated tribological stress evaluations (48-hr continuous marathon tests) of three PCMP clean PVA brushes, including two SOTC design brushes (types A and B) and one MTTC design brush (type C), demonstrate a different behavior of wafer-liquid-brush contact pressure, contact area, and dynamic COF.
Brush A and brush C showed a more consistent behavior of mean COF, whereas brush design B experienced catastrophic failure somewhere between two and eight hours. The total variation in the range of COF for brush C was least among the three brush types. Those brushes that experienced the least amount of deformation variability during the 48-hr test also exhibited the least amount of variability in their tribological or frictional attributes. The PCMP cleaning comparative evaluation of the above MTTC brushes in a Cu/low-k process was found to be similar or better than the fab POR SOTC brushes in a third-party evaluation.
This study demonstrates the importance of PVA brush design and the methods of tribological and PCMP cleaning performance characterization of the brushes for ensuring consistent frictional characteristics and wafer cleaning behavior throughout the brush lifetime. ??
Acknowledgments
The authors would like to thank Drs. Ara Philipossian, Ashwani Rawat, Ahmed Busnaina, Robert Donis, and Peria Gopalan for insightful discussions, and Craig Brodeur and Scott Moroney for their contributions to this study. Planarcore is a registered trademark of Entegris.
Reference
- Communication with Dr. Ahmed Busnaina of the NSF Center for Microcontamination Control at Northeastern U., Boston, MA.
Rakesh K. Singh received his PhD from the U. of Manitoba and MBA from U. of Massachusetts Amherst and is manager of the WW CMP applications group at Entegris, 129 Concord Road, Billerica, MA USA; ph 978/436-6556; e-mail [email protected].
Christopher R. Wargo received his BSChE from Carnegie-Mellon U. and is VP of the new business development group at Entegris.
David W. Stockbower received his MS and BS degrees from UMass Lowell and is a product manager for CMP products at Entegris.