Efficient, high-throughput post-STI CMP cleaning
12/01/2000
Andrew Black, Albert H. Liu, Philips Semiconductor, San Antonio, Texas
Linda T. Jiang, Diane Hymes, Lam Research Corp., Fremont, California
overview
A new high-productivity scrubber system provides precise control of chemical mixing and delivery. It enables using a single cleaning solution for the removal of all types of chemical mechanical polishing defects, Philips Semiconductor has successfully applied this system in production for post-STI CMP cleaning. The new process removes slurry particles, minimizes etch enlargement of STI CMP microscratches, and provides significantly higher throughput and lower cost of ownership compared to conventional scrubber systems.
Scrubber-based cleaning technology takes advantage of the synergetic force of mechanical scrubbing and chemical action. It uses a low volume of diluted cleaning solution, operates at room temperature, and features single-wafer processing [1]. It is thus cost-effective and environmentally friendly. Scrubber cleaning has been widely adopted for post- chemical mechanical polishing (CMP) cleaning of metal, dielectric, and polysilicon wafers [2, 3].
 Figure 1. Representative SEM images of cleaned STI patterned wafers: a) ammonia cleaned and b) HF cleaned. |
Conventional post-CMP cleaning uses a basic (high pH) cleaning solution, such as dilute aqueous ammonium hydroxide (NH4OH), to remove slurry residuals. If metal contaminant removal is required, the basic step is followed by acidic cleaning. The use of two cleaning chemicals in sequence increases the complexity of process equipment, decreases throughput, and uses more chemicals and rinse water. Consequently, moving to a single cleaning chemistry to remove all types of defects is highly desirable.
A new high-productivity scrubber developed by Lam Research enables using a single solution for removing all types of defects. This new cleaning system provides reduced costs and increased throughput. Chemical usage is reduced by 20%, and deionized (DI) water consumption is decreased by 48%, due, in part, to reduced idle-state water consumption. The new scrubber is applicable to all post-CMP cleaning.
Experimental design
Initially, to optimize and evaluate post-STI CMP cleaning, unpatterned wafers were processed on a Lam Teres polisher using Cabot SS-12 slurry. At Philips Semiconductor, we polished STI test wafers with an Applied Materials Mirra polisher using Rodel ILD 1300 slurry. All wafers were cleaned on a Lam OnTrak HPD Synergy scrubber, using the Lam-recommended process.
We gathered qualitative defect-level data immediately after CMP and post-CMP cleaning using scanning electron microscope (SEM) review and atomic force microscopy (AFM). Defect levels were measured quantitatively on patterned wafers with a KLA-Tencor Surfscan AIT and on blanket wafers with a Surfscan 6420. For etch rate tests, oxide film thicknesses were measured with Tencor Prometrix UV1280 and UV1250 optical metrology tools.
Tests on unpatterned wafers
At Lam, blanket oxide and nitride wafers were cleaned after CMP using aqueous solutions of 0.5% hydrofluoric acid (HF) or 0.5% ammonia. Defects on the polished and cleaned blanket wafers were measured with the Tencor Surfscan. For both HF and ammonia cleaned oxide wafers, the defect counts were <100 at 0.15mm for the oxide wafers and <50 at 0.20mm for nitride wafers. AFM images of the oxide and nitride films did not show any significant roughness differences between the HF and ammonia cleaned wafers. The AFM RMS roughness value was between 0.2-0.3nm for the oxide film and <0.15nm for the nitride film. These results show that dilute ammonia and HF were both effective at removing slurry particles from the blanket oxide and nitride surfaces.
Tests on patterned STI wafers
STI wafers are more difficult to clean than the unpatterned wafers because a direct STI CMP process requires stopping at a nitride layer with both nitride and oxide films exposed.
Since the CMP removal rate of oxide is greater than nitride, the trench oxide is slightly recessed [4], and CMP slurry particles become trapped along edges and corners of the trench at the nitride-oxide interface. It is difficult to remove these particles using conventional post-CMP cleaning methods such as ammonia scrubbing.
Microscratches are another type of defect often produced by STI CMP. The scratches are often formed on the trench oxide and are accentuated by the subsequent silicon nitride strip and pad oxide removal process. Enlarged microscratches may fill with silicon during a subsequent amorphous silicon deposition and may become conductive stringers, which are killer defects.
 Figure 2. Representative AFM images of STI patterned wafers: Top row cleaned with ammonia only; bottom row recleaned with HF. |
The STI CMP defects can be easily highlighted by conformal depositions and isotropic etch strip-clean processes. This can cause background noise during inline defect metrology and reduce the visibility of killer defects [5]. A post-STI CMP cleaning process must remove all slurry particles and, at the same time, minimize enhancement of microscratches.
At Philips Semiconductors, we performed two split-lot experiments on production wafers at the post-STI CMP clean process step. Half of the lot was cleaned with ammonia and half with HF. To examine defects on the STI wafers, SEM (Fig. 1) and AFM (Fig. 2) images were collected. The HF-scrubbed wafers had a clean, slurry-particle-free surface. By contrast, small slurry particles were observed on the ammonia-cleaned STI wafers. The particles were mainly located at the edges of the oxide-nitride interface (Figs. 1a and 2a). A quantitative comparison of ammonia and HF cleaning efficiency could not be made since 98% of the slurry residuals were 0.1-0.3mm, which is less than the 0.4mm detection limit of the production AIT recipe used.
Defect characterization
At Philips Semiconductors, we used the KLA-Tencor AIT to inspect defects on STI wafers. Microscratches tend to grow larger and deeper during the removal of the nitride and pad oxide, and slurry residues will be highlighted after amorphous silicon deposition. Therefore, to capture all defects and fully understand the impact of HF and ammonia cleaning, KLA-Tencor AIT scans and SEM reviews were performed at different stages of the manufacturing process. The defect counts are reported as the average of the five wafers in the five-wafer split. Similar defect trends were observed after each process step.
The total number of all AIT-detected defects after the amorphous silicon deposition was similar for the HF-cleaned wafers and the ammonia-cleaned wafers. The AIT data were sorted into size bins (Fig. 3). These data indicate that, compared to the ammonia-cleaned wafers, the HF-cleaned wafers had a larger number of smaller defects, but the same to significantly fewer larger defects. The bin No. 1 (smallest size/shape) defect count for the HF-cleaned wafers was 19% higher than the ammonia-cleaned wafers.
The defect types and origins were identified for all bins by Quest analysis (KLA-Tencor tool software) of AIT data and illustrated with auto defect classification (ADC), optical, and SEM review. Defects on the HF-scrubbed wafers were found to be microscratches. The ammonia-cleaned wafers had slurry residue as well as microscratches. The HF cleaning efficiently removed the slurry particles, while the ammonia clean did not.
The total defect counts were similar on the HF-cleaned and ammonia-cleaned wafers. The ADC and SEM review revealed that the slurry residues were removed after the HF cleaning. This indicates that the AIT detected more CMP-induced microscratch defects on the HF-cleaned wafers than on the ammonia-cleaned wafers. Oxide damage and microscratches are produced during STI CMP, and HF scrubbing may enlarge them and make them detectable to the AIT. Whether the surface defect becomes visible to the AIT or not depends on HF concentration and exposure time (scrub time). If no microscratches are present on the polished film after STI CMP, HF scrubbing will not add detectable microscratch defects, but will reduce slurry residues, and will, therefore, reduce the total number of defects. In general, the oxide film type (density and hardness), the initial defect level prior to CMP, and the CMP process parameters determine the amount of oxide damage and microscratches that are susceptible to being etch-enhanced and becoming visible to AIT.
Optimizing HF concentration, scrub time
The ideal way to reduce the defect levels on STI wafers is to optimize CMP to minimize both oxide damage and slurry residuals and to use a minimal post-CMP HF scrub to remove slurry residuals. For the HF scrub to be effective in removing slurry residue, a certain minimal amount of oxide must be etched away. It is important to minimize this oxide etching to prevent microscratch defect enhancement. To this end, we studied the HF etch rates of both high-density plasma (HDP) CVD oxide, and TEOS oxide films as a function of HF concentration and scrub time.
 Figure 4. Oxide removal versus scrub time with 0.5% HF. |
At Philips Semiconductors, we cleaned polished STI wafers with ammonia and measured the HDP oxide film thickness. The wafers were then scrubbed using 0.5% HF for varying times between 10 and 60 sec, and the oxide thickness was re-measured. The near-linear relationship of the oxide removal versus scrub time (Fig. 4) indicates that the HF etch amount can be readily controlled by varying processing time in the HPD scrubber.
 Figure 5. AIT defect counts (1s) after amorphous silicon deposition as a function of HF scrub time. |
The minimum etch amount required to completely remove all slurry residues was determined using AIT and SEM review. SEM review showed that a 10-sec HF scrub removed slurry particles smaller than 0.4mm, but did not remove all of the larger particles. AIT was used to examine the removal efficiency of large particles as a function of 0.5% HF scrub time. A split test of five HF scrub times ranging from 0-60 sec was performed. AIT scans were made after amorphous silicon deposition since the nitride strip and the silicon deposition steps highlight oxide surface damage defects. The defect counts (1s for each five-wafer group) are plotted in Fig. 5 as a function of scrub time. A 36-sec scrub or an HF etch of about 30-40Å of the HDP oxide film effectively eliminated slurry residuals, including the largest particles.
 Figure 6. Oxide removal versus HF concentration for a constant scrub time. |
To maximize throughput, oxide removal is best controlled by adjusting the HF concentration rather than by varying scrub time. The HPD scrubber used in this process provides the precise chemical mixing and delivery control needed. It precisely controls HF and DI water flow rates and delivery time. Using the new scrubber, tests have been performed to demonstrate how the HPD system can precisely dilute and control HF concentration and delivery time to provide the desired amount of oxide film etch.
Blanket TEOS oxide films were polished on a Lam Teres polisher and scrubbed using DI water. The initial oxide thickness was measured, and the wafers were rescrubbed using various concentrations of HF. The incoming HF concentration was fixed at 1%. It was diluted to 0.8%, 0.6%, 0.4%, and 0.2% on board using the HPD scrubber's chemical delivery system. Figure 6 is a plot of oxide removed versus HF concentration for a constant scrub time. By varying the HF concentration over the range of 0.2% to 1.0%, the amount of TEOS oxide film that was removed could be readily controlled to between 40 and 100Å. By changing the flow rates of deionized water and HF, as well as the initial concentration of HF, a wide range of HF concentration and oxide removal can be readily applied to the wafer. The actual oxide removal rate is also dependent on the properties of the oxide film.
A marathon test of 800 wafers was performed to determine HF etch uniformity. It was found that wafer-to-wafer nonuniformity was 10% and within-wafer nonuniformity 6%.
Device performance
Measurements of transistor parameters showed no significant differences between HF and ammonia post-CMP scrub processes (Table 1). All parameters were the same within ±2.5%, except for n-to-p well leakage, which differed by 7%. The absolute leakage is very low measured in pA so this difference is most likely due to normal test variation.
Device yield
A review of wafer sort maps for the two split lots revealed that yield was impacted by a known defect caused by a subsequent process step. Therefore, no conclusions could be made about the impact of post-CMP HF cleaning on the device yield of STI wafers. However, based on the known reduction in slurry residue and the negligible effect on device parameters, the effect on yield is expected to be positive. Similar work on post-CMP interlayer dielectric (ILD) cleaning of silicon-rich oxide films showed that, although an HF clean increased the KLA-Tencor AIT counts by a factor of four, the yield actually was increased by approximately 1% [6]. It appears that for ILD, large residual slurry particles have a greater negative impact on yield than microscratches. The impact of HF clean on the yield of STI wafers is yet to be determined.
 |
Cost of ownership
The HPD HF-only clean process increases throughput by 33% and reduces the consumption of chemicals, deionized water, and waste water that must be treated (Table 2). All this contributes to a lower cost of ownership.
 |
Conclusion
To achieve low defect levels, the post STI CMP cleaning process must remove all slurry residues and ensure minimum enhancement of oxide damage. The scrubber cleaning process using dilute HF showed effective removal of STI CMP slurry residues. Our experiments showed that a dilute HF scrub clean is efficient at removing slurry particles from even slightly nonplanar surfaces typical of direct STI CMP processes. However, if either the HF concentration or clean time is too high, the isotropic oxide removal will etch and enlarge any oxide damage defects caused by the CMP process. Careful attention to HF concentration and clean time during process qualification is essential for an effective post-CMP clean for STI. When HF concentration and scrub time are optimized, throughput is maximized and defectivity is minimized.
Acknowledgments
Teres and Synergy are trademarks of Lam Research Corp. Mirra is a trademark of Applied Materials Inc.
References
- W.C. Krusell, J.M. de Larios, J. Zhang, "Mechanical Brush Scrubbing for Post-CMP Clean," Solid State Technology, Vol. 38, No. 6, p. 109, 1995.
- L.T. Jiang, et al., "The Development of Chemical Mechanical Polishing and CMP Cleaning Process for Low-k Films," CMP-MIC, p. 278, 1999.
- D. Hymes, et al., "The Challenge of Copper CMP Clean," Semiconductor International, Vol. 21, No. 6, p. 117, 1998.
- J.Y. Cheng, et al., "Shallow Trench Isolation," J. Electrochem. Soc., 144, p. 315, 1997.
- J.J. Li, A.H. Liu, S.S. Hiemke, "Defect Detection Strategies for Chemical Mechanical Polishing Process in Shallow Trench Isolation Application," SPIE Microelectronic Manufacturing — The International Society for Optical Engineering, Sept. 22-23, 3884, p. 36, 1999.
- Andrew Black, Philips Semiconductor (internal report).
Andrew Black received his BS in physics from the University of Nebraska and his MS in management technology from the University of Texas. He is a senior process engineer at Philips Semiconductor, 9651 Westover Hills Blvd., San Antonio, TX 78251; ph 210/522-7045, fax 210/522-7302, e-mail [email protected].
Albert H. Liu received his BS in physics and MS in electrical engineering from the Texas A&M University. Liu is process engineering manager at Philips Semiconductor.
Linda T. Jiang received her BS from Fudan University, China, and PhD from the University of Toronto, Canada. Jiang is a senior process engineer at Lam Research, Fremont, CA.
Diane Hymes received her MS and PhD in materials science and engineering from Brown University. Hymes is director of product marketing, CMP/Clean Product Division at Lam Research in Fremont, CA.
 |
Residual slurry defects removed
HPD HF-HF scrub implementation yields clean wafers after CMP, and optimizes the polishing process and consumable sets to reduce surface damage. An HF clean, by itself, may highlight surface damage induced by the polishing process (microscratches) due to HF etching properties. The HF scrub process described here effectively removes slurry particles and minimizes etch-enhanced microscratches at a high throughput and reduced cost of ownership.
Miguel Delgado, director of yield engineering enhancement, Philips Semiconductors, San Antonio, TX