Issue



Development of a CMP pad using an unpatterned surface inspection system


10/19/2012







C. Y. Cheng, S. N. Peng, S.C. Chen, Semiconductor Technologies, Dow Electronic Materials, Dow Chemical Company, Miaoli, Taiwan. Larry Yang, Debbie Hu, Steve Lin, KLA-Tencor Corp., Milpitas, CA, USA


Wafer haze information was used to develop advanced Cu CMP processes.


As design rules continue to shrink, copper chemical mechanical polishing (CMP) remains a challenging component of copper dual damascene processes. Traditionally, inspection of blanket wafers to determine how different factors influence overall defectivity has been used for CMP process development. In this study, wafer haze information is used to reveal defect signatures not apparent on standard defect maps. New methods of advanced haze analysis are used to measure haze defects on copper blanket wafers for characterization and development of CMP pads.


Introduction


Chemical mechanical polishing is a critical process for wafer surface global and local planarization in IC manufacturing. Copper (Cu) CMP has been widely reported as one of the leading techniques for Cu interconnect applications [1]. Blanket wafer inspection has played a key role in helping engineers optimize Cu CMP processes. In the past, the standard blanket wafer defect detection system for CMP process es focused on particle or scratch count and characterization. However, the standard methodology is no longer suitable for detecting anomalous process defects while technology nodes continue to scale down. Accessing defectivity information below standard thresholds can be achieved by wafer haze analysis, which represents a powerful tool for capturing spatial signatures caused by CMP processes [2,3]. As such, wafer haze information can be used in addition to standard defectivity data to optimize CMP processes and characterize CMP defects.


The unpatterned wafer surface inspection system used in this study (KLA-Tencor's Surfscan SP2) works by scanning a laser spot over the surface of a wafer with normal, oblique, or dual incidence and then collecting the scattered light into wide and narrow collection channels. From that scattering signal, both light point defects (LPDs) and haze can be extracted. The LPD information is used to generate a map of localized defects ??? the predominant information gathered from a system of this sort. The haze portion of the signal is often regarded as nuisance when localized defects are the aim of the measurement. As a result, grazing angle systems with sophisticated algorithms were designed to suppress haze. The wafer manufacturers and their users discovered that haze maps also contain important information because haze correlates with surface roughness. However, the value of haze information has been limited by lateral resolution and sensitivity, the lack of haze standards, and the visible wavelength employed by most unpatterned surface inspection systems. To address these limitations and make the best use of the information contained in the haze signal, a haze map of high resolution called Surfimage was developed, together with Surfmonitor, an application providing analysis capabilities [4,5].


The goal of this study was to examine how wafer haze information could help determine optimal process conditions for advanced Cu CMP processes.


Experiments


Four types of CMP pads from two batches were used in an experiment to find the best process conditions for minimizing defectivity related to the Cu CMP process.




Table 1. Experimental matrix

Cu wafers from two different sources were polished on an Applied Materials Reflexion LK tool. Table 1 summarizes the matrix of experimental variables that were designed to modulate the CMP defect levels. All wafers were measured by the Surfscan SP2 and analyzed with SURFmonitor to check overall defect counts and wafer haze.


Results: Defect count and haze


The post-CMP sum of all defect counts and the haze defect count were analyzed for each variable in the experimental matrix. These analyses showed that there was no statistically significant difference in the sum of all defect counts and haze defect counts by polishing head or wafer source.


For different pads, there were no statistically significant differences in the sum of all defect counts. However, the haze defect count for Pad 1 was statistically lower than that of Pads 2, 3 and 4 (Fig. 1a). The mean of Pad 1's haze defect count is 1451, which is much lower than the 2333 defects corresponding to Pad 3.





Figure 1. Post-CMP sum of all defect counts and haze defect count by pad and pad batch.
Figure 1. Post-CMP sum of all defect counts and haze defect count by pad and pad batch.

There was also no statistically significant difference in the sum of all defects counts by pad batch. However, the difference between batches can be identified when the haze defect count is examined (Fig. 1b). A smaller value of haze defect count is found in batch X compared to batch Y.


These wafer haze results identified pad and pad batch as factors influencing the Cu CMP process. In the next section, we examine different methods for analyzing the haze data in order to gain additional feedback for CMP process optimization.


Advanced analysis of wafer haze


The high-resolution haze map (SURFimage) generated by the unpatterned wafer inspection system can reveal new information about the wafer surface. The haze analysis application (SURFmonitor) includes algorithms that can define abnormal haze map patterns as distinct defect objects. These analysis capabilities include: anomalous process defect; grid analysis; and cross-sectional analysis. These haze analysis capabilities aid process development and production monitoring.


Anomalous process defects. An anomalous process defect (APD) is a nonuniform defect on a wafer caused by process variations in, for example, a CMP module.. APDs are not usually visible on the standard defect map but show up distinctly on a high-resolution wafer haze map. APDs can be extracted as defective areas when process conditions change and can be handled just as other standard defects on a wafer. They can be exported in a standard results file and statistical process control limits can be set for APDs in a defect data management system. By monitoring APDs, excursions can be captured early, avoiding the propagation of yield killer defects through the line.





Figure 2. Anomalous process defects (APD) show a strong signature on the edge of a Cu CMP wafer.
Figure 2. Anomalous process defects (APD) show a strong signature on the edge of a Cu CMP wafer.

Figure 2 shows the inspection results from one Cu wafer. While the standard defect wafer map shows very low defectivity, the SURFimage highlights surface nonuniformities, and the APD wafer map shows a clear defect signature at the wafer's edge.





Figure 3. The data flow for analyzing high-resolution haze maps with grid analysis, and then utilizing RBB to separate the grid cells by haze statistical values.
Figure 3. The data flow for analyzing high-resolution haze maps with grid analysis, and then utilizing RBB to separate the grid cells by haze statistical values.

Grid analysis. Within the CMP module, a grid analysis of the high-resolution haze map can be used to reveal new information about the localized wafer surface roughness. This is done by superimposing a grid on the wafer haze map with a user-defined cell size. Haze statistics are calculated for each grid cell. Rule-based binning (RBB) can be utilized to enable screening of process conditions with different haze levels. The defective grids are then flagged in order to help process engineers identify defects of interest. A production monitoring system is set up based on the grid to ensure that any drift in process condition is caught before product lots are run. Figure 3 shows the data flow for analyzing high-resolution haze maps with grid analysis, and then utilizing RBB to separate the grid cells by haze statistical values. This is a powerful tool for quantifying the haze differences among different process conditions.


The inspection results for the Cu wafers from the experimental matrix were analyzed using the grid analysis feature. Cu wafers were divided into approximately 2000 grids for higher resolution analysis and eight haze bins were created. Bin 601 (red bar in Fig. 3) marked haze values > 0.74, while Bin 608 (yellow bar in Fig. 3) marked haze values less than 0.59. Wafers with high roughness (haze values > 0.74) can be easily found after binning is applied, as shown in Fig. 3.


Cross-sectional analysis. Figure 4a shows within-wafer cross-sectional haze of wafers polished using Pad 1 and Pad 4, indicating a significantly different haze mean between Pads 1 and 4. The mean roughness of Pad 4 is ~0.80 nppm, higher than the 0.60 nppm of wafers polished with Pad 1. The roughness range on Pad 1 (0.18 nppm) is similar to that of Pad 4 (0.20 nppm). The roughness is much higher at the wafer edge on both Pads 1 and 4.





Figure 4. a) Cross-section of wafer roughness profile by pad; and b) cross-section of wafer roughness profile by batch. Both pad and pad batch show higher roughness at the wafer's edge.
Figure 4. a) Cross-section of wafer roughness profile by pad; and b) cross-section of wafer roughness profile by batch. Both pad and pad batch show higher roughness at the wafer's edge.

Figure 4b shows within-wafer cross-sectional haze of wafers polished using different batches, with a lower roughness mean achieved by Batch X. In addition, the mean roughness of Batch Y is approximately 0.80 nppm, higher than the 0.62 nppm of wafers polished with Batch X. Both batches demonstrated higher roughness at the wafer edge. The roughness range of Batch Y (0.18 nppm) is similar to that of Batch X (0.17 nppm).


Conclusion


In this study, a new methodology that uses high-resolution haze maps, rule-based binning and anomalous process defect assignments is implemented to effectively detect defects of interest and quickly optimize CMP process conditions for minimizing defectivity. While we found no statistically significant difference in the sums of all defect counts by pad or pad batch, the haze defect count for Pad 1 was statistically lower than that of other pads, and the haze defect counts of pads in Batch X were statistically lower than pads in Batch Y. The high resolution haze image also revealed anomalous process defects at the wafer surface. Grid analysis differentiated the pads and pad batches by mean roughness and generally revealed higher roughness at the wafer edge.


Acknowledgments


A more detailed version of this manuscript originally appeared in the 2012 23rd Annual Advanced Semiconductor Manufacturing Conference (ASMC) Proceedings, p. 128???131, 2012.


References


1. A. Steinbach, C. Pelissier, and U. Mahajan, "Surface scattering technique for micro-roughness characterization," NCCAVS CMP User Group Meeting, 2007.


2. C.Y. Cheng, Larry Yang, S.N. Peng, et al., "Development of CMP process condition characterization on haze defect," International Conf. on Planarization/CMP Technology (ICPT), 2011, pp. 159-164.


3. C.Y. Cheng, Larry Yang, S.N. Peng. et al., "Investigation of wafer haze defect for CMP," International Conference on Planarization/CMP Technology (ICPT), 2010, pp. 113-118,


4. R. Brun and C. Moulin, G. Bast, G. Simpson, and P. Dighe, "The use of high resolution haze for control of SOI surface roughness in a volume production environment," IEEE SOI Conference, 2010.


5. L. Ither and D.Pepper, "The use of high resolution haze for control of critical films surface roughness in a volume production environment," ARCSIS, 2010.


C. Y. Cheng is a metrology engineer (CheneyCheng@dow.com), S. N. Peng is a technology center lab manager, and S. C. Chen is a process engineer at Dow Electronic Materials, Dow Chemical Company, and Larry Yang is an application engineer, Debbie Hu is a regional product manager, and Steve Lin is an application manager with KLA-Tencor Corp.


Solid State Technology | Volume 55 | Issue 8 | October | 2012