Defect Analysis and Metrology

Characterizing wafer defects with an integrated AFM-DF review system

Larry M. Ge, formerly of Digital Instruments, Veeco Metrology Group and Monteith G. Heaton, Digital Instruments, Veeco Metrology Group

Atomic force microscopy can distinguish between different defect types and provide quantitative 3-D structural information on the defects at the nanometer level.

Driven by the critical need for product yield management and process control of deep submicron silicon device manufacturing, defect engineering has become increasingly important. Commonly used laser scattering defect inspection tools provide a wealth of information about defect size and defect distribution at very high throughput. However, such optical systems sometimes fail to provide accurate defect sizes and correct defect types because of the fundamentally limited nature of this technique. Two high-resolution defect review techniques, scanning electron microscopy (SEM) and atomic force microscopy (AFM), can be used to characterize deep submicron defects.^1,2 While SEM provides information on lateral defect size and shape, it does not provide quantitative height information and its resolution is limited for sub-0.1-µm defects. AFM, on the other hand, measures three-dimensional structures of individual defects nondestructively, providing accurate information on defect size, shape, and height. It has higher resolution in all three dimensions than does SEM.

Despite its advantages, AFM defect review must rely on the accuracy of the defect coordinates from scanning laser scattering inspection tools, because the first step of AFM-based defect review is the importation of a defect map. The error of the coordinates is often >50 µm for unpatterned wafers, which is larger than the maximum AFM scan size required to find ~0.1-µm defects easily. This limitation makes AFM-only defect review difficult and time-consuming. To surmount this limitation, an integrated defect review tool combining a dark-field (DF) optical microscope with an automated TappingMode AFM has been developed by Digital Instruments, Veeco Metrology Group (Santa Barbara, CA) for point-defect characterization on silicon wafers. This system's dark-field optical microscope can locate defect events precisely down to 0.1 µm and below (within 2 µm), and the AFM tip can subsequently be positioned on individual defects and image the defects at high resolution.

This article demonstrates the AFM-DF's capacity to distinguish between different defect types and to perform defect review of crystal-originated particles/pits (COPs) on silicon wafers. Moreover, it discusses the relationship between the physical defect size measured by AFM and the light-point defect (LPD) size determined optically by scanning laser inspection tools, showing that the correlation between the actual COP size and the corresponding LPD size is poor.

Experimental Procedures

Both Czochralski (CZ)-grown polished silicon wafers and epitaxial silicon wafers were used for this study. COP characterization was performed only on CZ silicon wafers. The CZ wafers were cleaned so that they contained a high percentage of COPs among all the defects. All wafers used in the study were inspected using a Surfscan SP1 or Surfscan 6XXX from KLA-Tencor (San Jose) prior to AFM characterization. Dimension series AFMs from Digital Instruments/Veeco with an integrated DF optical microscope were used. The DF optical microscope has fixed magnification with a field of view of ~250 µm. The DF microscope can resolve ~0.1-µm point defects on polished, epitaxial, and oxide wafers. AFM-DF defect review was performed in four steps:

1. A defect map generated by a scanning laser scattering inspection tool was imported.

2. Wafer edge registration was executed, followed by optional coordinate offset correction using two or more optically visible defects.

3. The defects were located with the DF microscope and then the AFM tip was positioned to the defect site to image the defects.

4. Different defect data, such as particle/grain size, cross section, depth, and bearing were analyzed.

Results and Discussion

Defect Type Identification. Identifying defect types based on the morphology of LPDs captured by scanning laser scattering tools can easily be done by the AFM-DF defect review system. Common types of defects on bare/epitaxial silicon wafers include point defects, such as particles, COPs, and their clustered forms; extended defects, such as microscratches; and crystalline defects, such as stacking faults and slip lines. All of these defect types were characterized.

Figure 1 shows three high-resolution AFM images of particle defects. The particles were flagged as 0.10-, 0.067-, and 0.082-µm LPDs, respectively. These images show that actual particle sizes differ from the estimated LPD sizes provided by laser scattering tools. This could be caused by the different chemical compositions of the particles, which result in different optical scattering properties that elude the binning capability of the laser scattering inspection tools.

Figure 1: Three 0.5 x 0.5-µm AFM images of particle defects flagged by laser scattering inspection tools as 0.10-, 0.067-, and 0.082-µm LPDs, respectively.

Figure 2 shows three 0.5 x 0.5-µm AFM images of COPs, which were flagged as 0.157-, 0.150-, and 0.068-µm LPDs, respectively. Figure 3 shows a slip line and two microscratches. Figure 3a shows that the silicon crystalline planes are slipped at several adjacent locations by different levels from ~0.5 to ~3.0 nm normal to the wafer surface. The scratches in Figures 3b and 3c were flagged as 0.20- and 0.10-µm LPDs, respectively.

Figure 2: Three 0.5 x 0.5-µm AFM images of COPs flagged by laser scattering inspection tools as 0.157-, 0.150-, and 0.068-µm LPDs, respectively.

Figure 4 shows three AFM images of stacking faults on epitaxial Si(100) wafers. They were flagged as 0.20-, 0.15-, and 0.13-µm LPDs, respectively. All three stacking faults are of large lateral size (5­6 µm) and have an extremely small vertical range (2­15 nm). The four edges of the square-shaped stacking fault are deeper than the center area, and the four corners are even deeper than the sides. The actual measured defect sizes strongly suggest that the laser scattering technique cannot estimate the defect size of low-aspect-ratio (vertical/lateral) defects such as stacking faults. That technique substantially underestimates their sizes because of their low-scattering cross section.

 

Figure 3: Three AFM images showing (a) a slip line, (b) a microscratch, and (c) a longer microscratch-type defect.

Various data analyses can be performed on the captured defects. Sectional analysis is often used. For example, Figure 5a is the cross-sectional view along the x-axis of the particle defect shown in Figure 1b; Figure 5b is the cross-sectional view of the COP shown in Figure 2a; and Figure 5c is the cross-sectional view of the stacking fault shown in Figure 4b. Vertical, lateral, and side-wall angles can readily be measured.

 

Figure 4: Three AFM images of stacking faults on epitaxial Si(100) wafers flagged by laser scattering inspection tools as 0.20-, 0.15-, and 0.13-µm LPDs, respectively.

Figure 5: Three cross-sectional views (along the x-axis) of the defects shown in Figures 1b, 2a, and 4b, respectively.

These examples demonstrate that the AFM-DF defect review system can be used to perform three-dimensional measurements and can identify and characterize deep submicron defects flagged by scanning laser scattering inspection tools. This system can also identify and classify defect types based on topography. Actual defect sizes measured by AFM-DF often differ from estimated defect sizes captured by scanning laser scattering inspection tools. Such errors, caused by the limitations of the optical scattering technique, are usually larger for low-aspect-ratio, extended, and clustered defects.

COP Characterization. COPs result from grown-in voids on CZ silicon wafers during wafer processing. It has been shown that they have a significant influence on gate oxide integrity.³ Reducing the density of COPs on CZ-grown silicon wafers has become critical for wafer manufacturers. Monitoring and controlling them has also become important for semiconductor device manufacturers. Special defect control methods should be used to eliminate COPs, because simple wafer cleaning processes not only enlarge existing COPs but can also create new ones. Laser scattering techniques cannot reliably distinguish COPs from other LPDs on production wafers, although progress has been made in distinguishing COPs from particles.⁴ AFM characterization of COPs provides rich information about their structure and can help both wafer and device manufacturers to develop effective defect control solutions.

Figure 6 shows six different COPs. Some of them show crystalline facets. This study found that relatively deep COPs, because they are well developed (or etched), show clear crystalline facets, while shallower ones, because they have not developed fully (or been etched), do not show clear facets. COPs often have a square or rectangular cross section parallel to the wafer surface, although other shapes also occur. The well-developed ones have a side-wall angle of four facets close to 45°, indicating that the pits have facets in (110) directions on Si(100) wafers. COPs sometimes show octahedral or triangular cross-sectional shapes parallel to the wafer surface (as shown in Figure 2), indicating the formation of facets in (111) directions in addition to the formation of facets in (110) directions. A high percentage of double (or twin) COPs have been observed. The AFM characterization of twin COPs in this study suggests that twin COPs often start with two separated small pits that grow larger and become connected when they are well developed. The pits of double COPs can be as small as 20 nm.

 

Figure 6: Six 0.5 x 0.5-µm AFM images showing a collection of different subtypes of COPs flagged by laser scattering inspection tools.

To investigate the relationship between optically and AFM-determined COP sizes, 60 COPs were chosen for analysis. Particle/grain size and bearing analysis were used to calculate the enclosed volume (V) and the lateral cross-sectional area (S) parallel to the wafer surface of those COPs. Effective COP size (D^_eff) was determined from volume (V = 1/6D^3_eff), while effective COP lateral size (L^_eff) was determined from the cross-sectional area (S = L^{_eff 2}). Figure 7 plots the COP sizes determined by the Surfscan SP1 against those determined by the AFM-DF. Both plots show a weak correlation between the physical dimensions of the COPs and the optically determined LPD sizes. The effective physical COP sizes deduced from the enclosed volume and the lateral cross section show a similar lack of correlation (R² ~ 0.6). The LPD sizes estimated by the SP1 have a larger spread than the actual COP sizes. This spread (or error) is likely due to the different COP shapes and to the occurrence of double COPs. Both can affect the scattering cross section and the accuracy of the LPD size estimation.

 

Figure 7: (a) Plot showing the relationship between LPD sizes determined by SP1 and effective COP sizes measured by AFM-DF and (b) plot showing the relationship between LPD sizes determined by SP1 and effective COP lateral sizes measured by AFM-DF.

Conclusion

A new integrated AFM-DF defect review system is a powerful tool for high-resolution, quantitative three-dimensional characterization of defects on silicon wafers. It is an accurate measurement instrument that can identify and classify various defect types such as particles, COPs, microscratches, stacking faults, and slip lines.

Acknowledgment

The authors wish to thank Luke Ghislain of Digital Instruments/Veeco for his valuable input during the early stages of AFM-DF product development.

References

1. Y Matsushita et al., in Silicon Materials Science and Technology, ed. HR Huff, H Tsuya, and U Gosele, Electrochemical Society Proceedings Series, PV 98-1 (Pennington, NJ: Electrochemical Society, 1998), 683.

2. N Adachi et al., in Silicon Materials Science and Technology, ed. HR Huff, H Tsuya, and U Gosele, Electrochemical Society Proceedings Series, PV 98-1 (Pennington, NJ: Electrochemcial Society, 1998), 699.

3. M Miyazaki et al., Japanese Journal of Applied Physics 36, (1997): 6187.

4. M Akbulut et al., "COPs/Particles Discrimination with a Surface Scanning Inspection System," Semiconductor International [on-line], April 1999, available from Internet: http://www.semiconductor.net.

Larry Ge, PhD, is a staff process engineer at Intel's California Technology Manufacturing group (Santa Clara, CA). Before joining Intel, he was a senior applications scientist at Digital Instruments, Veeco Metrology Group (Santa Barbara, CA). He received his PhD in solid-state physics in 1993 from the University of Hawaii in Honolulu. (Ge can be reached at 408/765-6689 or larry.m.ge@intel.com.)

Monteith G. Heaton is director of marketing at Digital Instruments, Veeco Metrology Group, a position he has held since 1993. He has published more than 100 articles in the fields of semiconductors, chemistry, environmental science, and materials science. In 1988 he received his MS in chemistry from the State University of New York at Stony Brook. (Heaton can be reached at 805/967-2700, ext. 293, or monte@di.com.)