Single crystal silicon wafers are the fundamental elements of the semiconductor manufacturing sector. The Czochralski (CZ) process produces wafers that are high-quality, single crystalline materials with known defects that are modified by further processing or formed during the crystal growth.
Defects can be unfavorable for yield in some manufactured electrical devices, but groups of defects like oxide precipitates can have both negative and positive impacts on the final device. Scattering techniques can be used to detect the spatial distribution of these defects.
However, many crystal defects are either not detected or poorly classified due to limitations of scattering (i.e., light wavelength). As a result, it is essential to have high throughput and accurate characterization of their dimension and shape to evaluate the defects and proper classification.
Although high-resolution two-dimensional (2D) images can be obtained with scanning electron microscopy (SEM), atomic force microscopy (AFM) provides the highest vertical resolution among all techniques, and is required to obtain three-dimensional (3D) information of the defects of interest (DOI).
However AFM has several disadvantages, including limited tip life, low throughput, and laborious efforts to locate the DOI when it comes to defect review of 300 mm wafers. These limitations of AFM have been addressed with the introduction of automatic defect review (ADR AFM), which has been used in this work for studying DOI on 300 mm silicon wafers.
In this analysis, a 300 mm silicon wafer was carefully etched with a gaseous acid in a reducing atmosphere at a temperature and for an adequate period of time to decorate and grow the crystal defects to a size that can be detected as light scattering defects.
Laser light scattering (LLS) was used to inspect the distribution and relative size of the shallow structures formed by the etched defects. However, the LLS could not properly size some groups of defects due to low light scattering and very shallow depth.
Similarly, post-inspection defect review and classification of these very shallow types of defects cannot be achieved effectively with the SEM technique. Therefore, ADR AFM was used to accurately locate and image the DOI as well as to verify and acquire accurate shape and 3D information of those defects.
In ADR AFM, non-contact mode imaging is employed for non-destructive characterization and to preserve the sharpness of the tip for data reproducibility and repeatability. Locating and imaging the DOI is carried out automatically, with a throughput of 10 defects per hour.
Topography images of DOI have been obtained and compared with SEM images. ADR AFM is a non-destructive metrology tool developed for defect review and acquiring 3D topography information.
As integrated devices continue to shrink, the requirements for incoming bare silicon wafer defectivity have become increasingly stringent. The predominant way to inspect bare silicon wafers for surface defects is to measure the difference in LLS between a surface defect and a clean portion of the surface, where the intensity of the scattered signal is evaluated against the LLS of a standard latex sphere.
The actual surface defectivity can originate from topological defects, added particles, and crystal imperfections. The source of the defect must be known to reduce the number of defects.
LLS inspection can only provide defect counts and a relative size. In order to determine the origin and nature of the defects, defect review techniques such as AFM and SEM should be used.
In this article, focus is given to extremely small crystal imperfections which cannot be viewed easily by LLS without some means to make them larger. A decorative etching technique was used to highlight the crystal defects to be studied by AFM, SEM, and LLS methods.
Reproducible and accurate defect coordinate transfer between analysis tools is the only way to accomplish defect analysis. This article shows how the decorated defects can be successfully and reliably detected and characterized by AFM.
ADR AFM Procedure
Figure 1 shows the process in ADR AFM. The target defects are placed accurately and imaged non-destructively during this process.
Figure 1. The schematic shows the ADR AFM process for this study. After completing coordinate mapping, ADR AFM will automatically perform survey scan, zoom-in scan, processing, analysis, and classification for each defect.
In order to accomplish these objectives, two factors should be considered. First, proper linkage between the LLS inspection tool and the ADR AFM is required to reduce the positioning errors and accurately locate the defects.
Sample coordinate alignment is used to achieve the linkage for blank wafers. Generally, fiducials or alignment markers are not available on blank wafers to be used for sample alignment. Therefore, specialized vision is used in ADR AFM to carry out sample alignment properly.
Non-contact mode imaging is another key factor in AFM defect review. It is needed for non-destructive imaging of samples while simultaneously preserving the life of the AFM tip so that it can last throughout the process for multiple defects.
To ensure proper linkage between the stage coordinates of the LLS inspection tool and the ADR AFM, sample coordinate alignment is required. In the case of blank wafers, no alignment markers or fiducials exist on the sample to be used for sample alignment.
This challenge is resolved by performing a coarse alignment followed by a fine alignment. During the coarse alignment, three randomly selected peripheral features and the notch or an angular reference are selected to correct for rotational and translational errors.
This is followed by a fine alignment to remove positioning errors due to non-affinity between the stage coordinates of the LLS inspection tool and theADR AFM. Fine alignment is performed using several large defects with known inspection coordinates.
Since the defects are hardly visible in a standard AFM optical image, the defects in the ADR AFM optics are located with an enhanced vision process and the located detects are used as aligner markers. Following sample alignment, ADR AFM identifies more defects accurately.
During fine coordinate alignment, enhanced vision is used to locate the defects in the optical vision of the ADR AFM. The method is developed based on the well-established differential frame averaging of the optical frames obtained from the surface of the sample at two accurately isolated locations.
As a separated XY and Z scanners configuration is used by ADR AFM, the sample can be moved accurately. This architecture was originally developed to eliminate the crosstalk between the XY and Z scanners, which has been a regular artifact in tube scanner-based AFM systems.
In this arrangement, the sample is moved by the XY scanner while the tip is following the sample topography through the use of the Z scanner. In enhanced vision, the sample’s optical frames are obtained at two precisely isolated locations, and the final frame is subsequently created from the variation between the collected frames.
The resulting frame possesses an enhanced contrast of surface details, which can be hardly seen in the standard vision of the ADR AFM. Figure 2 shows a comparison between the frames obtained by standard vision against enhanced vision.
Figure 2. a) Standard vs. b) enhanced vision images of a small defect on the surface of a bare silicon wafer. The insets frames show a magnified view of the defect. The small defect is easily observable in enhanced vision. The larger image dimensions are 550 µm × 413 µm.
Non-contact Mode Imaging
The standard imaging mode in ADR AFM is non-contact mode. During the defect review process, it is essential that the sharpness of the tip is maintained from the first to the last defect located and imaged.
In addition to keeping the tip costs low, well-maintained tip sharpness guarantees consistent accuracy and image quality between the images of all defects during the process. As a result, the defect is located and imaged with no interruptions by the automated system with a high throughput.
Non-contact mode imaging can be performed by oscillating the AFM cantilever at its resonance frequency. This oscillating cantilever is brought close enough to the sample that the oscillation amplitude reduces to a pre-defined set point because of the van der Waals tip sample interaction.
The oscillation amplitude is maintained by the ADR AFM to prevent the tip from contacting the sample. As the tip scans the surface of the sample, the cantilever is moved up and down with the Z scanner to maintain the oscillation amplitude and to maintain the interaction between the tip and the sample in an attractive regime.
More information regarding non-contact mode imaging is described in reference. While the functionality of ADR AFM is based on non-contact mode imaging, it can perform in other contact or dynamic imaging modes if required.
Automatic Defect Search and Imaging
Due to its fully automated process, ADR AFM helps to acquire the significant improvements in the throughput of defect review. After entering the defect coordinates from the LLS inspection tool into the ADR AFM, coordinate alignment is carried out, the defect is located, and imaging is performed for the list of chosen defects.
Locating and imaging the defects is a fully automated process. The automation involves locating the defect, non-contact mode parameter optimization, tip-sample engagement, optimization of the scan size, a survey scan, processing, the final scan, and defect classification.
Defects can be divided into two groups: pits and bumps. Defects are normally situated within ±10 µm of their LLS coordinates.
Sample Preparation (Etching Process)
Bare CZ silicon wafers measuring 300 mm in diameter were treated with a gaseous acid in a reducing atmosphere at a temperature and for an adequate period of time to grow the crystal defects . Both the shape and size of the decorated defects rely on the nature of the original defect (see Figure 3).
Once decorated, the size of the defect can be detected as an LLS event. The LLS events are located and sized by the LLS inspection tool, which provides the coordinates to be used by the AFM and SEM.
Figure 3. Schematic of the process used to decorate crystal imperfections for defect inspection.
An LLS tool was used to inspect a wafer with surface decorated defects and 34 defects were chosen to be assessed through ADR AFM. After entering the defect coordinates into the ADR AFM, coordinate alignment is carried out and ADR AFM is used to locate and image the defects.
Before being studied by ADR AFM, the first 21 defects were imaged by SEM. However, SEM images only provide aerial 2D views of defects without adequate information on out of plane dimensions and the defects’ depth.
Despite the signal collected by the LLS tool, SEM could not locate the remaining 13 defects. Figure 4 shows the summarized results of the decorated defect study with ADR AFM as well as a comparison with the SEM results. In contrast, ADR AFM was able to locate all of the 34 defects including those that were not located by SEM.
Figure 4. The results of defect review with ADR AFM and a comparison with SEM are shown. ADR AFM was able to locate and image all the 34 defects.
The defects chosen to be assessed by ADR AFM belong to the eight types based on their LLS signal. The LLS tool’s tentative classification is based on the defect’s light scattering, which depends on depth, morphology, and presence of a central defect.
As the decorative etching process continues, crystal defects are revealed and etch at a different rate than the perfect crystal surface. When compared to the defects exposed late in the etching process, defects exposed at the inital stages of the etch are more developed and deeper.
Defects with an inverted pyramid shape posses a higher LLS signal and are usually deeper. These are categorized as "Facet". Curved shape defects that form during the late stages of etching are shallower and are categorized as "Shallow".
There are certain defects that are exposed at an intermediate point in the decorative etch and possess some degree of curved bottom with faceted walls. This type is classified as "Both". Some defects possess an extremely weak LLS signal as they have only started to be decorated, and are classified as "Too shallow".
Regardless of whether they have the center defect or not, all defects were categorized. Therefore, a total of eight types of defects were detected. Figure 5 shows the defect classification.
Figure 5. Defect classification based on the LLS, SEM and AFM data.
The LLS signal becomes weaker while moving from the left to right side of the table in Figure 5. This is due to the sharpness of the defect's edges and the depth of the defects.
The depth difference between various classes of defects is validated by AFM images. As Z heights were present in the AFM images, a banded color scale was used to show the defect’s surface topography more accurately in 2D view.
A comparison between the data obtained with SEM versus AFM for the same defect is shown in Figure 6. Primary SEM images gives an aerial 2D view of the defect, but the defect’s shallow depth reaches SEM limitations, showing poor contrast in the image.
SEM did not find shallower defects, as indicated in Figure 5. The center defect was identified by a secondary electron image. However, this is only possible if the defect was identified in the primary SEM image.
On the other hand, the AFM image provides an aerial view of the defect and also includes the depth/height values for each pixel. Using a contoured color scale or using a 3D representation of the AFM image, additional data can be acquired about the true topography of the defect.
Contoured color scales can help to understand the defect topography in aerial view (Figure 5). As mentioned above, AFM has the highest vertical resolution among all imaging methods and provides better contrast in images taken from an aerial view.
Figure 6. Comparison between the data collected with SEM versus AFM. The SEM image provides an aerial 2D view of the defect. A secondary electron image indicates the presence of center defect. The AFM image, in addition to providing an aerial 2D view, includes the 3D data. Therefore a line profile, 3D demonstration, and contoured color scale can be utilized to obtain more information.
ADR AFM detected all of the 34 defects, including13 that had not been detected by SEM. Figure 7 shows the AFM images, a defect that was not detected by SEM.
The depth of the defect is below 4 nm and includes a center defect feature. Once again, this example highlights the limitation of SEM resolution in an out of plane direction.
Figure 7. AFM data for defect #24, which was not found by SEM.
As mentioned above, the ADR AFM is a non-destructive imaging method that uses non-contact mode imaging for survey scan as well as final imaging scan. However, the SEM beam can still change the surface of the sample.
Sample contamination, due to the electron beam "burning" the surface during SEM imaging, is shown in Figure 8. These SEM burn-mark sizes correspond to the SEM magnification. As shown in Figure 8, a number of SEM magnifications were employed to study this defect.
Figure 8. AFM of a Facet defect with several SEM burn-marks.
This article has shown how ADR AFM offers quality 3D data for defect review on bare silicon wafers. This type of analysis is made simple yet powerful with automation and carries significant advantages over traditional characterization methods such as SEM.
References and Further Reading
- G. T. Smith, Industrial Metrology: Surfaces and Roundness.: Springer, 2002.
- Ardavan Zandiatashbar et al., "High-throughput automatic defect review for 300mm blank wafers with atomic force microscope," in Proc. SPIE 9424, Metrology, Inspection, and Process Control for Microlithography XXIX, 2015, p. 94241X.
- J. Libert and L. Fei, Method to Delineate Crystal Related Defects.: PCT Publication, WO2013055368(A1).
- Ardavan Zandiatashbar, "Sub-angstrom roughness repeatability with tip-to-tip correlation," NanoScientific, no. Winter, pp. 14-16, 2014.
- Ajay Kumar and Banqiue Wu, "Extreme Ultraviolet Lithography: A Review," Journal of Vacuum Science & Technology B, pp. 1743 - 1761, 2007.
- James A. Folta, J. Courtney Davidson , Cindy C. Larson, Christopher C. Walton, and Patrick A. Kearney, "Advances in low-defect multilayers for EUVL mask blanks," in Proceedings of SPIE 4688, Emerging Lithographic Technologies VI, 173, 2002.
This information has been sourced, reviewed and adapted from materials provided by Park Systems Inc.
For more information on this source, please visit Park Systems Inc.