Manufacturers of silicon wafers have been striving for years to produce wafers with flat, smooth, defect-free surfaces. Such wafers are produced by carefully polishing the surface to a level of smoothness that is significantly better than most optical surfaces. Unfortunately, not all defects are on the surface. Many defects are introduced by the surface-finishing process and reside just below the surface in a damaged layer. While the surface may be very smooth, the subsurface damaged layer distorts the crystal structure and can significantly affect the operation of active electronic circuits placed on the wafer. Much effort has been expended to produce wafers with smooth surfaces and no subsurface damage. Detecting and mapping that damage, although very difficult, is an important part of achieving high-quality, smooth wafer surfaces.

One way to improve the wafer surface is to deposit a thin, epitaxial layer of silicon. A single-crystal deposited epitaxial layer has a uniform structure that has proven to be ideal for manufacturing circuits when the layer quality is high. Nevertheless, questions remain.

Based on experiments using both surface scanning and the PBS subsurface defect-mapping system designed by VTI (Dayton, OH), this article addresses the question: How does damage in the wafer substrate affect the epitaxial layer? The experiments were performed at Okmetic, a manufacturer of silicon wafers located in Vantaa, Finland.

The PBS Measurement Technique

In a typical wafer-manufacturing process, the ingot is sliced using either a circular or wire saw. Subsequently, the wafer is lapped, etched, and polished. Lapping planarizes the wafer and removes most of the deep damage caused by slicing, while etching removes the damage caused by lapping. Lastly, polishing produces the final, top-quality surface suitable for device manufacturing.

Typically these processes remove a total of 50–80 µm of silicon per side. Lapping removes between 35 and 50 µm, etching between 15 and 25 µm, and polishing between 10 and 20 µm. In addition to determining the flatness of the wafer, these silicon removal processes largely determine the polished wafer's crystalline and surface quality. From the economic point of view, it is crucial to remove as little material as possible, while ensuring that all defects caused by the wafer-forming processes have been eliminated.

An epitaxial layer of silicon is commonly grown on the polished wafer before IC or MEMS device fabrication begins. During epitaxial deposition, surface defects and contamination on the substrate wafer can initiate the nucleation of crystalline defects. Depending on their morphology, these crystalline defects are usually called stacking faults, mounds, hillocks, pyramids, or spikes. Stacking faults are a major cause of yield loss in epitaxy. Therefore, it is important to understand the role of substrate damage in the formation of epitaxial defects.

The PBS measurement technique can be used to detect and map subsurface damage nondestructively, providing a useful tool for evaluating wafer quality. Because PBS measurements are nondestructive, measurements can be made both before and after epitaxy. The technique uses scattered laser light to detect subsurface defects in semiconductor wafers caused by slicing, lapping, polishing, and other surface-finishing processes.¹ The technique also can be used to evaluate epitaxial layers and other types of coatings. The measurement is displayed in a color map showing scatter-intensity distribution over the wafer. Map variations show the distribution of subsurface defects while scatter intensity relates directly to defect density and depth.

The measurement is made with a helium neon laser, whose probe beam is 0.3 mm in diameter, circularly polarized, with a wavelength of 632.8 nm. The measurement geometry is shown in Figure 1. A large, 55° angle of incidence is used to reduce the reflectance of the P-polarized component of the probe beam. Scatter is detected in the backscattered direction where the P- and S-polarized components of the scattered light are separated. This backscatter angle decreases scatter from the surface and increases the detection of scatter from the subsurface. The detector solid angle is limited to about 0.004 sr to confine the scattered light to the area directly in the plane of incidence.

Figure 1: Simulation of PBS geometry measuring subsurface damage on a silicon wafer.

The P-polarized component of the scattered light is measured as parts per million per steradian (ppm/sr) of the incident intensity and is used directly as the PBS measurement. Because the light is strongly absorbed by silicon, the measurement is sensitive only to shallow damage within about 2 µm of the surface. Both the P- and S-polarized components are used to separate surface and subsurface effects. The reflectance of P-polarized light is 15.2%, which means that about 85% of the light enters the wafer under these geometry conditions.

A perfect crystal structure scatters light only at an extremely low level, but any crystal defects act as particles suspended in a medium and scatter primarily in the forward and backward directions. A portion of the internally scattered light travels back to the surface and is refracted toward the incident probe beam and the detectors. Surface microroughness and the reflected beam result primarily in forward-direction scatter, which is isolated from the detectors.

Subsurface damage from typical surface-finishing processes is usually oriented in a specific way, with strings of defects forming parallel lines in the subsurface. This defect structure acts like an optical grating, strongly scattering the light in one direction perpendicular to the structure. Since defects can be aligned in any direction, it is necessary to rotate the plane of incidence to find the orientation of the defect structures on any particular wafer.

Because scatter is measured over a small solid angle in the plane of incidence, the azimuthal orientation of the plane of incidence at a given point on the surface determines which defects are seen. In other words, scatter in the plane of incidence comes from defects aligned perpendicular to the plane of incidence. The angle of the plane of incidence to the test surface is called the R-angle. The changing scatter signature as the R-angle changes shows the orientational characteristics of subsurface defects. That change is a particularly important aspect of the measurement process because it enhances the scatter from oriented defects. Figure 2 shows a 3-D drawing of this rotational arrangement and the location of the R-angle.

Figure 2: Schematic drawing of the plane of incidence and the azimuth angle R in relation to the X-Y axis of a test surface.

The PBS measurement is accomplished in two steps. First, several scatter-intensity versus R-angle measurements are made for a series of adjacent points in the center of the wafer. These measurements are displayed together on a single polar plot called a V-map, an example of which is presented in Figure 3. The V-map is composed of a series of concentric circular arcs, each of which represents a point on the wafer surface. The R-angle is measured counterclockwise from the 0° position, and the colors in the map show the scatter intensity. The measurement arcs are not complete circles because the high-speed equipment design blocks the measurement over a portion of the rotation. This is not significant since virtually all defect structures scatter with a 180° symmetry. A red band across the V-map indicates a defect structure.

Figure 3: V-map of an as-polished silicon test wafer.

Second, the plane of incidence is positioned in the high scatter direction indicated by the V-map, after which an X-Y scan covering the surface of the wafer is performed. Figures 4 through 9 show such maps. All the maps have a color scale on the right-hand side showing the maximum and minimum scatter levels.

While the PBS measurement technique primarily detects subsurface defects, some types of surface features, such as pits, particles, and haze, affect wafer quality and are detected by this method. To differentiate between subsurface and surface defects, the EPBS measurement was developed. This measurement combines the P- and S-polarized components of the scattered light using a proprietary algorithm to produce separate surface and subsurface EPBS wafer maps. These maps represent normalized scatter intensity from the surface or subsurface. EPBS results, while similar to PBS results, are not quantitatively comparable. Surface and subsurface EPBS maps can only be compared with each other when plotted on the same scale.

Experimental Procedure

Experiments were performed to determine the relationship between surface defects and subsurface damage. After being sliced with an inner-diameter circular saw, 150-mm wafers were lapped, etched, and polished to different thicknesses. Total material removed was between 25 and 80 µm.

Following polishing and chemical cleaning, the PBS technique was used to measure the wafers. Several wafers that apparently had been damaged during polishing were also measured. Subsequently, a 2-µm epitaxial layer was deposited on both damaged and undamaged wafers in a single-wafer reactor using a standard high-temperature atmospheric process. The wafers were checked for epitaxial defects with a laser surface scanner and then remeasured with PBS. Finally, a thin, 100-nm-thick disordered zone, a common feature of epitaxial layers, was removed using touch polishing, and the wafers were measured again.

The PBS measurements indicated that the substrate wafers had sustained varying degrees of damage depending on how much of the surface had been removed during the different process steps. This primarily directional damage is shown in the V-maps in Figures 4a and 4b, which compare two wafers that underwent different amounts of surface removal. Less surface removal resulted in a greater amount of directional damage. As expected, the area maps in Figures 5a and 5b, created by measuring in the maximum scatter direction, show that wafer damage remained when too little surface material was removed. The wafer mapped in Figure 5a, which underwent 69.5 µm of surface removal, had much lower defect counts than the wafer mapped in Figure 5b, which underwent only 25 µm of surface removal.

Figure 4: V-maps showing the effects of removing (a) a moderately large amount of surface material from a wafer and (b) a small amount. Removing 69.5 µm of material resulted in a low level of directional damage, while removing 25 µm resulted in significant directional damage, indicated by the stripes.

Figure 5: A defect map of the wafer that underwent 69.5 µm of surface removal (a) and a defect map of the wafer that underwent 25 µm of surface removal (b). The measurements confirm that removing less surface material results in higher defect levels.

The effect of surface removal is not always straightforward, since polishing affects the subsurface layer and can, to a certain extent, hide deep damage. In fact, since polishing seems to reorder the wafer near the surface, subsurface damage caused by slicing or lapping is not usually immediately evident near the surface. Some fairly uniform but directional subsurface damage remains even after polishing, but this damage can be detected with the PBS measurement system. In general, the direction of the damage is random and not related to the direction of the crystal lattice.

EPBS analysis can be used to distinguish between surface and subsurface damage. While the surface EPBS map in Figure 6a is featureless and the signal level very low, the subsurface EPBS map of the same wafer in Figure 6b has a much higher scatter level with notable features. The subsurface EPBS map is qualitatively identical to the PBS map in Figure 5b. These maps reveal that the oriented damage is located below the polished surface. The PBS technique cannot determine the exact depth distribution of the damage, but the small absorption length of the 632.8-nm light in silicon limits the probed depth to < 2 µm. Because of this shallow probe depth, the PBS technique is most sensitive to damage sustained in the final removal process, namely polishing. Surface defects such as pits on polished wafers generally appear as isolated red spots on PBS and surface EPBS maps.

Figure 6: A comparison of surface (a) and subsurface (b) EPBS maps confirms that removing 25 µm of material resulted in very slight surface damage and extensive subsurface damage.

Because the mechanical damage caused by slicing, lapping, grinding, and polishing is almost always directional and usually curved, the direction of the maximum scattering intensity changes across the wafer. Consequently, a PBS map may show only part of the damage. When a PBS area map is made, the R-angle is fixed so that only those defects that are aligned perpendicular to the probe beam, or defects that scatter in all directions, are detected. An examination of the V-maps in the foregoing figures indicates that the aligned defects are detectable only over a relatively narrow range of R-angles.

Curved damage is typically easy to identify on PBS maps because it causes a band of higher-intensity scatter across the wafer, as presented in Figure 7. A more-complete map of that damage can be obtained by repeating the mapping process using slightly different R-angles, which results in bands of damage in slightly different positions. PBS software allows maps to be combined to produce one map portraying the maximum scattering intensity for several R-angles. This process also is used when a V-map shows several distinct damage directions and separate maps are made for each direction.

Figure 7: A PBS map of a wafer with curved, polishing-induced damage. Curved damage causes a characteristic band of high scattering intensity across the map.

Close examination of the map in Figure 7 shows that the band is aligned exactly in the R-angle direction, parallel to the plane of incidence. The band itself is composed of many small line segments that are perpendicular to the band. These line segments are portions of the subsurface damage structure in the wafer. In this case, the curved structure covers the entire wafer.

PBS software also can estimate the radius of curvature of the damage. This is done by creating at least two additional V-maps at different locations on the wafer. The radius is calculated from the differences between the R-angles on the different V-maps. This feature can be useful for pinpointing the origin of the damage. For example, the polishing-related damage on the wafer shown in Figure 7 had a radius of curvature of about 750 mm, which is in fair agreement with the polishing machine's radius of rotation.

The Effect of Substrate Damage on the Quality of Epitaxial Layers

Further experiments indicated that after a 2-µm epitaxial layer was applied, the scatter level increased significantly, as shown in the V-map in Figure 8a and the area map in Figure 8b. This increase was the result of a disordered zone at the surface of the epitaxial layer.

Figure 8: A V-map (a) and an area map (b) created after the deposition of a 2-µm-thick epitaxial layer on the wafer that underwent 25 µm of surface removal. Nonuniform shallow damage is visible.

The depth of the disordered zone is believed to be determined by the ramping down of the deposition process. In epitaxial layers grown on damaged substrates, the scattering level caused by the disordered zone is very high. How the disordered zone forms, particularly the role of subsurface damage, is not known. However, the unusual nonuniformities that can be seen in the area map in Figure 8b provide a gauge for determining how well the disordered zone has been removed in the next step.

After the epitaxial layer had been applied and the wafers mapped, the wafers were touch polished to remove about 100 nm of material from the surface. The removal of this small amount of material was sufficient to remove the disordered zone, exposing the bulk of the epitaxial layer for examination.

The V-map in Figure 9a and the area map in Figure 9b show damage results after touch polishing was completed. The area map is featureless and the PBS scattering level is very low. In fact, the scattering level compares to that of a high-quality prime polished wafer. It seems that except for a very shallow surface disordered zone, epitaxial layers are not sensitive to the curved damage structures and isolated defects in the subsurface region. Moreover, while the exact thickness of the disordered zone is uncertain, it cannot exceed a few tens of nanometers. In addition, surface inspection results show that the surface remains free of epitaxial defects regardless of how much subsurface damage remains in the substrate after polishing.

Figure 9: A V-map (a) and an area map (b) created after 100-nm touch polishing. The V-map shows no preferred scattering direction, and the area map shows that surface damage resulting from the disordered zone has been totally removed.

Conclusion

The PBS measurement technique was used to characterize a large number of wafers with varying degrees of damage. This technique was found to be sensitive and convenient for mapping subsurface damage within about 2 µm of the surface of polished silicon wafers. The technique is most suitable for measuring polishing-related damage, but it also can detect directional damage caused by slicing and lapping. The contributions of surface and subsurface defects can be distinguished by using the EPBS algorithm. For directional defects, the radius of curvature can also be estimated with PBS. Because of the software's measurement speed, it can be used for routine production control.

It was found that 2-µm-thick epitaxial layers are remarkably insensitive to deep damage in polished substrates. The only signature of substrate damage that could be seen on epitaxial layers was a very shallow disordered zone within 100 nm of the surface. Once this zone had been removed, the remaining material was almost damage free. The curved damage structures and isolated defects seen on the substrates did not affect the epitaxial layer, indicating that wafer manufacturers have some latitude when selecting substrates for epitaxial wafers.

Acknowledgments

The authors would like to thank Tom Magee, formerly with Applied Materials, for his helpful discussions on epitaxial-layer growth and characteristics.