ANALYSIS & METROLOGY—POST-CMP

Developing effective inspection systems and strategies for monitoring CMP processes

Cheri Dennison, KLA-Tencor

While it is a radical departure from the typical clean technologies that make up most of the IC manufacturing process, chemical mechanical planarization (CMP) using multipart abrasive slurries has become a key technique in enabling the development and production of sub-quarter-micron devices. First used primarily for polishing intermetal dielectric (IMD) layers and then for treating tungsten plugs, CMP is being used in additional process segments, in particular in shallow trench isolation (STI). Because CMP inevitably creates substantial amounts of surface contaminants and can introduce many defects that are major threats to final yields, the processes must be closely monitored using sophisticated wafer inspection tools. After reviewing the types of defects that may occur, this article focuses on advances in inspection technologies. Examples of effective post-CMP inspection strategies are also provided.

Critical CMP Defect Types

Each of the three most common CMP process applications—IMD, STI, and tungsten—can introduce yield-limiting defects, but the types of defects produced varies from application to application.

Figure 1: Schematic illustrations of premetal oxide (top), which is often borophosphosilicate glass (BPSG), and IMD oxide just after metal 1 (bottom). The differences in the underlying structures seen here can lead to unique defect inspection challenges.

Oxide CMP. When CMP is used for oxide polishing, a distinction is often made between premetal oxides and the IMD oxides used at the back end of the line to separate the metal layers. As seen in Figure 1, oxide is being polished in both applications, but the underlying structures are different—a fact that must be taken into consideration when inspecting for defects. STI CMP is another subset of the oxide CMP application. Used in advanced-device manufacturing in place of the traditional local oxidation of silicon process, STI makes it possible to achieve the closer packing density necessary for 0.25-µm logic processes. In this application, oxide is polished down to a nitride stop; the nitride and underlying pad oxide are then etched away to leave trenches filled with oxide, as shown in Figure 2.

Figure 2: Schematic illustration of STI CMP. After the oxide is polished down to a nitride stop, the nitride is etched away, leaving trenches filled with oxide.

Whether it is used for IMD or STI, oxide CMP can introduce several types of yield-limiting defects, including residual slurry, surface voids, and surface particles. Residual slurry is caused by an inadequate post-CMP cleaning step, while surface voids may be caused by an embedded particle or a weak point in the oxide being ripped out or dislodged during processing. Surface particles can also result from inadequate post-CMP cleaning or from airborne contaminants. Other critical oxide CMP defects that require close monitoring are microscratches and embedded particles. Microscratches are created when a small particle or other debris is caught between the polishing pad and the oxide surface during polishing, while an embedded particle is one trapped in the oxide during metal deposition prior to CMP. Each of these defect types is illustrated in Figure 3.

Figure 3: Five types of oxide CMP defects.

Tungsten CMP. Devices with random patterns, such as logic, require thick dielectrics and multiple layers of interconnects, which, in turn, create deep vias. Because these vias are challenging to fill using standard metal-sputtering techniques, tungsten plugs are widely used as an alternative to sputtered metal. The tungsten is deposited on top of an adhesion layer such as titanium nitride (TiN) (shown in dark orange in Figure 4) and is then polished using CMP until only plugs remain.

Figure 4: Schematic illustration of tungsten CMP. The tungsten is deposited on top of an adhesion layer and then polished until only plugs remain.

Among the most significant defect types associated with tungsten CMP are tungsten puddles, surface voids, tungsten-filled microscratches, and cored plugs. A tungsten puddle is created when residual tungsten is trapped in a surface void (caused by a scratch or ripout) and is then polished planar to the oxide surface. A tungsten-filled microscratch occurs when a microscratch created at a previous oxide polish step is filled with tungsten. Plug coring is a void in the plug and is usually related to the slurry chemistry. Other metal CMP defects include empty microscratches, residual tungsten stringers, residual slurry, surface particles, and recessed plugs. Examples of all of these defect types are shown in Figure 5.

Figure 5: Nine types of tungsten CMP defects.

Meeting the Challenges of Post-CMP Defect Inspection

Detecting these different types of defects can pose serious challenges for wafer inspection systems, primarily because of the film thickness variations caused by the CMP process. During CMP process optimization, film thickness—both across a wafer and between wafers—may vary by as much as 25%, and even a well-optimized process can show film thickness variation on the order of 3%. For bright-field image-processing tools, nonuniform film thickness interferes with the reflected light, resulting in a change of color. In gray-scale representations, which are used by these tools for die-to-die processing, this color variation translates into a variation in gray-scale levels, creating "noise" that can reduce the sensitivity of the inspection system. For laser light—scattering tools, film thickness variations present a different set of challenges. The way a defect and its surrounding surface scatters light depends partly on the surface's local reflectance. Because surface reflectance is affected by film thickness, different areas may scatter more or less light, making defect detection difficult. Other noise sources that can diminish the sensitivity of inspection tools are underlying or surrounding grainy films. Grainy features can create additional noise for both bright-field and light-scattering systems, potentially resulting in nuisance defect counts.

Laser Light—Scattering Systems. Despite these noise problems, advances in wafer inspection technologies aid in effective monitoring of IMD, STI, and tungsten CMP processes. For example, the unique design of a double-dark-field (DDF) laser light—scattering inspection system (Figure 6) dramatically reduces the effect of film nonuniformities. Scattering theory states that scattering background varies with the cosine of the incident angle and cosine-squared of the scattering angle.¹ The figure shows the instrument's incident and scattering angles of light, where 0° refers to normal incidence and 90° is in the plane of the wafer. From this it can be seen that light scattered by the surrounding surface (background noise) is minimized by the use of a grazing-incidence laser beam combined with collection optics located near the wafer horizon.

Figure 6: A double-dark-field inspection system design that suppresses background noise for high defect capture rates on post-CMP inspections.

DDF technology also allows the use of circular polarization of the incident laser beam, which further enhances defect capture on nonuniform films by reducing the variation in the defect's scattering signal. Figure 7 shows how scattering cross sections vary with film thickness for three incident-beam polarizations, C (circular), S, and P. The different curves on each graph represent different sizes of polystyrene latex (PSL) spheres, from just over 1 µm on the top to 0.157 µm on the bottom. As the figure shows, the amount of light scattered by a PSL sphere of a given size depends on the thickness of the film on which the sphere rests. This means that the same sphere will scatter very differently depending on whether it is sitting on a thin or thick area on the film. The smaller the sphere is, the larger the variation will be. However, as the graph on the right of the figure shows, over a typical variation in film thickness the variation in scattering cross section can be minimized by using circular polarization.

Figure 7: Comparison of scattering cross sections for various-sized PSL spheres using three incident-beam polarizations.

Bright-Field Image-Processing Systems. Inspection systems based on bright-field digital image-processing technology, which is widely regarded as the most sensitive and comprehensive defect inspection technique, have always provided a high rate of defect capture. Recently, the capabilities of one such system were expanded with the addition of an ultrabroadband (UBB) illumination source in combination with significantly improved bright-field optics and a unique segmented autothreshold (SAT) technology. These improvements increased the instrument's defect detection sensitivity, especially on wafers with significant color variation.

Figure 8: A digitized image of a post-CMP device (top) and a segmented image of the same device created by the SAT algorithm (bottom). In this example, two segments were used, shown in blue and yellow.

Figure 9: Diagram of SAT operation. Without SAT, the high fixed threshold level would be used to minimize nuisance defect counts; with SAT, lower thresholds are automatically applied to each segment that has lower noise, maximizing the capture of real defects while reducing nuisance defect counts.

The SAT image-processing technique was developed especially to increase sensitivity on wafers with the grainy metal and color variations seen following CMP processes. SAT algorithms segment the wafer image based on the mean and range value of each pixel; separate thresholds are then dynamically set for each segment, rather than having a single threshold for the entire wafer. The top part of Figure 8 is a digitized black-and-white image of a device following oxide CMP. The wide, light areas are metal, and the thin, dark lines are the areas between the metal. The high level of contrast in this image poses a problem when a single threshold value is used. The SAT algorithm enhances defect detection capability in such situations by segmenting the different images used in die-to-die processing, as seen on the bottom of the figure. In this example, two segments were used, shown in blue and yellow. Figure 9 further illustrates the value of the SAT algorithm. Without SAT, the high fixed threshold level shown in this figure would be required to minimize nuisance defect counts caused by the highest-noise areas, but its use could result in a failure to detect legitimate defects. In contrast, SAT automatically applies lower thresholds to the segments that have lower noise, maximizing defect capture while reducing nuisance defect counts.

Figure 10: Comparison of the intensity and wavelength range of UBB and narrowband illumination sources.

Figure 10 reveals that a UBB illumination source produces a wider and more uniform range of wavelengths than a narrowband source. Color variation following oxide CMP is the result of interference from varying thicknesses of the transparent oxide layer, and UBB technology is able to optically average out this interference, thereby increasing the instrument's sensitivity to CMP-related defects. The images in Figure 11 illustrate this capability. The upper row contains time delayed integration (TDI) digitized images of three die used for die-to-die image processing with a narrowband illumination source; the bottom row shows the same die with a UBB source. Each image represents the same patterned area on the respective die. Comparing the two rows reveals there are more significant contrast differences between each die when the narrowband illumination is used, requiring that higher thresholds be set to avoid nuisance defect detection. Because the UBB technology optically averages out intensity variations, contrast differences between die are lower. In addition, because of the reduction in color variation, a larger pixel size can sometimes be used for inspection, thus increasing system throughput.

Figure 11: TDI digitized images of different die used for die-to-die image processing, using narrowband (top) or UBB (bottom) illumination sources. The UBB source decreases color noise through wavelength averaging, resulting in reduced contrast variation between the die.

Effective Post-CMP Inspection Strategies

Both DDF laser light—scattering and UBB bright-field image-processing techniques offer advantages that can be exploited for CMP inspection applications. Either tool can be used as a single CMP inspection solution, or, in fabs with the required resources, they can be combined in a complemen-tary strategy to maximize the inspection strengths of each technology.

STI Inspection Strategies. Two inspection points are commonly used for monitoring STI CMP processing: directly after polish, and after the nitride strip and pad oxide removal. Each point has its advantages and disadvantages. Inspecting directly after polish provides immediate feedback on the CMP process. At this point, however, high process noise can often limit the sensitivity of the inspection, resulting in some defects—particularly microscratches—going undetected. Inspecting after a later process step usually ensures higher defect capture because there is less process noise, but CMP process feedback is less immediate.

Figure 12: Results of an inspection immediately after STI CMP, in which a microscratch appears as a faint dot (left image), and of an inspection of the same die area after nitride strip and pad oxide removal, which shows the microscratch quite clearly (right image).

Figure 12, which shows results of a case study from Advanced Micro Devices (AMD) using a KLA-2100 series bright-field inspection system, illustrates how images from these two inspection points differ in their depiction of microscratches.² The left image is the result of an inspection immediately after STI CMP, and the microscratch appears as a faint dot. The right image shows the same die area inspected after nitride and pad oxide removal; the microscratch is now much more visible, making it easier to detect and classify. Detection is improved at this step because the microscratch undergoes a significant change during the nitride strip and pad oxide removal, as illustrated in Figure 13. Just after CMP, the microscratch may be very shallow (for example, <100 Å), making detection extremely difficult. During the nitride strip, however, the microscratch grows slightly, and then during pad oxide removal, the HF etch causes the scratch to grow even more and become "decorated." This latter phenomenon is why inspection just after CMP is sometimes called predecoration, and inspection after the nitride strip and clean is called postdecoration. Figure 14 shows additional results from the AMD case study. Here, predecoration detection of microscratches is compared with the postdecoration detection of microscratches. As can be seen, almost 100 times more microscratches were revealed in the postdecoration inspection. The results were consistent for both densely and sparsely patterned regions of the die, as is also shown in the figure.

Figure 13: Schematic illustration showing how a very shallow microscratch grows and becomes decorated during nitride strip and pad oxide removal, making its detection easier.

Figure 14: Predecoration detection of microscratches compared with postdecoration detection of microscratches.

IMD Inspection Strategies. The two inspection points most frequently used for monitoring the IMD CMP process are directly after polish, and after Ti or TiN deposition. Again, inspecting directly after polish has the benefit of providing immediate process feedback, but detection sensitivity can be impaired by the presence of significant process noise caused by underlying grainy layers. DDF technology, however, is well suited to this application, enabling effective oxide IMD monitoring directly after polish.

Figure 15: Critical defect types identified in a case study from a logic and memory manufacturer that inspected 100% of its lots, two wafers per lot, immediately after oxide IMD CMP.

Figure 15 depicts typical results from a major logic and memory manufacturer that used the Surfscan AIT dark-field system immediately after oxide IMD CMP, inspecting 100% of its lots at a sampling rate of two wafers per lot. The defect types that occurred most frequently and that were of major concern were surface voids (or ripouts), microscratches, and particles. Figure 16 shows results of another case study from a different major logic manufacturer that also used double-dark-field technology to inspect directly after oxide IMD CMP. In this study, however, only 50% of the lots at four wafers per lot were sampled. The major defect types identified were residual slurry, microscratches, and particles.

Figure 16: Critical defect types identified in a case study from another logic manufacturer that also inspected after oxide IMD CMP.

The other inspection point commonly used for monitoring the oxide CMP process is following Ti or TiN deposition because the opaque Ti/TiN layer masks the underlying grainy layers, eliminating color variation and other noise sources that can inhibit inspection sensitivity. In addition, the Ti/TiN layer decorates the defects, which improves their detectability. The inspection results illustrated in Figure 17, which were obtained using a KLA-21XX bright-field system, show that defects from both the current and previous process layers can be revealed at this point—in this case, defects from the CMP process, the contact patterning processes, and the Ti/TiN deposition.

Figure 17: Critical defect types identified following TiN deposition, which include microchatter and microscratches from the CMP process, missing and closed contacts from the patterning process, and particles from the TiN deposition.

Tungsten CMP Inspection Strategies. When inspections are performed following tungsten polishing, the wafer surface is largely transparent oxide broken up with small tungsten-filled contacts or vias. As with other CMP inspection points, underlying process noise from previous-layer grainy films can negatively affect the sensitivity of the results, and this issue must be addressed by using a system with advanced inspection capabilities, such as those described above.

Figure 18: Critical defect types identified in a case study. When a problem with cored plugs (top) was solved by switching to a new slurry, a higher level of microscratches occurred (bottom). An uncored plug is shown at top right in the figure.

The capabilities of the KLA-21XX for tungsten CMP inspection have been illustrated in a case study presented by Cypress Semiconductor.³ After experiencing yield loss resulting from cored plugs, Cypress engineers traced the problem to the chemistry of the CMP slurry. (The images at the top of Figure 18 show examples of a cored plug and a good plug, respectively.) When Cypress switched to a new slurry the coring problem was solved, but it was soon discovered that the reformulated slurry introduced a different problem: a higher level of microscratches (as seen in the bottom image of Figure 18). In order to address this new problem, Cypress used its bright-field inspection system in array mode to inspect split lots. Results of these inspections helped the fab determine that new filters and pads were needed to reduce the occurrence of the microscratches. The progress of its process optimization is presented in Figure 19.

Figure 19: Process optimization results from the case study illustrated in Figure 18.

Conclusion

Because it enables both increased levels of interconnect and tighter packing densities, CMP has become critical for the development and production of sub-quarter-micron devices. The implementation of CMP processes, however, has introduced many yield-limiting defects and noise sources, which in turn have created significant challenges for optical in-line wafer inspection systems. Recent advances in both dark-field laser light—scattering and bright-field image-processing technologies address these challenges, allowing effective monitoring of IMD, STI, and tungsten CMP processes.

Acknowledgments

The author would like to thank Bobby Bell, Raleigh Estrada, Becky Pinto, and Kern Beare for their help in preparing this article.

References

1. Elson JM, Rahn JP, and Bennett JM, "Relationship of the Total Integrated Scattering from Multilayer Coated Optics to Angle of Incidence, Polarization, Correlation Length and Roughness Cross-Correlation Properties," Applied Optics, 22(20):3207—3219, 1983.

2. Zika S, and Bains G, "Microscratch Detection Strategies for After Oxide CMP at Shallow Trench Isolation," presented at KLA Yield Management Seminar, Geneva, Switzerland, April 1997.

3. Shih YC, and Wang JF, "Polishing Pads and Process Effects on Tungsten CMP," in Proceedings of ISMIC 1997, pp 237—240, 1997.

Cheri Dennison is a technical marketing specialist for the wafer inspection group at KLA-Tencor (San Jose). Previously, she was an applications engineer and a regional applications manager for Tencor Instruments. She also managed the Intel account for the wafer inspection applications organization. Before joining Tencor, Dennison was a defect reduction engineer at Motorola's Phoenix facility. She has a BS (1989) in materials science and engineering from Stanford University. (Dennison can be reached at 408/875-7601.)

MicroHome | Search | Current Issue | MicroArchives
Buyers Guide | Media Kit