Bright field-bright future:Material defect detection with a laser scanning system
09/01/1997
Bright field - bright future: Material defect detection with a laser scanning system
C. Thomas Larson, Mehdi Vaez-Iravani, KLA-Tencor, Wafer Inspection Division, Milpitas, California
In wafer manufacturing, inspection of the finished material assures that the wafers produced meet customer expectations. In the past, some defect classes could not be detected by automated inspection systems - known generically as scanning surface inspection systems (SSISs) - and visual inspection was all that was available. A new technique integrates, for the first time, an automated Nomarski Differential Interference Contrast technique with an advanced SSIS. It can detect these defects with more sensitivity, at a faster speed, and with higher accuracy, allowing these defect data to be combined with other inspection data to facilitate automatic classification of the entire range of defects on the wafer surface.
Integration of the Nomarski Differential Interference Contrast technique with an advanced Laser Scanning Inspection System [1] provides a significant advance in the automation of the entire inspection process. This technique enhances the productivity of the inspection process on many levels. First, the higher sensitivity allows the identification of defective process steps that would otherwise proceed. Second, automating a traditionally visual process improves consistency. Third, the speed of the inspection step is substantially increased by simultaneously capturing the whole range of defect types. Finally, improved accuracy due to automation moves the inspection process from a relatively qualitative capability to a substantially quantitative measurement.
This paper describes the scanning Nomarski technology as integrated onto a Surfscan SP1 and provides details of the system level applications. We present data obtained from well-characterized artifacts and data demonstrating the application domain of the approach. Finally, we close with a discussion of the future direction of the method.
Background information
Silicon wafer manufacturing. The manufacture of silicon wafers for the integrated circuit industry is experiencing tremendous growth as manufacturers try to produce enough high-quality starting material to meet the voracious demand of the IC industry. The IC community`s need to produce devices with more sophistication at increasingly high yields demands that wafer manufacturers improve their process technologies and raise their manufacturing productivity.
Inspection of the finished material assures the manufacturer that the wafer produced meets the expectations of the customer. Currently, portions of the inspection process are performed on automated inspection systems and the remainder is performed visually by operators looking for specific categories of defects. Depending on the sophistication of the inspection, it can identify specific defect mechanisms and the manufacturing process can be corrected. Higher inspection accuracy provides faster feedback and thus saves yield, time, and resources. As IC requirements tighten, the sensitivity to smaller and smaller defects becomes increasingly critical.
Automated inspection. Decreasing dimensions of critical defects, combined with the need to qualify silicon material quickly, have led to the wide use of automated inspection in the wafer manufacturing environment. Speed, sensitivity, and accuracy are the principal benefits. SSISs were first introduced to detect, size, and count particles adhering to the wafer surface. Particles have a measurable yield impact on the devices produced with the wafers. In the past, circuit linewidth requirements set the particle sensitivity at around a few microns in diameter. With circuit dimensions in the near future approaching =0.25 ?m, the particle sensitivity needed is below 0.10 ?m.
Laser inspection systems also deliver macroscopic information about a surface quantity called haze. This topic has been covered elsewhere [2], but haze is in general related to the amount of light scattered by a surface illuminated by a given amount of light. As the surface characteristics change (in fact, even as some sub-surface morphology changes), the light scattered will change in relatively predictable ways. Monitoring this change in surface scatter can reveal longer range process defect mechanisms, such as those related to polishing.
Visual inspection. Visual inspection is generally confined to locating macroscopic defects, such as scratches and some relatively large yet low-contrast defects in the silicon wafer manufacturing process. Although the visual inspection step may never be removed entirely, there is a compelling need for an automated technique that can detect these defects with more sensitivity, faster, and with high accuracy, and then combine the data with other inspection data to help in the automatic classification of the entire range of defects on the wafer surface.
Bright field: Nomarski Differential Interference Contrast
General concept. In general, optical inspection techniques can be classified into two categories: bright-field inspection and dark-field inspection. In both cases, a beam of light is shone on a surface. Bright-field inspection examines the directly reflected light, while dark-field inspection examines light that is scattered away from the direct reflection. Dark-field and bright-field inspecting systems are sensitive to different subsets in defect types. There may be some overlap between them, and indeed this overlap may provide additional information about a given defect, but in general each detection method is optimal for its own defect types.
The Nomarski bright-field inspection method addresses a class of defect modes rarely detectable in a dark-field system, because the Nomarski method is sensitive to local slope variations. The method also provides a direct measure of the defect dimensions. For example, a dimple is an artifact of the silicon polishing process and can be characterized as a shallow dish in the surface with a very small depth-to-length aspect ratio. The lateral extent is sometimes on the millimeter scale. This type of defect has very low contrast, and so is not seen in the light-scattering, dark-field channels. However, since the defect has a slope variation to it, it is detectable with the Nomarski method.
Figure 1 illustrates the basic Nomarski concept [3]. Traditionally, Nomarski is a static measurement method incorporated into a light microscope. It is normally an imaging method in the sense that the whole illuminated area, or a portion of it, is observed at once, either by the eye, a detector array, or a photographic film. It takes advantage of interference between two coherent light waves that intercept one another.
 |
Figure 1. Basic Nomarski concept: 1) beam splits into two beams; 2) surface induces a phase shift; 3) interference between the two beams causes intensity variations; and 4) intensity variations are measured.
Assume that a light beam from a laser is separated into two coherent beams with a well-defined phase relationship between them. Each of the two beams is incident on a surface, and the two subsequently reflected beams are recombined just prior to being detected. If the spots from the two beams are vertically separated, then the intensity of the image, as measured by the detector after recombining the beams, will be modulated. This modulation is derived from vertical surface variations that induce a phase shift between the two coherent beams. The phase shift, manifested in an intensity fluctuation at the detector, can then be correlated to the original surface variation. In the traditional application, the operator observes the modulated intensity, or interference fringes, over the whole image. The spatial distribution indicates the type of defect causing the interference.
Integration into a laser scanning inspection system. One of the key elements of the Nomarski technique is its inherent sensitivity. Since it examines the phase difference between two light waves, the surface feature causing the phase difference can have a displacement length that is a small fraction of the wavelength used to illuminate it. Additionally, because it is a differential technique, it has selectivity to local variations, while significantly attenuating common-mode deviations such as sample vibration.
When Nomarski is combined with a laser scanning system, the scanning process automatically determines the shape characteristics of a defect. For example, stacking faults in silicon can result in localized crystal growth above the nominal surface. Although some stacking faults are detectable in the dark-field system, the ability to determine that a particular defect is above the surface, without the need for another piece of equipment, considerably reduces cycle time.
Light scattering cannot, by itself, detect the polarity of the objects causing the scattering. The signals measured in the dark-field, or scattered-light channels are always unipolar. An object either scatters light or not and the detection method measures these scattering events. With an appropriately configured optical system, the Nomarski illumination path and the dark-field illumination path can be the same, thus providing the Nomarski measurement at the same time and at the same speed as the dark-field measurement.
The Scanning-Nomarski technique provides more than just the polarity of the material defects, however. The output of the Nomarski detection channel is configured so that both unipolar and bipolar signals are delivered. Unipolar signals can be detected when the Nomarski beams cross over a step (Fig. 2). Because Nomarski is a differential method, the phase difference between the two beams is always zero when the two beams are at the same vertical position. As the first beam steps up, the phase difference between the two beams increases until a maximum is reached when the first beam is entirely on the step and the second beam is entirely off the step. As the second beam moves up the step, the phase difference decreases until both beams are again at the same level. The result of the differential output is a positive pulse. A step down would result in a negative pulse. A bump results in an initially positive bipolar signal, as the first beam is climbing the front slope of the bump with the second beam lagging below. The signal turns negative as the first beam moves down the back slope with the second beam lagging above and finally returning to zero when the two beams are again level. A scratch or longer-range concave surface profile results in the inverse signal behavior from bumps.
 |
Figure 2. The Scanning-Nomarski technique delivers both unipolar and bipolar signals, providing the ability to distinguish "bumps," scratches, and steps.
Examples and applications
Many classes of defects are appropriate for detection in an automated Nomarski arrangement. This system is currently used to look at a subset of those defects as we learn the full range of defect types that can be detected.
Slip lines in epitaxial silicon (epi) are one of the most common and destructive defect modes in epi production. Figure 3 displays a high-level, full wafer map and histogram of events on an epi wafer detected in the Nomarski system. There are approximately equal numbers of negative and positive unipolar signals. This means that there are approximately equal numbers of slip lines that step up as step down. The signal strength (as indicated by the histogram) is related to the height of the slip lines.
 |
Figure 3. Full wafer map and histogram of events detected on an epi wafer using the Nomarski channel: a) full wafer map of all the detected events; b) magnified view of a region; c) "Phase-Haze" represents the arithmetic average slope variance across the wafer surface; d) histogram of the unipolar events detected in the main map. Zero is located in the center of the histogram. Equal numbers of negative and positive unipolar signals indicate that there are an approximately equal number of slip lines that step up as there are that step down.
Slip detection sensitivity can be characterized both in terms of the vertical displacement sensitivity and by the lateral resolution. Figure 4 shows a set of slip planes with a small lateral translation at the ends of a few of them. The lateral positioning is not yet calibrated, but an upper bound is approximately 5-10 ?m. The shading of this image also allows a good view, on the leftmost pair of slip planes, of slip that steps up and slip that steps down.
 |
Figure 4. Slip detection sensitivity can be characterized both in terms of the vertical displacement sensitivity and by the lateral resolution.
The integrated Nomarski channel is designed to detect defects that do not scatter light. In Fig. 5, a wafer was scanned in the inspection system and both the dark-field data and the bright-field data were collected simultaneously. The dark-field map, or scattered-light map (shown on the right), displays a few detected events randomly distributed across the surface. The bright field map, obtained with the integrated Nomarski channel, clearly has detected a significant amount of defects.
 |
Figure 5. a) 2-D MicroView of a defect. The defects are rings of residual material left behind after a cleaning step; b) Nomarski optical microscope image of the defect; c) cleaning residue on silicon.
Figure 6 shows a MicroView of a scratch. This image is bipolar, with both negative and positive components to the signal, thus differentiating it from the signal from a slip line.
 |
Figure 6. A MicroView of one of the scratches detected in Fig. 5. This image is bipolar, differentiating it from the signal of a slip line.
A dimple with less dimensional quality than scratches is also detectable (Fig. 7). As mentioned above, the ability to detect the vertical position of a defect is critical to the fast recognition of the defect mechanism.
 |
Figure 7. Two renderings of a microview of a material defect: a) Raw data shows slope, while b) integrated data on right shows surface representation. Events such as this one correspond to a depression caused by a polishing process anomaly.
The sensitivity to vertical displacements can be quantified using known reference materials. Since automated Nomarski systems are new, standards are as yet undefined, although a number of approaches are currently under way. One of them is derived from an artifact created by depositing titanium lines, roughly 2 mm in width and 300 ? in height, in a traceable manner and standardized.
Finally, the term Phase-Haze was mentioned above as a measure of the average slope variation in a unit area on the wafer surface. The Nomarski signal has certain spectral content that could potentially be used to qualify the roughness of the wafer in a relatively unstudied spatial frequency regime. This component of the Nomarski channel will be the subject of further work as an understanding of the information obtained from such data develops.
Future directions
The future of this technology is clearly bright. The most immediate work revolves around understanding the interactions between various defect types detected in the Nomarski detection channel and the dark-field scattered light collection channels. This work includes examining in detail each of the defect types of importance to the wafer manufacturing community. In the near term, the interactions in each of the detection systems form the basis of a learning set by which classification mechanisms can automatically detect, recognize, sort, and disposition wafers based upon the specific defects detected on the rejected wafer. Once these components are combined with spatial analysis packages, a tremendous wealth of information will be available in a very short measurement time.
Acknowledgments
The authors would like to thank Lionel Kuhlman, Mark Noakes, and Stan Stokowski of KLA-Tencor for their help in preparing this paper; Howard Huff of SEMATECH for many stimulating discussions about silicon; and finally, Dave Ruprecht from MEMC for his valuable insight to the nature of silicon growth defects.
References
1. Surfscan SP1, KLA-Tencor, One Technology Drive, Milpitas, CA 95035
2. C.T. Larson, K.P. Gross, S.E. Stokowski, "Noise Sources and Their Influence on Surface Particle Detection," Proc. Microcontamination 94, 440-454, 1994.
3. G. Nomarski, A.R. Weill, Rev. Metallurgie L11, 121, 1955.
C. THOMAS LARSON received his BS degree in physics from Arizona State University in Tempe, and is working on an MS in physics from San Francisco State University. He is product marketing manager for KLA-Tencor`s unpatterned surface inspection systems, which include the Surfscan 6000 Series and Surfscan SP1 product lines. Prior to joining KLA-Tencor, he was a research associate of the Thin Film Lab at San Francisco State University, where he developed a scanning tunneling microscope. He is a member of the ASTM committee developing a test method for distinguishing particles, haze and microroughness, as well as the Particle Microroughness Task Force with SEMATECH. KLA-Tencor, Wafer Inspection Division, One Technology Drive, Milpitas, CA 95035; ph 408/875-4200, fax 408/434-4270.
MEHDI VAEZ-IRAVANI received his PhD degree in electronic and electrical engineering from University College London. He is a principal research scientist at KLA-Tencor`s Wafer Inspection Division, working on both patterned and unpatterned wafer inspection systems. Prior to joining KLA-Tencor, he was on the faculty of the Center for Imaging Science, Rochester Institute of Technology, where he was engaged in research in the area of near-field scanning optical microscopy, other probe techniques, and ultrahigh sensitivity detection systems.