Defect Detection and Review Enter New Era
In-line and off-line ADC, spatial defect signature analysis, adaptive sampling, yield impact assessment and expert ADC will be key in reducing time to decision and product at risk.
Alexander E. Braun, Associate Editor -- Semiconductor International, 5/1/1998
Because today's defects are getting smaller and more critical, it is increasingly difficult to get a robust recipe that works over variable production layers to find these defects, with a minimum of nuisance rates. The new platforms are reducing and automating setup time. The trend is to move to a common look and field and a common setup interface across all wafer inspection platforms, but it is not exactly proceeding at breakneck speed.
Patterned wafer tools are more sophisticated and harder to use. Since that is true for all areas of process equipment, the trend is toward streamlining software -- the more expert the system, the less expert the technician that runs it.
Inspection systems are advancing on an evolutionary, not revolutionary, track. Top equipment manufacturers already have or are about to have next-generation unpatterned inspection platforms that meet today's 0.18 µm (180 nm) design rule requirements in terms of sensitivity. Work is under way to extend this to 0.13 µm (130 nm) design rules.
Unsurprisingly, the industry continues to follow lithography's path, getting more out of optics than once believed possible. It still mostly operates with white light, 550 nm xenon illumination. This probably will be sufficient to go below 0.18 µm, and 488 nm lasers are already in use. Just as steppers migrated to g- and then i-line before making the painful and expensive shift to DUV, inspection platforms have followed the same trail.
Although many prefer using unpatterned wafers and monitor wafers, there is a strong move toward patterned wafer tool monitoring. Fabs working their way to 300 mm must consider wafer cost and therefore patterned wafer monitoring. At first, this appears as the more expensive way to do things until the cost of monitor wafers is factored in, together with a much quicker signal and response to process excursions. The use of patterned wafers will become established, because real product wafers permit more accurate simulations of what is really happening in the process. However, blank wafer inspection, review and analysis still seems to be the fastest and most direct method to get at tool-level defects. Optical review, the old white-light standby, has taken the industry down to 0.3-0.4 µm geometries. Beyond that (and Intel is pushing the limits to 0.13 µm), problems arise. A solution is laser confocal technology, a technique between optical and SEM. Laser confocal technology will find applications for 0.18 µm. However, at 0.13 µm and beyond, UV must be used since it provides twice the resolution, and SEM begins to come into its own.
Defect detection vs. defect count
Defect count is not as important as it once was. Most fabs now want to know defect types and their yield impact. Off-line classification is particularly valuable for excursions. Many foundry fabs are doing less baseline and more excursion monitoring. They avoid doing defect classification except when there are no excursions. It is here that off-line accurate defect classification (ADC) be-comes attractive.
The challenge is fast time-to-results with ADC. This may be accomplished through in-line binning or a high-resolution ADC that although takes some overhead, does not significantly affect the inspector platform. Considerable capability can be gained by getting quick results on defect classification on-line. Clearly, resolution -- as is throughput -- is a challenge, which is why SEM is attractive.
Some experts distinguish between defect detection and review methods and techniques like cross-sectional SEM, FIB analysis and FTIR. As one expert, who asked not to be identified, put it, "Defect detection and review typically are done in an nondestructive fashion on production wafers. The only time you want to do cross-sectional SEM, FIB or FTIR is if you really have a problem and you don't mind trashing a few wafers to do it." Not everybody is enthusiastic about cross sectioning. "People say FIBs are good enough to enable you to cut out a die and look at it, leaving the rest undamaged -- I don't believe it. You have a huge particle source spouting out material as you blast the hole, and if you go through a couple of layers, who knows what byproducts are going to result. The odds you may end ruining an entire wafer are good."
There are a number of laser-based defect detection tools being used for
analysis where high sensitivity --
0.1 µm -- is necessary and
throughput is not an issue. Other systems are aimed at production
monitoring uses, where sensitivity requirements are lower and throughput
matters. They may use a darkfield-only architecture to pick up light
scattering off the target and its defects. A system like that is
designed for high-speed production monitoring to spot excursions. One
platform adds brightfield capability to spot excursions quickly.
Simply detecting the presence of something does not mean one knows what it is. Here is where defect classification comes in. There are two drivers that are moving equipment manufacturers toward new products: smaller geometries and automatic defect detection (ADC).
Systems from various manufacturers provide different ways of doing ADC. Some come with defect classifications already programmed in, while others apply user-defined categories. ADC gives the engineer a quick way to slot defects into categories. Different systems work essentially in similar ways: They find a defect, take its picture and run it through the software, which then decides what it is. The problem is that on an optical system, both the software and the engineer have to make an informed decision by essentially looking at a fuzzy dot.
Coping with the fuzzy dot
When confronted with anything below 0.25 µm, the best optical microscopes can only produce a fuzzy dot. This can reduce the best opinion on what the defect is to little better than an informed guess. In-line defect review so far has been limited to light optics. With the cost of defect inspection systems as much as five times that of off-line optical review stations (RSs), the choice is to use RSs wherever possible for defect review and classification. This changed with the availability of on-line ADC on the inspection tool itself, where defects are reviewed and classified after inspection. Provided the ADC is accurate, this approach is faster and less costly.
The off-line RS has changed over the years. With the need to look at increasingly smaller defects, confocal systems have become popular to improve the resolution of light optics and extend its use in defect review, much like phase shift masks, OPC, etc., are used to extend optical lithography's life. This has been complicated by 0.25 µm design rules, however, and it is evident that light optics alone can no longer do the job.
SEM ramps up
E-beam is the obvious next defect review step. SEM review has been around for years, with key players such as AMRAY (Bedford, Mass.), Hitachi (Tokyo, Japan) and JEOL (Tokyo, Japan). These vendors have had full wafer capability and the ability to read and use defect files. Until now, their use and need was perceived as limited.
SEMATECH (Austin, Texas) recognized this problem and argued for the
development of a SEM-based automatic defect review, defect
classification system. Last year, it identified SEM-based ADC as an
important 0.25 µm technology advantage and contracted with Applied
Materials' (Santa Clara, Calif.) Process Diagnostics and Control Group
to develop this capability, resulting in Applied's 9300 DR-SEM platform
(lead photo). The system is designed for rapid in-line classification.
It selects a recipe and runs the tool automatically: It loads the wafer,
aligns it, redetects the defect (defect redetection being necessary to
get a high-magnification
image), captures the SEM image and performs
ADC (if chosen) before moving onto the next defect. It is specified at
12 sec/defect.
According to Prasanna Chitturi, product manager for review products at Applied, this high-speed imaging and review capability makes possible the use of a SEM for rapid in-line classification. "It brings the SEM into common use in manufacturing lines. Several operations have abandoned optical review for 0.25 µm and 0.18 µm lines." The 9300 uses standard maps from defect inspection systems to pinpoint the location of defects and provides imaging and analysis capabilities with its multiple perspective SEM imaging (MPSI) proprietary technology. It can include EDX capabilities for defect compositional analysis. MPSI's multiple detector array simultaneously images a defect in multiple perspectives. Each perspective contributes different information about the defect's nature.
As a result of its recent acquisition of AMRAY, KLA-Tencor (Milpitas, Calif.) now has both SEM- and laser confocal-based review technologies for patterned and unpatterned wafer inspection systems. The SEM review station has full-tilt/EDS for materials analysis, while the laser confocal review station provides the imaging capability needed for process monitoring and troubleshooting on advanced devices. There is also an off-line ADC feature.
ADC is becoming the standard with inspection systems as well. Automated inspection is easier to use, and the data themselves are more valuable when distilled into a more usable form. A staple complaint has been that if it were possible to inspect at more points in the fab, quicker value could be obtained from the information and the data. Historically, just lining up review stations and operators to classify that information has been a costly exercise. It still is done, and it is useful, but it is becoming more difficult to extract useful information.
Within a couple of years or so, 300 mm will be on-line in a reasonable number of fabs. Nobody is doing production defect detection tools in the 0.18 µm and 0.15 µm (150 nm) range, but 0.18 µm is starting, and people are looking below that. At that point, SEM capabilities and review are going to be far more critical, and ADC more important, with less reliance on people and more on computers and expert systems to look at defects and make decisions.
CMP and laser scattering
|
2. The ILM-2230 is a high-performance, laser-imaging darkfield patterned wafer inspection system for production line monitoring. It combines oblique-angle darkfield illumination with small-pixel, high data rate image processing to capture yield-critical defects on 0.25 µm and smaller devices.(Source: KLA-Tencor) |
Defect inspection has be-come more sophisticated, with multiple systems in use for a number of years. Laser scattering-based inspection tools and high-sensitivity, high-resolution image processing technology tools have been adopted for different, somewhat complementary, applications, but requirements differ across the fab. For that reason, both technologies will continue being used where they best fit the requirements.
ADC and in-line monitoring made it possible to see all critical defects at production speeds. The big next transition is the integration to an intelligent line monitor, integrating detection with classification and analysis for fast answers with little operator intervention. This will be an attempt to improve system productivity with respect to engineering time and the operator time necessary to get information and value out of the systems to make decisions about yield.
According to Dave Icke, vice president of marketing for the Wafer Inspection Group at KLA-Tencor, technology has changed with the shift from 0.5 µm to 0.35 µm and now into 0.25 µm. "We have customers doing 0.25 µm, at least in development. For them, the advantage lies in a modular architecture that offers multiple sensitivity settings in the form of pixel sizes and different objectives." Icke added that as new device generations appear, engineers must go to smaller pixel sizes. "The capability to align images accurately to subpixel bases allows them to inspect below 0.25 µm levels."
Not too long ago the consensus was that as design rules got smaller, laser scattering would disappear. Now, according to Icke, "It is a dominant technology for tool monitoring, first with unpatterned wafers, and now with the use of patterned wafers for tool monitoring, particularly when qualifying $8 million and up litho cells."
Laser scattering systems' high speeds are well suited to monitoring applications because an engineer can quickly get a result showing whether the lot can be run. For years, laser scattering dominated planar layer inspection, post-deposition, and is now a powerful CMP inspection tool. Film layer numbers continue to increase, so there is a solid need for laser scattering systems in tool monitoring, which makes up for some of the method's limitations such as the size of the laser spot and the data rate used to process the information.
Last March, KLA-Tencor introduced its ILM-2230 laser-imaging darkfield patterned wafer inspection system (Fig. 2). The platform combines oblique angle darkfield illumination with small-pixel, high data rate digital image processing. Aimed at defect capture on 0.25 µm and smaller devices, it is designed for advanced interconnect process-inspection applications, such as CMP. "CMP introduces new yield-limiting defect types and noise sources," Icke said.
The new system is optimized for the detection of all unique oxide, metal, poly and shallow-trench isolation CMP defect types, including microscratches, pattern deformations, residual metal/tungsten and microbridging. It provides adjustable illumination angles, allowing it to be configured to fit the wafer's surface grain and pattern noise. It has an up to 20 wph throughput.
Imaging at the atomic level
3. Shown is a high-resolution TEM image of GaAs-Si interface (8 million times magnification). Lattice mismatch at the interface and resulting strain have caused V-shaped defects that are visible directly as discontinuity in the atomic arrays. (Source: Philips Electron Optics) |
As dimensions get smaller, the characterization of smaller volumes of materials is becoming necessary to the point where, according to John Fahy, product manager for TEM at Philips Electron Optics (Mahwah, N.J.), "You should be able to image and characterize several rows of atoms." An interface or deposited layer may be only at most half a dozen or a dozen atoms row wide. The imaging system resolution has to be better. However, the sample itself becomes a limitation because determining what is the chemistry is normally done either by using X-rays generated in the sample by the e-beam or by the electron energy lost in transmission through it.
The problem with X-ray generation is that a thick sample allows X-rays to come from a far larger area than the beam actually hits. This is because as the e-beam spreads in the sample, it causes secondary fluorescence because of X-ray generation, which then generates other X-rays. So even though a thick sample is hit with a beam a few angstroms in diameter, the information comes from up to 1 µm because of this fluorescence effect.
Since it is desirable to localize information, particularly chemical data, it is necessary to go to thinner materials. This leads to imaging in transmission mode, where it is possible to combine the resolution to see individual atom rows with the ability to collect X-ray data that the user knows will only come from an area another diameter larger than the area being hit.
"Ultimately," predicted Fahy, "it will be necessary to know the interface's electron structure, to know structurally what kinds of atoms are there and to know what the bonding state is."
Although TEM can do this, systems of this capability are intended for the use of high-end research people -- Ph.D.s with years' practice. The industry's challenge is to make this level of technology available on a shift basis on the fab -- to make the instrument relatively simple to use and easy to interpret. In short, a Ph.D. tool must be turned into a technician's tool, possibly through some sort of expert software (Fig. 3).
"Macros and very careful optics design could make it possible to do pushbutton operation between two or three principal modes, to use the machine like a SEM with a very simple pushbutton control," Fahy pointed out. "The instrument will be a transmission instrument, but not necessarily a TEM; it might be a scanning transmission instrument." The jury is still out on which would be the best, though the weight of experience lies on the side of TEM. Fahy said he believes that the requirement, when doing scanning transmission, will be that the probe must be kept at atomic dimensions.
Recent advances in focused ion beam technology now allow TEM samples to be made without breaking the wafer or moving it from the fab.
As matters stand today, both in-line and off-line ADC, spatial defect signature analysis, adaptive sampling, yield impact assessment and expert ADC will play key roles in reducing time to decision and product at risk. Defect detection is migrating closer to the defect source, and as such, development to integrate defect detection into process equipment is being accelerated. On-wafer defect detection and characterization in situ process control sensors will become necessary for cost-effective, high-volume 300 mm and beyond manufacturing. Unfortunately, the necessary technology currently needed for finding defects and correlating them to yield is not yet available.
