How CD-SEMs Complement Scatterometry
Like competing Siamese twins, CD-SEM and scatterometry enable and push each other to higher levels of capabilities and evolving applications.
Alexander E. Braun, Senior Editor -- Semiconductor International, 6/1/2009
Ever since scatterometry became available to IC manufacturers, it has competed with CD-scanning electron microscopy (CD-SEM), with each technology continuously improving (Fig. 1). From a roadmap perspective, this competition has yielded good results. A CD-SEM can now analyze a couple of lines and get the critical dimension (CD); it is progressing toward simultaneously analyzing several lines, which scatterometry already does, and both techniques are getting faster. SEMs now produce tilted images to image down sidewalls and get sidewall angle data, matching one of scatterometry's strengths.
Now the two technologies are converging, according to Alain Diebold, Empire Innovation Professor of Nanoscale Science at the College of Nanoscale Science and Engineering (CNSE) of the University at Albany (New York). "As SEMs improve their electron optics and image analysis software, scatterometry adds wavelengths and moves toward the UV, gets better modeling capability, becomes faster, and adopts improved optics," he said. "With the advances in CD-SEM, both methods now measure the CD of more than one line at a time, giving them improved precision."
Development and production
CD-SEM and optical CD (OCD) scatterometry face steeper hurdles than just dimensional shrinkage. These come from double patterning, whether spacer double patterning or techniques with more processing steps. This procedure to sustain 193 immersion lithography for still smaller features complicates metrology (Fig. 2). In double patterning, often two sets of CDs will have sidewall angles, making analysis difficult for either method.
|
| 2. Double patterning creates significant challenges for CD measurement. In this top-down image, the line shapes and CDs of two sets of lines are clearly visible. (Source: University at Albany) |
"There shouldn't be any concern over these two technologies hitting fundamental barriers anytime soon," Diebold said. "We still have tricks up our sleeves to improve them. Some of the work is going on here at Albany, and we're always looking to do a better job on CD measurements, as with all metrology."
However, there are concerns that metrology for double patterning is not being sufficiently advanced. "There isn't enough R&D and advances in scatterometry or CD-SEM to help us attain the needed precision, or enough metrology development work for looking at these double-patterning processing schemes," Diebold said. This work requires the masks, equipment and wafer runs — expensive metrology activity that is perennially underfunded.
"We can do it here under the appropriate circumstances; we have the equipment," Diebold said. "We hope to do additional work with our partners, exploring scatterometry and CD-SEM's limits for double patterning. However, it's more complicated than in the old days, and there are unresolved issues."
The damage that electrons or photons can inflict on a sample while measuring it is one such issue. With CD-SEM-induced photoresist shrinkage, a long-standing problem, clever approaches were developed. Several other approaches may exist, but they must be researched. As Diebold put it, "It'll require a focused R&D effort just to determine how serious these problems are, and then solve them for the coming sets of smaller dimensions and novel materials."
Today, most fabs have scatterometry capabilities, but the CD-SEM remains the workhorse. It provides better data because it can be used on all device layers, and if there are any issues it is easier to track the problem's origin. Spatial resolution problems remain the same: getting CD-SEMs and scatterometers to yield higher-resolution measurements. New capabilities, better lenses and optics for scatterometry help push back the wall.
Challenges can be divided into engineering and production, said Ram Peltinov, CD-SEM product line manager for process diagnostic and control at Applied Materials (Rehovot, Israel). "Probably the most attractive application for scatterometry is its ability to become integrated, allowing wafer-to-wafer control together with sidewall information."
Peltinov added that development has become "increasingly metrology-intense, from OPC model calibration and verification, scanner and resist optimizations, test chips optimizing process integration, up to the production ramp. SEM is central to all these applications, time to market is key, and for that you must see your process and understand your variability budget from edge roughness, through local to global variation. There are hurdles, but solutions can be had within a very short turnaround time."
Limited throughput is a roadblock, said Joseph Race, product marketing manager for semiconductor TEM products at FEI Co. (Hillsboro, Ore.). "Also, variation in sample preparation processes significantly affects S/TEM results, driving the need for more repeatable, efficient methodology," he said. "We're working with our applications development teams and strategic partners to streamline the process, from advanced, automated sample preparation to a variety of contrast-enhanced imaging and high-sensitivity analytical techniques." Although TEM tomography shows increasing promise for providing needed 3-D information on interfaces and defects, Race said that critical data cannot come at a cost of $10,000 or $20,000 per image.
From a productivity standpoint, OCD scatterometry comes down to the metrics of cost per sample or samples per hour. The goal is not just shorter measurement times but shorter time to obtain a valid measurement — ensuring that the output number from an incredibly complicated 3-D model is correct and correlates with real data.
Key differences
Scatterometry provides information such as undercut, depth, sidewall angles and profile information that top-down SEMs cannot provide, noted Craig MacNaughton, senior marketing manager at KLA-Tencor Corp. (San Jose). These parameters are critical to device yield, and must be controlled inline.
"Scatterometry provides a better estimate of process mean. With a SEM you measure effectively a 1 or 2 μm square picture of a set of lines, spaces or structures," MacNaughton said. "Line-to-line variability within a SEM can have up to 4% variation from one to the next. With scatterometry, you're averaging over a 40 or 50 μm bar, which provides a better mean process estimate." Although scatterometry's precision and matching is not quite an order of magnitude better than SEM's, it certainly is 2–3× better (Fig. 3).
New materials cannot be disregarded, MacNaughton noted. "Look what happened with CD-SEM and 193 resist. First, there was an ongoing battle to prove that you couldn't damage the resist, and then it became a matter of making the damage consistent. We see that more and more with other materials, such as ultralow-ks being brought into the back of the line. Some e-beam techniques affect them almost as much as the 193 resist did. Not only does it become difficult to measure due to charging issues, but measurement uncertainty accuracy comes into question."
Noam Shintel, corporate marketing director for Nova Measuring Instruments (Rehovoth, Israel), thinks that scatterometry is ready for both 32 and 22 nm. "Scatterometry has replaced CD-SEM for most advanced applications for measurement and control device manufacturing. OCD measurements take about one second to provide angstrom-level precision for a full high-k gate profile, including sidewall, undercuts, thicknesses, width and rounding," he said, adding that in periodic structures, scatterometry measures critical process profiles invisible to CD-SEM, which is being gradually relegated to niche applications such as optical proximity correction (OPC) and line edge roughness (LER) monitoring (Fig. 4).
|
| 4. Scatterometry excels in the capability to measure bowing in 3-D structures, which are invisible to CD-SEM. (Source: Nova Measuring Instruments) |
Improved control, taking into account both wafer-to-wafer and within-wafer variability components, will require increasingly denser sampling plans, Shintel said. "Because of its faster measurement time, lower cost per point, and a wealth of profile information, scatterometry is better-suited than CD-SEM for dense sampling plans, making it useful for complex process control," he said. "Additionally, scatterometry can be integrated into various process tools, enabling both feedback and feedforward control schemes."
For scatterometry, averaging is built in: it measures global structure properties over tens or hundreds of lines in die, or on test structures — critical for process control. Its main challenge is modeling the geometry's complex features. A critical prerequisite of a good measurement is accurate spectrum response with high signal-to-noise obtained from the optical system. The most stable, accurate and highest signal-to-noise system is a polarized normal-incidence spectral scatterometer that does not suffer from the angle-of-incidence uncertainties of the common spectral ellipsometer. Normal-incidence polarized reflectometry also provides angstrom-level tool-to-tool and fleet matching.
For a long time, scatterometry has worked at a regime where the half-pitch is an order of magnitude smaller than the wavelength used. Going from 45 nm to 16 nm changes nothing in this respect; on the other hand, CD-SEM is largely based on the premise that the e-beam spot size is significantly smaller than the features measured. As this assumption breaks down with shrinking CD values, the CD-SEM will be forced into modeling the complex interaction between the beam and the sample to get accurate metrology data. Scatterometry thrives on smaller geometries — as the pattern density of the features increases, more diffraction signal is available for extraction of geometrical information.
The modeling quandary
Scatterometry and CD-SEM are complementary, said Brennan Peterson, a technologist at FEI. "The primary hurdle in scatterometry is that it's difficult to go to anything real with its spectral information. It comes down to how much you trust — and how easily you can develop — your models. People use cross-sections from a SEM, TEM or STEM to determine how close the actual device's physical layout is to the scatterometry models, to resolve whether these truly represent them and can be tracked through time." Figure 5 shows a comparison of TEM and scatterometry measurements.
Peterson added that e-beam metrology acts as a reference for scatterometry's spectral data. "We're working with a scatterometry company to get true integrated validation between the systems, because there are scale and calibration differences," he said. "It all depends upon the validation needed, with the added complication that you must know your material's composition and variations." For example, scatterometry results will differ depending on whether an interface is rough or interdiffused.
FEI's modeling perspective is that manufacturers will start applying direct physical calibrations by overlaying SEM and TEM data. They need better reference data rather than better modeling, Peterson said. "This is a more reasonable approach than trying to make your own reference," he said. "The fundamental calculations won't get any easier; true physical referencing will always be a difficult issue." You must be clever about applying the data that you can get now, whether that is tomography and holography on the TEM side, or 3-D reconstructions from the SEM or TEM data. "You can overlay those and either directly, analytically, or just by looking determine if the model fits the physical condition."
Metrology developments are no longer solely a matter of new platform technology, but of smarter application of the data they provide. Complex code must be created for models, but in reality, most of what is meant by imaging — at least in 2-D — is grayscale imaging, and it is a matter of doing a better job to ensure these images turn into real data. As Peterson put it, "I don't think we'll have fundamentally new technologies; we'll just continue improving them and make them applicable to their respective domains."
3-D modeling is much more complex. "I'm not even sure if much of what we think of as measurements makes much sense in 3-D," Peterson said. "What does it mean to measure the distance between the sidewall of two different 2-D planes? Roughness is material on a different level; the way we think about statistics, line edges and equivalent four planar edges isn't really the same. There's equivalence, certainly in the math; but how do you minimize errors to get the needed output? For the full 3-D modeling of datasets we can build models, but I don't know what the next step is."
Peterson indicated a need for an industry-wide standard of what 3-D modeling results should look like. It took a long time to establish how to use LER vs. linewidth roughness (LWR), two different variables. Going to 3-D squares the number of factors involved. These are not fundamental technical hurdles, but more a matter of thinking through problems more clearly and understanding which data is needed, Peterson said.
32 and 22 nm
One of 22 nm's biggest changes is the implementation of complex 3-D structures such as finFETs and buried word line structures for memory, according to Tom Gubiotti, product marketing engineer for KLA-Tencor. "There's always the option to create test structures on the wafer that can be measured with other methods, but there really isn't a way to directly measure the final product," he said. Scatterometry can close this gap; it provides non-destructive measurements on the actual structures.
Matt Pietrucha, marketing manager for the metrology business at Rudolph Technologies (Flanders, N.J.), is convinced that CD-SEM and OCD scatterometry will stick around. "Our customers use both — the OCD for more of the inline, high-throughput, high-volume monitoring requirements, which provides a better CoO; but then there are capabilities that CD-SEM has that OCD lacks."
Chris Morath, vice president of operations for Rudolph's metrology business unit, concurs. "I think customers would like to have linewidth, LER capability in an optical tool, but nothing really exists right now, so that's definitely a CD-SEM application."
There is interest not only in measuring the structure or the CD, but also film thickness and optical properties as a combined OCD measurement. The CD-SEM cannot do this. With OCD there is proven sensitivity, but being capable to meet stringent gate requirements for production control is still a tall order. "For high-k it isn't just thickness, but composition, and it gets more complicated with advanced high-k gates," Morath said. "There are monolayers of lanthanum oxide, a workfunction tuning metal. Ideally, one should do these measurements on patterned structures. However, because the manufacturers themselves do not fully know what their processes are going be like in three years, they want as many options as possible."
Non-Visual Defects Impacting Semiconductor Yields
Ralph Spicer, Qcept Technologies, Atlanta
Wafer surface condition is becoming increasingly critical to achieving enhanced device performance at smaller design rules. Since wafer cleaning and surface preparation are the most repeated process steps in semiconductor fabrication, yield management must ensure that these steps result in a well-controlled surface condition.
In the past, yield management focused almost exclusively on physical defects such as particles, microscratches and bridging. However, as device performance improvements are increasingly being driven by materials innovation (e.g., high-k/metal gates, dual work-function materials, low-k dielectrics and SOI) and process innovation (e.g., single-wafer wet cleans, edge/backside cleans and atomic layer deposition), process margins associated with wafer cleaning and surface preparation are becoming tighter, giving rise to non-visual defects (NVDs).
 |
| Typical NVDs caused by cleaning and surface preparation steps. |
NVDs include contamination (both metallic and organic) and process-induced charging (Figure). NVDs cause changes to the wafer surface's electrical or chemical properties that can lead to poor adhesion of subsequent layers, or degraded device parametrics or reliability. Although they can cover large areas of a wafer's surface, they are often sub-monolayer or electrical in nature, having a Z-height of a single atom or molecule. Feedback from several leading-edge semiconductor fabs indicates that NVDs may account for up to 30% of all defects. Because NVDs have an increasing impact on yield, inline NVD detection is becoming an important component of semiconductor yield management strategies.
