Metrology for Litho Seeks Standards, Integration

That metrology for lithography will not get easier is a consensus shared by many, including Anantha Sethuraman, vice president of product marketing, FEI Co. (Hillsboro, Ore.): "Even now, with high-end devices, there's a push toward correlating physical CDs with electrical CDs." Bents Kidron, director of marketing and product manager at Nova Measuring Instruments (Rehovoth, Israel) worries that immersion technology is knocking at the door, but its metrology requirements are still undefined.
 
According to Kevin Monahan, vice president of technology, Patterning Solutions Group for KLA-Tencor (San Jose), pattern-limited yield in the immersion lithography area will be a big challenge for metrology. As he puts it, "At the 65 nm node and beyond, it'll be the key manufacturing problem in early production, requiring more sampling and higher capital expenditure on process control." John Heaton, CEO of Nanometrics (Milpitas, Calif.) views incomplete and uncorrelated process information resulting from using only one number as a serious obstacle. "In lithography, there are only three major measurements in the metrology space: CD, overlay and film thickness."

FEI's Sethuraman sees the correlation of physical CDs with electrical CDs as being of major importance. "It begins the convergence of design, process and test — the start of the global DFM/DFT revolution. Devices are as good as the designed structures and how the patterns turn out. If a pattern doesn't fit with the final dimensions, it's worthless." There will be resist questions, because it is uncertain yet which chemistry will emerge triumphant. "Resist chemistries will change, creating different issues, until someone hits upon a formulation that stands up to exposure and is robust in a number of focus windows."

For metrology, resist e-beam exposure and degradation is a major issue (Fig. 1 ). Another is structural metrology's implementation — controlling features that, in actuality, are 3-D structures. "We've been lucky to get away with 2-D measurements," Sethuraman said. "Now we must control at 3-D. How? Can one use an AFM, scatterometry? But how do you reference and monitor these tools?" Scatterometry and the AFM provide enough data for quick and dirty process control information. "But do we know if these instruments provide what's required? There's no single standard showing how it's done — no known solutions."

1. At gate level, there is a short between two tungsten lines. Observable are very long lines in the aisles. The shorts were found using e-beam inspection, and the holes were caused by residual carbon residue from a stripping step. (Source: FEI)

Sethuraman wonders how nanoimprint lithography will be controlled. "The process is deceptively simple," he said. "You're making wafer waffles — putting the resist on the wafer and punching it out. To control it, we need a different methodology and move away from in situ metrology." Sethuraman believes molecular imprint will require standalone metrology; possibly a combination of destructive and non-destructive measurement. "To control the process and ensure that the imprint has created the needed aspect ratio, cross-sectioning may become necessary. Here's where physical and electrical CD correlation become a line control factor."

KLA-Tencor views pattern-limited yield as a manufacturing obstacle. "There's a design-to-process yield gap resulting from the interaction of more complex designs with shrinking process windows," Monahan said. "Different feature sizes are being integrated on a reticle using OPC (optical proximity correction), PSM (phase-shift masks) and off-axis illumination in conjunction with everything else. With immersion lithography, the additional complication of polarization dependence — resulting from high NA (numerical aperture) — is added. Focus depth is declining, as are exposure latitudes, and the lithography process window shrinks with every node."

Considerable CD error will originate from hidden systematic variations. This includes intra-wafer dose and focus errors — across the wafer and across the field — that produce broadened CD and profile distributions, affecting performance and yield. Although immersion technology may create a temporary depth of focus (DOF) windfall at the 90 nm node (Fig. 2 ), the DOF problem reappears with a vengeance at 65 nm and below. At the 90 nm node, users will not go to hyper-NA (>1). However, as the effective NA becomes greater than unity, edge-die and scan defocusing effects are expected across the wafer, in addition to intra-field issues, all of which will require monitoring. "Since we can model expected yield losses, we can show how ultrasensitive, 1-2 nm resolution line-end shortening (LES) focus-exposure monitors and spectroscopic ellipsometry — new technologies for CD and profile control — can drastically reduce the influence of hidden error influence on pattern-limited yield," Monahan said.

2. One of the realities of immersion lithography is that focus monitors will be critical at <75 nm. (Source: KLA-Tencor)

In the era of pattern-limited yield, metrology convergence is becoming important. An example is the use of technology such as spectroscopic ellipsometry (SE). Because of its high signal-to-noise ratio, SE has remarkable technology node and layer extendibility. In addition, it can perform multiple measurements at one site. "The classic problem is STI (shallow trench isolation). Manufacturers have yield problems with it, sometimes losing 2% or more in die output," Monahan said. "Much of this loss is due to oxide voiding, a result of poor dimensional control. STI control requires accurate, precise measurement of CD, profile, trench depth and film thickness. The drawback has been that, in a standard line, a CD-SEM is used for CD, an AFM for trench depth, and an ellipsometer for film thickness." Three tools using different technologies, measuring at different times, create measurement correlation problems.

"At one of TI's logic facilities, we simultaneously measured all critical parameters for STI process control with a single, in-line measurement," Monahan said. "We used a state-of-the-art standalone tool. This single, in-line measurement cut cost, improved capability and eliminated three specialized tools, because one technology could make multiple profile measurements, including CD, trench depth and film thickness. We've also been successful at the gate level, looking at footing and notching — unobservable with CD-SEMs. Notching and footing correlate well to drive current measurements. If you compare SE measurements to traditional electrical CD measurements, SE can give a 60 to 70% correlation to drive current, where the rest of drive current is controlled by other causes, but ECD measurement — which supposedly informs about drive current — may have a <10% correlation."

"At smaller geometries, there are correlations and relationships between CD, overlay and film thickness measurements," said Nanometrics' Heaton. "Nobody has really combined these three metrologies to determine where these correlations exist. These are three different measurements, but there's considerable coupling between them that must be explored. Presently, 1-D or 2-D information is insufficient, and 3-D and pattern fidelity data are becoming critical. One CD dimension number — X, Y or Z —is no longer enough, but there's considerable 3-D profile information available in these features' topography."

Metrology targets no longer reflect real patterns' process sensitivity. Traditionally, designers have placed a small test feature in the scribe line and provided process control mechanisms. Now, the move is to measure in the active area. "Here's where micro changes take place within the die," Heaton said. "CMP is an example — there are substantial changes between tightly and broadly spaced lines, which remove at different rates. Looking at a scribe line measurement pad doesn't represent what happens inside the die."

At <100 nm, calibration standards are non-existent. When optical CD (OCD) technologies were introduced, these were compared to SEMs. "They correlate well, but only after several SEM measurements," Heaton said. "It's difficult to solidify what a nanometer or an angstrom is on these different features. With three or more different metrology platforms, correlating all this data becomes a problem." (Fig. 3 )

3. A 91-point CD uniformity die map (left) with 11 nm total CD variation is compared with a 25-point CD uniformity map (right) of the same die. It shows a 9.5 nm total CD variation caused by a smaller data set. (Source: Nanometrics)

Traditionally, users viewed SEM as a type of point-by-point metrology. They go on a wafer and check certain die areas for whatever dimensions they seek, whether CDs or film thickness. "You lose considerable information that way," Heaton said. "There's a misunderstanding about how much data is needed. Users worry that new metrologies, particularly optical critical metrology, give too much." The problem is not data overabundance, but interpretation.

"Engineers get so much data and so many samplings from these metrologies that they have difficulties determining its significance, because no one has done a good job on the software," Heaton said. "Plus, essentially there are no standards — you really don't know what a 'nanometer' is — so everything becomes statistical metrology." The reverse is how to look at data and draw conclusions from 5000 wafer measurements. "It's possible to do 1000 measurement points within one die to understand, on a small scale, what's happening in the process. Then you repeat this across the wafer. At the leading edge of 65 nm and below, sampling philosophies have had to change to understand what's going on with die."

The sampling required to characterize and monitor a process will spiral up substantially, as will the capability to measure data across the wafer — making it easy to measure 10,000 and 20,000 sites. "Engineers will be able to determine the relationship between space on a wafer and the CD," Heaton said, "to get an understanding of the hot plate, for example, or of how the photoresist is being deposited or spun on. We've always done this point-by-point — you take five or nine sites' measurements and proceed. Now, many will require considerably more measurements, as well as correlating film thickness, the CD, overlay, all together and being able to separate data." This additional metrology could impact throughput; however, if it is moved to each individual tool, standalone metrology platform bottlenecks are eliminated.

"Traditional metrology cannot measure parameters such as line-edge roughness or some reticle factors," Heaton said. "Reticles have become complicated, and their characterization relative to the wafer, the exposure tool, the resist, the track process — the whole integration — hasn't been tightly controlled. Once, litho and etch relationships concerned us; now it's the relationship of the exposure tool and the reticle. Although it's been possible to scale current lithography with advanced masks, these have also become complicated: they're multilayer, have different materials. There are several kinds of films, and how those are judged and how they're going to print on the wafer still isn't well understood."

As Nova's Kidron sees it, metrology for lithography faces five problems. "The first is getting more data on patterned wafers in real time — not just CDs but also sidewall angles, complete profile, resist thickness on patterned wafers, focus vs. exposure error effects — to correct the process using APC. The second is the need for real-time excursion control and high-throughput patterned wafer inspection for macro and micro defects. Then there is the systematic biasing of the results due to systematic process/tool errors and inadequate modeling, maintaining the measurement accuracy in the presence of process variations, and finally, shrinking real estate's effect on overlay and CD measurement." Kidron believes that the growing need for more data in real time can be addressed by the introduction and assimilation of faster metrology tools integrated into process platforms.

Robert Crowell, product marketing manager for Clean Track systems at Tokyo Electron America (Austin, Texas), sees the challenge for CD and macro-defect inspection as taking standalone capabilities, such as state-of-the-art 300 mm processing at the 90 nm technology node, and miniaturizing metrology so it fits into the track and operates it at that run rate.

"Customers want standalone tool functionality and better detection capabilities — to go from a 50 µm to a 20, 10 µm detection capability," Crowell said. "The technology is pattern recognition, where wafers are scanned using a high-resolution video camera. The output is analyzed using a system that automatically identifies and classifies defects such as a poor developer, hot spot, missing flash or missing pattern." Crowell added that users will not trade throughput for working devices. "Even if all devices pass, if testing takes longer, the user rejects the trade-off. Although sampling rates are rising, sampling in the track is different. If a user operates at 180 wph, and the tool can operate at only 150 wph, we'll inspect every second wafer, which is still more than what off-line tools can do."

This is different for CD. Users typically intensely sample one or two wafers per lot. They will sample 20 or 30 sites on the first wafer. "We'll sample perhaps 10 sites, and then every other wafer," Crowell said. "Thus, we maintain litho cell throughput without slowing down the track and scanner." In-line metrology tools aim at matching the scanner's throughput. There are, however, limitations for overlay. "The overlay modules run at about 80-90 wph, with maybe a five- or six-site sampling." Conventional metrology is targeted at specific wafers in one lot, instead of all. Even then, it might not be a sampling of the entire image field area, so there is insufficient information to determine whether variations are occurring in the stepper lens, or in imaging capability. A benefit of integrated metrology is providing that information on a production-worthy and practical level beyond what is possible today with standalone tools.

Timbre's ODP (optical digital profilometry) offers additional advantages with respect to integration in process tools over the CD-SEM approach and optimized tool algorithms. With scatterometry, usually more representative information is obtained than is possible on a CD-SEM, with more reliable data. Since a grating is used, the result is an averaging — more representative information about what is going on — as opposed to looking at just one specific line. ODP scatterometry has shown increased value in process characterization of monitoring tests such as analyzing Bossung curve behavior. The issue is whether the metrology can be made faster, smaller, and placed in the harsh track environment, with hot plates, chemicals, vibration — everything that standalone metrology tools have been isolated from.

Jon Opsal, CTO at Therma-Wave (Fremont, Calif.), sees scatterometry-based OCD metrology emerging as a solution for lithography metrology's limits today, and even evolving into a universal metrology solution. "Fundamentally, the challenges in metrology for lithography stem from how to measure and characterize the ever-smaller CDs being fabricated," he said. "Higher-yield and higher-speed devices using less power require better CD metrology and detailed CD profile information. We're seeing considerable significant work in developing advanced optical methods for CD metrology (OCD) as the viable alternative." According to Opsal, the need is for an ever-tighter ratio between precision (P) — the metrology tool's measurement variation — and tolerance (T) — an IC's specification limits. "One just needs to refer to the various CD line items on the ITRS to see that there are many red flags for the P/T ratio in future technology nodes. Increasingly, there's high confidence in the combination of scatterometry with SE as a robust solution for those roadmap blips."

To fully appreciate OCD's capabilities for advanced applications, it is necessary to consider how the engines behind optical metrology have evolved. "During optical metrology's early stages, numerical algorithms — the models — were the challenge," Opsal said. "Models today are very complete, capable of handling full conical diffractions from line-space structures, as well as 3-D structures, such as contact holes, and the more complex line-space structures that are still 2-D but have layers on top." Today's OCD metrology is built on more mature algorithms — be it rigorous coupled wave analysis, finite difference time domain, real-space finite difference, etc. — which provide significant benefit and data for production lithography. "We're even at the point where, with a given metrology tool, we can make choices between algorithms based on which one is optimal for the particular structure being measured," he added.

The reverse of this refined-algorithm capability is that today's OCD tools can be cost-effectively equipped with sufficient computing power to solve complex metrology for lithography situations; for example, 20 GHz processing today, with perhaps 100 GHz in a few years. A contact-hole metrology scenario is a true 3-D scattering problem. Doing it to the required degree of precision takes about 100× the computing power compared with a classic 2-D scattering problem.

"As the industry moves into the 45 nm node, we'll look at structures with dimensions of 30 nm or less," Opsal said. "You need repeatability and precision on the order of 0.1 nm or less. The challenges involve the data's precision and also how much data or information one has. It's like looking at something extremely small, which you cannot tell whether it is rectangular or circular, but if you move around and look at it from another angle, it becomes possible to begin mapping out its shape. This analogy describes what happens with optical scatterometry's content-rich nature, which is at the heart of advanced OCD metrology today. Scatterometry has been configured to provide rich information content about CDs and film in all kinds of advanced IC structures." Opsal noted that, as metrology for lithography advances, it is important to ensure that the data collected is relevant to the problem, and that there is a sufficient amount of it.

It would appear that, over the next three years, a major challenge will be the transfer from 90 to 65 nm and the capability to carry on — and over — some of the tooling for 90 nm, as well as the capability to implement 65 nm.