Metrology Adapts to Meet CD Measurement Needs

While most agree that critical dimension (CD) metrology — whether off-line, in-line, or in situ — is a necessity, traditionally, deep in their non-politically correct hearts, most manufacturers have looked upon it as a "necessary evil," a "non-value-added expense." Traditions change and, in this case, motivated by the need to keep control of increasingly testy (pun intended) processes, metrology is being considered in a new light, as an integral part of the process itself (Fig. 1).

"CD metrology requires more than just measuring a line's bottom. It's what we call 'shape metrology,'" said Yogev Barak, director of SEM technologies at Applied Materials (Rehovot, Israel). "Users need to measure a feature's entire shape, whether two- or three-dimensional. They must know the bottom and top CD, slope, roughness."

"The CD-SEM is losing steam," said Steven Berger, vice president of microelectronic products at FEI (Hillsboro, Ore.). "This isn't related to repeatability — its main criterion — but to the fact users require different information. They must understand what's going on in the profile."

"We must make measurements on a range of new materials — such as low-k dielectrics," said Chris Mack, vice president for lithography solutions at KLA-Tencor (San Jose). "Dielectrics charge up, interfering with the CD-SEM's capability to image patterns and make accurate measurements. Some time ago we developed a charge equalization technology for our CD-SEMs to enable mask manufacturers to handle glass substrates, which take tremendous charges in the CD-SEM. This is now useful for new materials."

 
1. Various technologies, working separately or complementing one another, will be needed to meet the challenges of new materials and processes. (Source: Philips Analytical)

No single technology today does everything — the solution lies in a combination of technologies. Some will come from R&D labs, and others currently off-line will go in-line. Applied expects SEM platforms to be fab workhorses for at least another five years, through technologies that extend their usefulness. "Several measurement technologies are available, such as scatterometry, SEM-based ones capable of measuring multiple CD perspectives — 3-D. There are dual-beam technologies — FIBs — also various AFMs," Barak said. "Most are R&D devices, with some serving as in-line metrology tools. Each has advantages and disadvantages."

Complementary metrology eventually will be integrated into the process tool itself, he said. "Some metrology capabilities will be integrated, others will remain stand-alones. SEMs are the fab engineer's eyes because they allow him to image the process and understand what's happening beyond just parametric measurements."

He added that all parametric measurements cannot be integrated into a single result. "Manufacturers must combine them instead of using one as they are used today."

Neil Hanson, Applied's director of silicon etch product management, views CD control as critical for etch. "Before, etch equipment concentrated on CD-bias results, not the incoming or outgoing CD. Etch equipment only cared about the difference between lithography and etch."

Applied has been working with optical critical dimension (OCD) companies and integrating their technology into its platforms, such as its patterning system, to influence the outcome of the final CD. "This could have happened years ago, but computers of the day weren't fast enough," Hanson said. "Measurements would have taken minutes with the computers of the time. Now, it can be done in real time."

Not too much profile information can be had from a top SEM image/measurement. "An OCD system gives profile information, shows what the line looks like coming into the tool," Hanson explained. "This profile information could be used to make very subtle changes to the final CD. OCD can easily be integrated into etch tools because OCD works at atmospheric environments — the wafer goes into the factory interface and OCD measures it on its way into the etch tool."

However, because the OCD uses a grating structure on the wafer, unlike a SEM, it cannot measure fully isolated lines. Manufacturers need better repeatability and resolution and, as we go into 193 nm photoresist, e-beams can cause CD changes in photoresist (Fig. 2). The OCD diffraction optics technique eliminates this problem, making it easier to feed data forward into etch tools.

 
2. As CDs shrink and new, more sensitive photoresists are introduced, dosage latitudes change significantly. (Source: ITRS 2001)

Applied partnered with Nanometrics for this solution, which uses a normal incidence method that looks straight down at the wafer, allowing real-time calculations based on the data gathered from the measurement.

Therma-Wave (Fremont, Calif.), recently introduced an optical, non-destructive, real-time measurement method that reduces the need for cross sections or AFM measurements. Its Opti-Probe RT/CD uses proprietary software in a real-time mode to provide CD measurements for detailed cross-section profiles, including CD information at select locations on the profile, as well as height, depth and sidewall profile.

The CD system is the first to use spectroscopic ellipsometry with real-time regression to calculate and display results. It eliminates the need for off-line modeling and solution libraries, enabling profiles to be calculated in seconds.

The spectral ellipsometry approach provides more information content on sidewall profile and shape than is typically available from normal incidence-based techniques. The technology is compatible with 0.13 µm technology node needs, and is extendible below 0.10 µm for a wide range of process applications.

John Heaton, president and CEO of Nanometrics (Milpitas, Calif.), views integrated metrology as inevitable. "We're centered on metrology in the process tool. We focused in on CDs to integrate metrology in the etch tool to monitor and control CDs on an ongoing basis for Applied's Transforma silicon etch system, and came up with a tool for pre-measuring CDs prior to processing within the etch chamber."

Nanometrics developed a normal incidence system that uses broadband polarized light and a real-time regression engine (Fig. 3). It calculates from the spectrum what the sidewall angle is, the top and bottom CDs, and information about the profiles in general. Because most CD-SEMs look from the top, they are blind to undercutting and incapable of resolving the footing at the feature's bottom — it cannot recreate the profiles without a cross-sectional inspection.

3. The OCD architecture enables the user to obtain profile information, which can then be fed back to make changes in final CDs. A disadvantage of the system is that it cannot measure fully isolated lines. Its greatest advantage is that, unlike e-beam, it does not cause CD changes in photoresist. (Source: Nanometrics)

Lithography is a metrology driver — smaller CDs equal tighter process windows. "Information is required about effects that turn up in the profile as you get closer to the process window edge," FEI's Berger said. "Wall angle, which starts to drift outside of what is typically needed for etch, is an example. But it isn't just that, it's also what's going on at the resist profile's foot: scumming, undercutting — all affecting etch."

In the back end, copper may change the whole philosophy of CD measurements. "Damascene is inherently a 3-D structure," he explained. "Understanding what goes on in it requires a cross section: You must see if over-etch damaged the low-k material, whether you've got your barrier material down, the seed's properly laid, that there aren't any voids, etc." Also, when parameters change (the plating bath's chemistry, for example), resulting defects are not as well understood as tungsten and aluminum's. Understanding defects, because of the damascene process's nature, requires more profile information.

"CD-SEMs won't disappear, but we'll see other tools on the fab floor; scatterometers, for example," Berger said. Scatterometry is a clean way of measuring profiles, but it is limited because it can only look at specific test patterns in the curve, making it a "garbage in, garbage out" measurement.

"The certainty that dual beam gives — you cut it open, get an image, take measurements, and see what you measure — has a place," he added. "Picking up defects in the copper damascene process, whether through voltage contrast or other electrical techniques, is insufficient. You still don't know what it means until you cut it open and see that, for instance, the etch didn't go through or the metal didn't go down."

The concept that a single CD number is sufficient to understand and describe all the necessary data is vanishing. CD metrology goes beyond just CD; it is shape metrology. We are looking at complex, 3-D shapes on the wafer that must be controlled within a narrow margin to get good product. One CD is inadequate to describe a 3-D shape.

"CD-SEMs are two-dimensional; they look top-down and give the wafer's X-Y plane," KLA's Mack said. "You need a shape metric — 3-D information."

To this end, the top-down information provided by the CD-SEM is complemented by scatterometry, which also provides a two-dimensional measurement, albeit a different one: X and Z. "This gives the shape of the resist profile and a cross section," he added. "It's essential information for defining process windows. The window of which range of focus and exposure is acceptable relies on more than how the CD responds to focus and exposure, but also on how the profile's shape responds to it."

As CDs shrink and new materials become increasingly touchier and fragile, metrology is evolving toward more indirect means of measurement, relying on software. "Our CD metrology group is working on what we call 'structured scanning,'" Mack said. "This is the art of minimizing a sample's total dosage when making a measurement. Before, the standard CD measurement required a picture of the sample, which was used to extract the CD. When you do this, you're giving the sample a respectable — often unnecessary — dose."

With a CD-SEM this can be done more precisely. It is possible to control where the beam is scanned, and procedures such as focusing off the feature of interest can be done, focusing the scans very quickly, illuminating solely the area of interest.

Philips Analytical (Chandler, Ariz.) is doing considerable ellipsometry work, pushing performance — repeatability and reproducibility, said Alec Reader, semiconductor line business manager. "The sigma values we're getting at the moment are typically at the 0.02 Å level — truly subatomic."

Philips is characterizing monolayers and sub-monolayers with X-ray diffraction. "Manufacturers are looking at SiGe channels in MOS structures at the moment," Reader said. "Some of the channels that they are growing are typically 50 Å thick."

Manufacturers want to measure the thin-layer stacks, not only for the gate stacks — oxide and polyoxide stacks. "We're pushing to get better sigmas," he said. "We've modeled the ellipsometer's entire optics train, and can improve it from 0.04 to the 0.02 level by looking at things like heating transients. The 0.25 σ plate had a heating hysteresis, which led to a miniscule variation in the sigma value. By modeling the whole train, we improved critical components."

Rudolph Technologies (Flanders, N.J.) is considering the barrier sidewall coverage question. "Looking at post-CMP copper line arrays, we discovered evidence of acoustic vibrations of the nitride sidewall — the section between the copper seed and the oxide, between the copper lines," said Chris Morath, director of product development.

When the sidewall signals were detected, Rudolph was studying the vibration of copper line arrays. Currently, they have simulation capabilities of the structures' vibrational modes, and are researching how to connect this with experimental data. "Basically, you look at a time-resolved reflectivity signal and when you excite a line array using a 10-15 µm spot size — for a narrow copper line array you're looking at 0.2 µm linewidth — all the lines vibrate and the resulting reflectivity signal is a sum of all the lines and dielectrics between them," Morath said. The result is a complicated pattern with several frequency components.

The approach looks at the frequencies in the data and relates them to the line structure's simulated vibrational modes. The simulation varies different line dimension properties — width, spacing, sidewall angle and thickness — to find a combination that matches experimental data to the simulation's frequencies. Presently, Brown University (Providence, R.I.) researchers have shown it is possible to obtain very close agreement — within a couple of percentage points — to a SEM measurement of a structure.

Although optical techniques such as OCD and scatterometry are coming on-line, the CD-SEM remains today's standard for CD metrology, and other technologies are viewed as complementary rather than competing. "The CD-SEM is the best option for most critical processes, and the only one for measuring critical contacts and vias," said Pete Knutrud, director of marketing at Schlumberger Verification Systems (Concord, Mass.).

He concedes that there are hurdles. "Recently, MIT presented papers at SPIE showing the capability to measure 40-50 nm and less gate structures. The problem is that 130 nm process control requires additional measurements of in-circuit features, more characterization of what goes on in the actual circuit. The imaging of high-aspect-ratio structures, 10:1 and greater, especially in DRAMs, is a major CD metrology requirement. Finally, with the increased use of reticle enhancement techniques such as off-axis illumination, OPC, and phase shift, features on the wafers must be analyzed for fidelity. This can only be done with image-based CD metrology techniques such as the CD-SEM."

The major problem facing CD-SEMs is 193 nm resist loss under e-beam exposure. This is a function of landing energy. Low landing energies are required to avoid damage. Today's CD-SEMs run reliably at 400 V. To go below 400 V it is necessary to consider the CD-SEM's total design — not a minor thing. Future CD-SEMs must be designed to handle sensitive resists required for 100 nm processes, not just 193 nm.

Metrology is fast becoming part of the manufacturing process and being designed into it from the beginning. The entire test structure — measurement points in the process flow — is being designed while considering sampling, how often measuring takes place. Metrology is mutating into a comprehensive system, no longer just an add-on used once in a while to see how the manufacturing process is proceeding. By designing metrology as part of the overall flow, it seamlessly becomes part of characterization and APC.