Metrology Develops to Measure Thinner Films Better

By 2007, 65 nm will be the leading-edge technology for manufacturing, while 45 and 35 nm will be in R&D. Requirements multiply with each node, driving toward more thin-film metrology integration, although standalone thin-film metrology will not disappear. These added control parameters increasingly require additional film characterization measurements. Some thickness-only measurements will be integrated into process platforms using established technologies such as ellipsometry, opto-acoustic pulse, optical reflectometry, or eddy current as a core.
 
"Conventional spectroscopic ellipsometry fades at about 220 nm — the spectrum's UV portion," said Murali Narasimhan, senior director of strategic marketing, films and surface technology division at KLA-Tencor (San Jose). "There's a 190 SE extension available, routinely used to monitor ARC layers or ultrathin ONO in flash memory, but even it is running out of steam. Process engineers are looking at further extensions into the so-called 'vacuum UV,' where optics operate in a vacuum, avoiding moisture and oxygen absorption of photons between the 190 and 150 UV range."

Sean Jameson, vice president of sales and marketing at Jordan Valley Semiconductors Inc. (Austin, Texas), points out that most state-of-the-art inline thin-film characterization metrologies use some form of X-ray. "This includes inline XRR, XRF, XRD, SAXS and XPS," he said, adding that advanced applications require not only thickness but also density or composition measurements. "Process control technologies that simply provide thickness and uniformity cannot control the total process. Without the capability to measure these parameters, material property assumptions can cause errors in thickness measurements."

"At the 180 and 130 nm nodes, metrology was primarily used for statistical process control (SPC), with measurements made at various process steps to ensure each was in spec," said Ajit Paranjpe, vice president of marketing and applications at Therma-Wave (Fremont, Calif.). "Many measurements were performed on blanket and monitor wafers. There wasn't much emphasis on measurements on product wafers, because it was difficult to do and technology was mostly unavailable. Also, for the time, SPC was adequate."

KLA's Narasimhan notes that many insulating films, especially high-k dielectrics, have properties leading to better absorption and optical response for vacuum UV measurement. "Their high bandgap is close enough to the spectrum's 150 vacuum part (Fig. 1 ). Thus, we measure not just thickness, but also composition," Narasimhan said. "With high-k materials, composition plays a considerable part in their electrical properties, and this correlates to the dielectric constant."

1. Improved measurement capability may be had at smaller design rules, using 150 nm wavelength spectroscopic ellipsometry (150 SE). On the left, 150 vs. 190 nm sensitivity to nitrogen concentration in 90/65 nm node SiON dielectrics is shown. On the right, SE sensitivity to compositional changes in 65/45 nm node high-k dielectrics increases below 190 nm, enabling inline optical composition monitoring. (Source: KLA-Tencor)

Over the years, on the electrical side, the equivalent oxide thickness (EOT) — a combination of the thickness times the dielectric constant — has been measured and monitored inline to control gate oxide. At 90 nm, gate oxide thickness typically is <20 Å and heavily nitrided to minimize leakage between the gate and the substrate. Increased leakage makes capacitance measurements more difficult. Now, engineers are resorting to actual leakage measurements, because leakage has become increasingly critical. There is so much reliance on these measurements that capacitance-based EOT measurements are being abandoned.

"With high-k materials, the emphasis on measurement changes to interface traps, because when high-k materials touch down to the silicon, the interface gets complicated," Narasimhan said. "High-k material deposition techniques are still immature and leave behind interface defects. Thus, there's more emphasis on density of interface traps (D_it) measurements, and we're working to improve the measurement capabilities in these areas: leakage at 90 nm and D_it measurements for 65 and 45 nm."

Early new metrology technology applications lie with the gate dielectric oxynitride. The reason for this is that, while optical ways of correlating changes in refractive index to composition are a fast, inexpensive method of monitoring composition, there is no fundamental way of measuring concentration and composition if an excursion occurs. "Optical methods are often correlated, especially when you have a sub-20 Å film and something changes, you don't know whether it's the thickness or the composition. Optically, you can get non-unique solutions that won't show which way to turn the reactor knobs to return to baseline," Narasimhan said.

E-beam stimulated dispersive X-ray (ESX) spectrometry can independently measure X-rays from the oxygen and nitrogen atoms in the thin gate dielectric. "Since you know how many atoms of oxygen and nitride there are, you know the composition and can accurately differentiate between thickness and compositional changes by using this technique," Narasimhan said. "Some of our customers use the optical method and our ESX monitoring solution for calibration, others drop optical altogether and go 100% with ESX." For ALD layers, especially HfAlO₂ in use in production for 80 nm DRAM, thickness and composition can be independently monitored that way. Composition often correlates well to the dielectric constant, which controls the DRAM's charge storage capacity.

"With transparent films, there's a small number of techniques used, and much commonality," said Michael Gostein, chief technologist for Philips AMS (Natick, Mass.). "Therefore, the user selects from among several suppliers on the basis of reliability, cost, service, etc. For instance, you can select an ellipsometer from any number of suppliers."

However, with newer metal films there are several technologies (many proprietary to single suppliers), many of which are specialized for a certain set of applications. In the case of copper-film measurements, mature technologies provided by many suppliers are being used, such as X-ray fluorescence and X-ray reflectivity. However, these are not being used online in high-volume production, because they are not fast enough or they lack a small enough spot size for product wafer features. "When you consider solving an online measurement problem, options are more limited and the enabling technologies become more proprietary," Gostein said. "Overall, the challenge isn't so much how you're going to do a particular measurement, but how to do it cost effectively. One major point to consider is how the metrology ties in with APC. With metal films, a key production application is the ability to monitor plated or post-polished copper thickness. However, an engineer doesn't necessarily need to sample wafers one at a time. He wants to feed the data into an APC system, mine it for information, and use the results to make real-time process adjustments. Until recently, the ability to cost-effectively implement APC hasn't been achievable with online copper metrology. However, with some of the new measurement techniques available for copper metrology, and a measurement strategy for which of the process issues most need to be monitored, APC can be fully implemented for copper."

Chris Morath, director of marketing at Rudolph Technologies (Flanders, N.J.), sees metal thin-film metrology under control for 90 nm. "But shrinking barriers and the changeover to low-k dielectrics will present challenges at 65 and 45 nm. To maintain low resistance, manufacturers may switch from relatively thick Ta/TaN barrier layers to ultrathin ALD materials. ALD film metrology requirements will pose a challenge to both X-ray-based and opto-acoustic approaches. ALD barrier integration with low-k dielectrics is additionally complicated by complex surface chemistry and pore structure. Therefore, measurements of barrier/dielectric interfacial properties, such as roughness and adhesion, will become increasingly important."

Morath added that, for effective low-k process control, it is not sufficient to measure film thickness. "It is also critical to measure the film's capability to withstand CMP and packaging stresses. Opto-acoustic techniques have the capability to rapidly and nondestructively measure low-k elastic modulus, which is well-correlated with mechanical performance."

"Interconnect processes are evolving quickly, but even more changes are occurring in the gate stack," Morath said. "Sub-20 Å nitrided gate oxides are the norm at 90 nm and require unprecedented long-term reproducibility of a few tenths of an angstrom. Ellipsometry can provide the required results and, combined with DUV reflectometry, can also provide nitrogen concentration measurements that show excellent correlation with electrical techniques." (Fig. 2 )

2. The combination of laser ellipsometry (633 nm) and DUV reflectometry enables thickness measurements of both layers of high-k/SiON gate stacks. Here, the high-k HfO2 layer was measured to be 34 Å (2T), and the underlying SiON layer was measured to be 7 Å (1T). The fit between the measured and modeled results provides accurate and repeatable measurements. (Source: Rudolph Technologies)

At 65 nm, Morath anticipates device makers will begin moving to high-k dielectrics such as HfO₂ . "These materials will probably require an ultrathin interfacial SiON layer. Optical techniques can provide measurements of these advanced gate stacks. But gate advances won't end there. At 45 nm, high-k dielectrics may be accompanied by metal gate electrodes. Characterizing these gate stacks will require a combination of opaque and transparent metrology technologies."

Accurate measurement of thin films requires consideration of three factors, according to Dr. Moshe Finarov, CTO of Nova Measuring Instruments Ltd. (Rehovoth, Israel). "The first is metrology performance (repeatability, stability, tool-to-tool matching as well as small spot size) of the optical measurement tool — ellipsometer, reflectometer or any combination. The second is optical modeling — an accurate optical model of the measured structure to enable correct fit of measured and simulated optical parameters. Such models account not only for major effects like n and k dispersion and polarization, but also effects like surface roughness, optical non-uniformity within each layer, internal scattering, etc. Also, sensitivity of measured parameters — ellipsometric angles, reflectance at basic polarizations, diffraction efficiency, etc. — to the measured material properties is very important. Usually, the shorter the wavelength, the higher sensitivity to film thickness and optical constants."

Since sensitivity to very thin-film thickness and optical properties is low, the major problem is process variation effects on measurement results, which are usually not covered by any applied optical model. If measuring oxide residues of several nanometers' thickness on a thicker silicon nitride, any nitride thickness and composition variation will have a stronger effect on the measured reflectance spectrum or ellipsometric parameters than the oxide residues themselves. "There's an injection method for cases where underlying stacks are measured in the same measurement sites, prior to producing the very thin layer on top," Finarov said. "The stack parameters are 'injected' to the final stack's optical model, and process variations are well accounted for." Integration metrology enables the best implementation of such an injection approach. (Fig. 3 )

3. The 2-D map on the left shows oxide residue distribution over a 300 mm wafer after STI CMP, measured in the range of 0-240 Å. The graphic on the right shows the total uncertainty of the thickness measurement, which includes dynamic repeatability (3σ) and tool-to-tool matching, as a function of oxide residue thickness. Through the applied injection method, this value can be as small as a single angstrom, even for ultrathin oxide residues. (Source: Nova Measuring Instruments)

In two or three years, Finarov believes there will be problems with high-k gate dielectrics, true low-k dielectrics, and interface layers (such as barrier and cap), which have complex and probably unstable compositions. "It'll be difficult to determine film thickness and optical properties separately," he said. "Probably it'll be necessary to combine different measurement technologies, such as ellipsometry and XRR."

Therma-Wave's Paranjpe sees increased emphasis on product wafer measurements. "The reliance on monitor wafer measurements is rapidly dropping. Although commonly used to monitor equipment state, they aren't used to completely monitor the process state, since a test wafer isn't necessarily representative of what takes place on a product wafer. Thus, the metrology must be capable of performing measurements on product wafers, and measure those parameters that eventually tie to device performance and yield."

Paranjpe also sees process control as evolving from simple SPC to APC. "We're seeing many of these advanced process control loops, especially for some key steps such as gate patterning and maybe the STI step. We're likely to see this in some BEOL metallization steps. This will require standalone metrology with higher sampling rates, to make measurements on product wafers without affecting overall cycle time." Integrated metrology is emerging because it allows higher sampling rates than standalone metrology, without degrading cycle time. Also, measurement latency, in terms of turnaround or measurement impact on an actual process step, is shorter. Thus, a correction may be made much more rapidly to the tool.

"We're seeing rapid increases in the number and types of materials we must measure," Paranjpe said. "Before, most measurements were performed on simple dielectric stacks that didn't involve several layers. Now, in the dielectric stack itself, we're seeing increased complexity with multiple layers that have unique attributes. Often, we must discern subtle differences in optical properties between the layers. Also, there are fairly dramatic differences in thickness between stack layers. So you're trying to measure thin and relatively thick films with high precision and good linearity."

Another aspect is new materials. Some, like high-k dielectrics, are not yet being used in logic devices, but are in use for DRAMs. Others are materials such as SiGe for the source drain, requiring that several parameters such as germanium content, doping level, and perhaps even strain be measured. Silicides are also getting increasingly thinner — NiSi has been introduced as a replacement for CoSi₂.

In patterning, many antireflective coating materials are being introduced. "ARC materials consist of both organic spin-on coatings and inorganic films," Paranjpe said. "The latter are basically silicon oxynitrides of differing stoichiometries. These are fine-tuned to achieve low reflectance for the particular stack of interest. Because stack reflectivity must be low, it's important to measure these stacks inclusive of the increasingly complex underlayers that are being patterned. Before, for example, the gate pattern was simply a poly gate pattern using resist. Then a BARC was added, then a hard mask underneath that, increasing stack complexity. You must measure all parameters of interest and preferably on product wafers."

Antireflective coating materials often exhibit unique optical properties, including anisotropy, which complicates the measurement, since anisotropy affects total reflectance and the film's optical characteristics. A metrology solution that incorporates multiple optical technologies provides an accurate determination of optical properties including anisotropy. By measuring and fitting the spectra using independent optical techniques such as spectroscopic ellipsometry (SE) and beam profile reflectometry (BPR), an unambiguous determination of thickness and optical dispersion can be made. For this application, an SE- or BPR-only measurement results in significant measurement uncertainty (up to 15%), which is unacceptable for advanced antireflective coatings.

"BEOL dielectric stacks for copper-based interconnects are becoming more complicated. Before, perhaps it was a simple barrier layer, a low-k material, and another barrier layer for a three-layer stack," Paranjpe said. "Most today are using about four layers, possibly going up to five or six. Each is some form of subtly different oxide, either doped with nitrogen or with carbon. Also, it's important to know the low-k's porosity — all this, plus the other parameters that must be measured simultaneously. Note that measuring these layers individually isn't truly representative, because film characteristics change when processed. As film is processed, its porosity might change and interlayers may form due to reactions. Often, low-k films are densified via a surface treatment to improve adhesion to the hard mask. Metrology must be capable of picking up all these subtle variations and yet accurately measure thickness."

Thinning is a problem in measuring thin-film layers, according to Jordan Valley Semiconductor's Mazor. Channel, gates, barriers and conductors are all thinning, resulting in density and stoichiometry variability. "Increased resolution is an answer — a smaller incident wavelength, for example," said Mazor, "and measurement technologies that don't rely on material assumptions. Sensitivity to process parameters like density (low-, high-k films), crystallizations, crystal orientation, roughness, etc., seem best solved by measurement technologies uniquely targeted at them. These aren't parameters controlled through correlation to other resultant material changes."

Stack complexity and stack separation examples are BEOL copper or barrier layers, and FEOL SiGe and strained silicon substrate. "Thinning complicates matters," Mazor said. "You need enough signal to identify these as stacks vs. some effective single layer. Increased resolution helps, as does a technology that doesn't require fixed or assumed material constants, or there can be significant error."

The worst thin-film metrology obstacle will probably be measuring the full gate stack. Currently, it is poly on a nitrided oxide, and the nitrided oxide and poly steps are done separately. There is discussion about integrating the dielectric and gate electrode depositions, so there may be the complexity of trying to measure these thin layers that currently are nitrided oxide, but which could change to ALD high-k materials — a multilayer stack with subtle compositional variations including multiple constituents: silicon, oxygen, nitrogen and carbon, all of which change subtly.