Interconnect Metrology Confidently Looks at 32 nm

As the pursuit of Moore's Law forced the dominance of interconnect delay over gate delay, aluminum gave way to copper's lower resistivity. This introduced a new deposition process, dual damascene, where the dielectric is patterned before the metal and coated with liner and seed layers to protect the copper and facilitate deposition. The copper is electrochemically deposited to fill these trenches and holes, followed by chemical mechanical planarization (CMP) and a dielectric cap on the resulting surface to protect it. The process worked, but the need to keep it in spec brought metrology to center stage (Fig. 1).

1. As new materials are introduced and integrated, maintaining control over these films to attain device specs is more dependent on the capabilities of inspection and measurement tools. While current metrology technology seems capable of continuing over the next two nodes or so, it is clear that new developments will be needed past the 32 nm node. (Source: KLA-Tencor)

"CMP metrology arose to keep parameters within desired process windows and to monitor consumable use," said Wei-Yung Hsu, managing director of the Metal planarization division silicon systems group at Applied Materials (San Jose). "Metrology and process control are critical CMP enablers, although the focus now is increasingly on intrinsic process-related enablers. For copper CMP in BEOL interconnect processes, maintaining within-wafer and wafer-to-wafer uniformity is critical to tight sheet resistance distribution and high Cpk. Within-wafer uniformity of copper removal is essential to control within-wafer dishing and erosion variation, and enable a tighter sheet resistance distribution. Variations arise primarily due to consumables' performance drift; for example, removal rate drifts during the barrier and dielectric polish step of copper CMP typically originate from pad wear." (Fig. 2)

"At the 90 and 65 nm node, just measuring CD in the BEOL was challenging, because one measured on a low-k material. The question was getting a precise measurement from CD-SEMs on that low-k material," recalled Wayne McMillan, marketing director of optical CD products at KLA-Tencor (San Jose).

"Sidewall angles are getting steeper. This affects barrier/seed because over the next two or five years, a technological end approaches for PVD, tantalum nitride, tantalum barrier plus copper seed and electroplating film. All this is complicated by lithography changes that will pose fundamental challenges," indicated Brennan Paterson, key account technologist at FEI Co. (Hillsboro, Ore.).

2. CMP is a process that requires in situ measurement of copper thickness during planarization to attain the required specs. Increasingly, other processes are following suit. (Source: Applied Materials)

Via and contact measurement is more complicated, because 3-D modeling is more involved. Software using more advanced algorithms helps, and metrology OEMs are developing these applications, principally for 45 nm. Users want to measure via line depth and determine if the etch did not go through the low-k causing an open or, if it overetched, causing contact or via resistance issues.

Another focus area is copper CMP. Some monitor this optically. Optical critical dimension (OCD) measurement is attractive because it provides precise copper thickness measurements after polishing, and the data can be fed back to the polisher for optimization. "We can also measure copper volume of the cross-sectional area within the line, which gives a good reading on the copper lines' sheet resistance," McMillan said. By measuring volume, electrical performance is predicted and the polishing is optimized for the best electrical performance.

At 32 nm, porous low-k will be used, with possible air gaps for vias. Present copper technology — barrier/seed and the copper itself — will persist for at least the next couple of nodes. There are new film applications with cobalt tungsten phosphide (CoWP) coatings to cap the copper, which helps with electromigration. There are other materials for etch, such as system-on-a-chip (SoC), that are tricky to measure on film multistacks.

According to Ahmed Khan, general manager of the optical films division at KLA-Tencor, the requirement for thickness measurement is shifting into thickness and composition. "Users want to measure thickness, composition, porosity, pore size and distribution to do a controlled low-k film," he said. "At 45 nm film thickness, composition and porosity are sufficient; but at 32 and 22 nm, with the possible introduction of air gaps, it'll be important to do pore size distribution measurements." Spectroscopic ellipsometry can be used because it enables the simultaneous measurement of these materials' thickness and index. From that, compositional properties and porosity may be calculated.

Metrology challenges

"There are two main metrology hurdles for 45 nm and beyond. First, new materials and second, integration," said Adrian Wilson, marketing director for films products at KLA-Tencor. New materials are becoming more complex and thinner. The inter-level dielectric (ILD) liner and stop layers and the low-k capping layers now require not only thickness measurements, but also composition measurements. Many of these films' adhesion and stopping properties are composition-dependent. For smaller design rules, composition is critical because it determines film behavior. Likewise, with TaN and capping layers such as CoWP, their impacts on film interconnect and electromigration is influenced by composition thickness. Pore sizing and distribution are also becoming important for low-k.

"At 45 and 32 nm, all those properties must be measured, but on patterned wafers or in die, not solid scribe lines or proxy structures; the metrology tool must manage path topography and conduct measurements," Wilson said. A dielectric pattern metrology module is being used with spectroscopic ellipsometry (SE) technology, which enables on-pattern structure measurements and derives the film thickness, index, co-associated composition and porosity.

Some film types are affected by the pattern itself — its dimensions. A line and space ratio change can alter some films' composition and that of the more advanced capping layers. As one deposits on a small structure, one composition results; done on a larger structure, the composition itself changes. "This leads to the integration challenge," Wilson said. "If we take a sample of the integration of some of the layers, we find it's dependent on film composition and stress. Additional stresses take place at 45 and 32 nm, becoming critical measurement requirements." Tools are now provided with stress capability, and produce film maps to determine the degree of stress locally or across the wafer.

Steeper sidewalls

Barrier layer changes will require improved metallurgical control of interfaces and the stack, although barrier and seed layers are composed of other materials. "Currently, barrier layers are moving toward ~2 nm on sidewalls, and at their thinnest point, seed layers are equivalent," Paterson said. "There's a 4 nm stack with at least three different phases of material that must be accurately controlled.

"A superior scatterometer and scatterometric model must be developed, showing what the stack will look like," Paterson stated. "Better models will be more in demand than better metrology, and be less expensive." Anyhow, data can always be obtained electrically; the problem is interpretation — electrical data must be turned into actionable engineering shapes. TCAD or equivalent modeling must improve. Post-etch will have huge metrology demands. This will be hard if there are air gaps in the dielectric, or yet another low-k. Extended use of scatterometers is almost certain to ensure shapes are equivalent.

At 45 nm, everyone worries about what happens at the front end and gate transitions. A change in barrier/seed architecture or air gap dielectric adoption requires considerable development. The party may end at 32 nm. These changes may be postponed until 22 nm, but new technologies will be needed then.

Complexity, data

The shift from one technology node to the next is resulting in increased complexity. "For metrology suppliers, this is good news, as more measurements are needed both in the FEOL and BEOL," said Noam Shintel, corporate marketing manager at Nova Instruments (Rehovot, Israel). "Before, CD was the reference measurement. At 90 nm and beyond, one must know a structure's full profile — sidewall angle, rounding, foot and notch, and undercuts."

While front-end-of-line (FEOL) processes are more demanding and get considerable attention, BEOL process challenges are not much different. Specifically, in damascene processes, copper line resistivity is correlated to profile; therefore, the trench profile must be measured before copper electroplating. After CMP, the copper line thickness is measured and results are fed backward to control the CMP process.

Shintel thinks that for these requirements, metrology systems that are fast and provide accurate full-shape profile will be needed. Optical methods give the best overall solution in throughput, accuracy, fleet matching and cost of ownership, while others like CD-SEM fall behind on throughput and capability to measure some attributes of the full profile after etch or copper line thickness after CMP.

Traditionally, hardware was the dominant part of a metrology solution; however, application complexity requires that software tools for highly automated 2-D/3-D modeling and application development become an integral part of the solution.

Ye Feng, director of key account applications at Nanometrics Inc. (Milpitas, Calif.), sees major challenges for interconnect metrology in providing data for resistance-capacitance (RC) and resistivity measurements. "The data needed is layer thickness, sidewall angles, CD, etc. As we approach 22 nm, scaling and complexity become problems. On scaling, we face three hurdles: new materials, new structures and different design rules — all requiring more efficient scatterometry."

Scaling brings with it tighter specs. "As we talk about angstrom-level precision and tool matching specs, accuracy is no longer a well-defined metric," Feng said. "If a spec states accuracy down to 1 Å, nobody really understands this, because it's tool matching that matters." Even if the perfect structure was made into a perfect model and crunched through a perfect engine, it would still be off from experimental spectra because the optical constants of materials of micron-sized dimensions can no longer be uniquely determined. "We're modeling thickness dimensions together with optical constants," Feng said. "When it comes to modeling, understanding the process and material variations is important; otherwise, the metrology capability suffers." A solution is additional OCD sampling steps to better decouple the parameter and get more data.

"Another scaling factor is complexity," said Kevin Heidrich, product director at Nanometrics. "We can have a copper grating or resist grating on top of a copper grating. Or something like an electroplated CoWP layer. What was a problem in one layer is more complicated in two, with additional intricacy added by new materials. Additional sampling is required to enable angstrom-level precision."

Measuring particles

Prevention is necessary when it comes to particles. "For the 45 nm technology node, the ITRS specifies defectivity levels on various types of films for particle sizes as small as 54 nm," said Bob Havemann, vice president of process integration and applications at Novellus Systems (San Jose). "However, for the 32 nm technology node, it's 45 nm that presents a significant metrology challenge." Havemann added that with some films, roughness comes into play. "To lower resistivity, you may need larger grain size, which may require a rougher film, complicating, in turn, particle differentiation from the film's background texture." (Fig. 3)

3. On rough films, background noise complicates particle detection. This poses a significant challenge for measurement of <50 nm particles at the 32 nm technology node and beyond. (Source: Novellus)

In terms of film thickness, most optical systems do a good job, but the answer is as good as the data coming in and the model used. Materials and material stacks are becoming optically complicated. "The index of refraction for oxide has a smooth inverse monotonic curve as a function of wavelength," Havemann said. "But with some of the new materials' stacks, you start getting significant deviations from that curve, and each material's optical properties must be modeled to get an accurate thickness calculation. Changing deposition temperature or any process parameters can lead to an erroneous answer, so you must nail down the material properties and hold them constant to get repeatable film thickness results."

Brad Bartilson, metrology product manager at Rudolph Technologies (Flanders, N.J.), views both sides of the issue. "With copper, we've had a thickness reduction of both seed layer and electroplated copper, which brings both problems and opportunities. My focus is on opaque — primarily metal — systems measurements using picosecond acoustic technology. We have single- and dual-wavelength systems. We brought in a dual-wavelength system to address copper, and expanded its capabilities to include some dielectric measurements within the low-k area." Early on, Rudolph focused its 400 nm system on copper. By combining that with 800 nm capabilities, it enabled measurements that previously were difficult with copper and copper barrier/seed.

For the upcoming 45 and 32 nm nodes, Rudolph sees anopportunity to increase its selectivity for copper by tuning the wavelength. While improving the signal and associated accuracy and repeatability, this could offer further improvement in throughput — vital to addressing cost of ownership reductions requested by CMP managers.

"Further out, we're considering mixed technologies, new optics to allow them to work together," Bartilson said. "Definitely metal films since, when thinner, they become transparent. We still haven't resolved whether it'll have to be X-ray, ellipsometry or a stretching of current acoustic technology. Some fundamental limits are being approached, but we don't envision that every layer will have only ultrathin films."

Reflectometry waves

To control the RC time constant of interconnects, line separation must be controlled. "This can be done by monitoring ILD thickness. For process control, it's preferable to monitor film thickness in the pattern area," said Arun Aiyer, chief scientist for emerging technologies and market development at Verity Instruments Inc. (Carrollton, Texas). Traditional small-spot reflectometry must be done on metrology sites in the scribe. Large-spot normal incidence Fourier transform spectral reflectometry can be integrated into the process tool to look at pattern areas and parse out ILD thickness. Contribution from an underlayer could be a problem if it has a comparable refractive index; however, if a low-k film is deposited over the etch-stop layer, separating the thicknesses may become easier.

As the industry moves to 45 and 32 nm nodes, vias and trenches (especially those in the lower metal levels) become smaller. As the pattern shrinks, the reflectometer's signal from these features is reduced. Measuring etch depth using reflectometry becomes harder because of the lower-signal-to-noise ratio and the resulting poorer fringe contrast. The signal-to-noise ratio may be enhanced using short-wavelength reflectometry, but implementing this for in situ process monitoring is difficult and expensive, and will require advanced modeling.

It is certain that to continue on our progression path, robust modeling and a high degree of shape and interfacial control are needed. We are reaching the end of a technical era. Some metrology tools are reaching the end of their run, and we will doubtless see a period of disruption in the field until new technologies take over.