OEMing metrology

Alexander E. Braun, Associate Editor -- Semiconductor International, 6/1/1999

As architectures shrink further, films get thinner. However, the same requirements still hold for uniform thickness, smooth and flat surface, uniform composition, no stress and integrity. This imposes thin-film metrology demands that are not easily met, particularly as some fundamental, physical barriers loom in the near horizon.

"We're at an inflection point with gate material and control issues," stated Gary Bultman, vice president and general manager of KLA-Tencor's (San Jose, Calif.) Film and Surface Technology Division. "We've lived with SiO₂ for 30 years and understand it. Companies like SiO₂ and are doing 0.13, 0.10 µm level research, pushing thicknesses down to 20 -- even 17 Å. But now we're talking about a film that's anything but homogeneous SiO₂."

John Heaton, president and CEO of Nanometrics (Sunnyvale, Calif.), views thinner film as a metrology driver. "Requirements have gone beyond refractive index and film thickness. Compositional analysis has become important. No longer is the ability to measure these thin films enough, but also computing their compositions --how much nitrogen, how much oxide, roughness, doping concentration."

Lee Smith, Therma-Wave's (Fremont, Calif.) vice president of strategic marketing, believes the next challenge for gate metrology will be how companies use reference standards to achieve matching and synchronization of oxide thickness across different fabs. "Accuracy's been somewhat of a phantom up to now," said Smith. "Company A may have a yield-optimized chip process running with a slightly different 'Angstrom' than company B. Within one organization this doesn't matter. But with mergers and alliances, eventually everybody needs to measure with the same ruler. That ruler must measure consistently with TEM and electrical measurements too. But the new focus for gate oxides will be overcoming organic and moisture films' metrology impact. This'll be key to achieving synchronization of gate oxide among multiple fabs spread throughout the world." (Fig. 1)

Fig. 1. Obtaining accurate optical thickness measurements is become increasingly more complicated as inhomogeneous films come into use. Not only is adsorption becoming a challenge, but standards' definition is a major concern because of manufacturers worldwide mergers and alliances requiring common standards. Equipment manufacturers are integrating metrology capabilities to meet these new demands. (Source: Therma-Wave Corp.)

Inhomogeneous films pose difficulties in obtaining accurate optical thickness measurements. Added to the problem is that outside the process system, a film adsorbs hydrocarbons and moisture from the cleanroom atmosphere -- approximately 1 Å over an hour period. For most applications, this is not a concern because 1 Å does not make that much difference when dealing with a 5,000 or 10,000 Å film. Besides, the adsorbed film desorbs when the wafer goes to the next process step and is heated.

"The problem," pointed out Bultman, "comes when you're trying to control a 20 Å gate to ±1.5 Å. We've systems today with 0.1 Å long-term stability -- metrology capability isn't the challenge. It's when you adsorb 1 Å of surface layer garbage and measured optically it shows 21 Å, but there is really only 20 Å in terms of electrical performance." Bultman believes surface adsorption will complicate process control procedures. "A 40 Å film is typical state of the art today. If in over half an hour it changes by 0.5 Å, that's a manageable problem. But when you're trying to control it to within 1 Å, a 0.5 Å change is a major problem." This requires different thinking about thin-film metrology tools. Since the additional surface film is electrically inactive, being a fairly porous film, an emerging approach is to do capacitance measurement of the gate dielectric. KLA's Quantox line is an example of this. Suitable for 20 Å, it has been used successfully at 12 Å.

Non-SiO₂ gate films are mostly terra incognita. As Bultman put it, "There's silicon nitride, maybe tantalum pentoxide. Depending on whether these are individual films or sandwiched into a structure, it'll be difficult to resolve. Nobody's figured out what the film's going to be, how it is going to be deposited and what process control considerations are going to be. If there were a 15 Å silicon nitride, with a 40 Å tantalum pentoxide and a 15 Å silicon nitride on top, and we were asked to determine individual layer thicknesses, we'd be somewhat unsure about how we'd do it."

New low-k films, such as spin-on and silks, are appearing. Bultman indicated that most of these films can be measured today. "The biggest headache we have is directly determining porosity. There are indirect ways of correlating to it through refractive index and the absorption coefficient, n and k. But users want a size distribution of the air bubbles in the film --how many and what size!"

Most low-k films are measurable today. Particularly with spectroscopic ellipsometry, where it is possible to obtain good information about how the full broadband wavelength of light interacts with the film, enabling accurate determination of thickness, refractive index and absorption coefficient.

"What I see going on both in the intermetal dielectric as well as the gate is a measurement world beyond thickness," stated Bultman. "Ten years ago we measured the passivation nitride's refractive index because it correlated to stress. With each generation, more of the film's attributes must be considered and controlled. We're now looking at refractive index and absorption on a majority of films because it correlates to something -- whether porosity, the concentration of two elements or something else. The n and k performance requirements are tightening up, pressuring precision and stability."

Bultman is optimistic about metrology's future. "The industry's mature, and we're doing a better job of antici pating issues. The next wave of problems will come with copper dual damascene. We're entering a period of very intense oxide etch. Trying to get a metal material we've never dealt with into the via and trench, and then polish it back -- it's tough."

Contamination is an initial worry with copper. How to keep it out of the gate and control it in production.

There already are systems that have proven to have good sensitivity to copper contamination in the silicon or the oxide. Beyond is process control, which separates into three issues: barrier/seed control, fill control and polish back control.

For barrier/seed control, manufacturers want to measure the film's continuity on the sidewall. Nobody can do it yet. As Bultman put it, "We're approaching the problem on the inspection side with our SEM inspection tools, investigating how to get down into these holes and see if the film is continuous or has a break." Bultman calls this "teething problem," and is convinced copper metrology will be ready when production begins.

OEMing metrology

Nanometrics' Heaton believes metrology is being challenged by the P/T ratios it is being asked to perform with. "People in 0.18 µm, especially now that they're getting into these processes, have new sets of stringent requirements on repeatability."

Nanometrics sees combined systems as metrology's future. The company introduced a platform with Fourier transform IR (FTIR) technology to resolve measurement problems beyond a combined ellipsometer and reflectometer's capabilities. "Epi silicon measurement is becoming a pronounced film particularly on 300 mm substrates, as is dielectric doping concentration as it relates to interconnects," said Heaton. Since material softness relates to doping concentration, in many low-k dielectrics CMP removal rates are affected. It is not electrical performance, but CMP removal rate that is linked to FTIR technology. (Fig. 2)

Fig. 2. No longer is one measurement enough when dealing with today's complex materials. Not only are several measurement capabilities being integrated on to single platforms, but work is underway to integrate metrology closer into the process cycle iteself. As requirements get more demanding, metrology will eventually be OEMed into processing platforms. (Source: Nanometrics)

Heaton sees metrology shifting quickly toward integrated measurements. "If you consider close-loop feedback in process tools, there's a compelling economic argument for moving metrology into the process tools. In situ's probably a decade away, but more integrated metrology is coming." Heaton bases this on the tack being taken by leading tool manufacturers. "Applied Materials has made a strong commitment to integrated metrology and is working with us," he said. "Eventually, the recipe will be selected, the tool will do its process, measure the thin film and ensure it's running at the correct thickness. Tighter specifications are going to lead to the OEMing of metrology."

Avoiding destruction

Michael Joffe and Michael Gostein, scientists at Philips Analytical (Boston, Mass.), view on-product measurement of metal films and traditional transparent ones as important for the company. "We're trying to do measurements with non-contact and non-destructive methods," stated Gostein. "As wafer surface area value increases, the fundamental changes of small spot size and on-product measurement capability become important."

The issue is understanding how spot size affects measurement and how to optimize it. "System resolution is linked to the capability to measure film-edge profiles. With a four-point probe, you don't do very well looking at properties near the edge. With a small-spot optical method, you get increased resolution and can profile edges and push dies farther out, buying real estate on the wafer," added Gostein. The future challenge is to make the spot size small enough to measure an individual wire.

Philips' Impulse system measures opaque films, and ellipsometry systems are used to measure transparent films, but Gostein sees these capabilities combining, without a division between opaque and transparent films. "Nothing's perfectly opaque or transparent. A film that's 50% transparent and 50% opaque could be better evaluated using both techniques. Today, neither fully covers all films," said Joffe. (Fig. 3)

Fig. 3. Systems like the Philips Impulse platform measure opaque films. The surface is illuminated by a laser pulse that launches an acoustic wave that propagates across the film, providing density information that can be extracted and processes to determine the film's properties. (Source: Philips Analytical)

There is more to thin films than thickness. "We zap the surface and get an electronic signal back," said Joffe. "How it's processed and what information is extracted is basically up to us. Currently, processing is geared toward thickness and density." Density is important because it gives an idea of the stoichiometry of a material that may be a non-stoichiometric compound, like tantalum nitride. "Density's important even for metal films, especially copper, because it depends on process conditions such as thermal anneal or electroplating," he added.

Joffe and Gostein view density as important, but do not stop there. "The trend's to expand the processing for density and parameters such as microroughness and perhaps properties such as thermal conductivity or doping level," said Joffe. If electrical conductivity is measured with a probe or other conventional technique and the stoichiometry changes, it is unclear if it is because of changes in the material's electrical resistivity or changes in grain size, for example. All these parameters are coupled.

It seems conceivable to develop an understanding of how these parameters affect the signal. The question is whether the full information is extracted or not. "With our optoacoustic measurement system," said Gostein, "in principle, we can determine thickness, density and stiffness. In practice, we must currently select two of these parameters to measure. In the future, it may be feasible to expand the number of parameters extracted simultaneously."

"The reality is that uncovering all desired metal film parameters by any metrology today is impossible. The challenge is to continually come up with practical solutions to real-life problems of process metrology," concluded Joffe.

Covering a wider spectrum

Rudolph Technologies' (Flanders, N.J.) newest contribution is its "laser spectroscopic ellipsometer." As George Collins, director of marketing, describes it, "The system covers the reflectometry domain of optical information from DUV, 190 nm, through near-IR, 980 nm -- it's full-spectrum reflectometry information." Rudolph also has laser ellipsometry that goes from deep blue, 458 nm, through near-IR laser at 905 nm, with simultaneous multiple angle of incidence data collection. So the system provides spectral information, reflectance information and the delta and psi from an ellipsometer obtained at the multiple angle of incidence.

"An area where we expect the multiple angle of incidence to make a significant contribution originates from the fact many low-k dielectrics being proposed for layer dielectric ILD are polymeric materials," explained Collins. "Because they need good high temperature performance, these materials have parylene-type structures or aliphatic-linked benzene rings. This tends to give them anisotropic optical properties, so the n and k measured in the film's plane differs from the one measured normal to the film. Optical properties can vary with deposition and curing methodologies -- optical process control needs optical properties acquired at multiple angles."

Collins indicated that with the focus on copper interconnects, sometimes people forget that many fabs will still be using aluminum for the next five to eight years. "In the excitement over the copper dual damascene processes, we also tend to forget that many of the initial steps in both copper and aluminum interconnect processes are the same. They both require silicide formation, the laying down and etching of an interlayer dielectric, and deposition of an ultrathin barrier and adhesion layer multiple times to build the multilayer metal film stack. Many new tools have been developed to measure the thick copper layers that'll be used in copper interconnect processes. However, some of these systems can't measure thin liners and silicide phase, and thus cannot support the entire process. Copper layers are only one third of the total."

TEM metrology

Colin Smith, president of SELA (Israel), believes that with the multi-process layers being used along with continuing shrinks, top down thickness measurements are almost limited to blanket layer test wafers. "Verifying metal/oxide thickness in specific product areas relies mostly on electron microscopy microcleave cross sections," pointed out Smith. "For high-resolution thin-film measurements, we recommend the automation of the initial microcleaving and TEM sample to expedite preparation. Then, we'd complete the final thinning with the FIB and verify the thin film thickness with the TEM."

SELA views sample preparation as a major film measurement obstacle. The time, cost and skill needed to prepare samples is considerable, requiring the use of expensive equipment such as the TEM and specialized technicians, especially for thin films less than 100 Å and site-specific areas with layer interface anomalies. "Copper will be a bigger challenge to process engineers controlling contamination and migration, than the ability of an analytical technician to microcleave, prepare the sample and measure the copper film," said Smith.

Characterizing and monitoring

Jean-Claude Fouéré, general manager at SOPRA (Acton, Mass.), sees spectroscopic ellipsometry as a solution to new material metrology. "Today, when you say 'thin gate oxide,' you're looking at more complex materials -- oxide, oxynitride, etc. They aren't just thinner, but more complex." As Fouéré pointed out, deposition processes were monitored in terms of thickness. "Now it's more than thickness, it's also material characterization."

"With thin oxide -- which may be thin oxide or thin oxynitride -- we're working on combining complementary techniques," said Fouéré. "For example, spectroscopic ellipsometry with grazing incident angle X-ray reflectometry or X-ray diffraction/ reflectometry. The X-ray measures thickness precisely and accurately. Once we've the data, we can do regression analysis on the ellipsometry data because we know the thickness. Then we can do the regression of the optical constant. This is a powerful way to characterize materials and may become common in the future."

SiGe is another example. There are several analytical techniques that can be used to characterize SiGe such as RPS, SIMS and X-ray. Unfortunately, none lends itself easily to manufacturing implementation. Some, like SIMS, are destructive, while others work with larger spot sizes such as X-ray diffraction.

Fouéré views spectroscopic ellipsometry as a solution to characterize and monitor the SiGe process, with the advantages of an automated, non-contact, non-destructive small-spot system. "When you work with a thin gate oxide, at about 20 Å, you want to characterize thickness within a fraction of an Angstrom," he stated. "You may measure a thickness of 25 ±0.05 Å. Even using TEM, resolution is maybe 1 Å, 0.5 Å, so what do you measure? As long as it's repeatable and the layer's electrical characteristics are the same, it doesn't matter -- precision, not accuracy is the key."

Combining technologies

Therma-Wave's Smith views technology combinations as a strategic issue for the future. "We pioneered combining technologies into one film tool with the Opti-Probe back in the early 1990s. That strategy led to the industry's present trend. We're now up to five separate technologies fully integrated into a single footprint in the fab. The end isn't in sight, although IP is surfacing as a major differentiator among tool suppliers." (Fig. 4)

Fig. 4. More and more different film metrology capabilities are being integrated on to a single platform. Requirements have evolved to widespread production measurement of four to six layers simultaneously. Measuring multilayer film stacks is these combined technologies' prime application. In this case, the Opti-Probe system incorporates spectroscopic ellipsometry and absolute ellipsometry. These are integrated with beam Profile ellipsometry, beam profile Reflectometry and a DUV spectrometer. (Source: Therma-Wave Corp.)

Film metrology has rapidly evolved from single or bilayer requirements to widespread production measurement of four to six layers simultaneously. Measuring multilayer film stacks is these combined technologies' prime application. "As six- or seven-layer stacks become common, measurement techniques must produce enough independent data to solve the equations robustly -- no single technique is capable of that task any longer. Measurements that combine several techniques -- in our case, beam profile reflectometry (reflectance versus angle) with spectrometry and ellipsometry -- are routinely performed today in fabs across the world."

However, ultrathin gate oxides are one key film requiring a different specialized approach. Only ellipsometry can precisely measure gate films with thicknesses migrating from the 20-30 Å range into the teens. "Ellipsometry's been optimized to give the best possible repeatability on this film due to its criticality to chip performance. Absolute ellipsometry measurements now give milli-Angstrom-range repeatability. If you put in a wafer of 15 Å or 20 Å and measure it 30 times, you get a one-sigma of 0.002 to 0.003 Å. This is over two orders of magnitude better than, say, five years ago," said Smith. "Achieving acceptable P/T ratios isn't a roadblock even for these very thin gates."

Looking ahead

Strong technical and economic drivers exist for combining multiple measurement techniques. New materials will need monitoring of parameters other than thickness and refractive index. Dopant concentrations in CVD films need monitoring to ensure proper (low) k values, and cure rates of spun-on low-k films and resists must be determined. Combining FTIR, spectroscopic ellipsometry and spectroscopic reflectometry, increasing information content by providing thickness, index of refraction, extinction coefficient, cure rate and dopant concentration, seems a design imperative. TEM and FIB metrology techniques will certainly continue playing an important part.

As process techniques evolve into more exotic materials -- wafers grow larger and architectures smaller and denser -- porting different metrology capabilities to single platforms will reduce overall costs and increase equipment effectiveness. It appears likely that most of us in the industry today will see metrology's transition from stand-alone status to full integration with other capital systems.