Characterization Challenges of New Materials

Qualifying innovative materials for next-generation devices often entails more testing and, in some cases, the development of new techniques and technologies.

Ruth DeJule, Contributing Editor -- Semiconductor International, 6/1/2007

Progress in the semiconductor industry means faster devices with greater functionality. Getting there begins with selecting the right materials to ultimately accommodate smaller linewidths and tighter constraints. Each new technology node brings demands for novel, and increasingly complex, materials, and with it the need to characterize them (Fig. 1 ).

New materials typically undergo basic testing of thickness, thickness uniformity, impurities, composition, uniformity and thermal stability, although additional characterization has become more frequent. This is reflected in the characterization workload in recent years, which has greatly increased as R&D and production interests and demands have grown, noted Hugh Gotts, director of R&D at Balazs Analytical Services (Dallas).

A series of tests performed on new low-k dielectric materials typically calls for further characterization beyond the basics, such as determining mechanical integrity in the form of tensile elasticity (Young's modulus) and hardness. The technique most commonly used, nanoindentation, applies a small, hard point against the material to create an indentation, while simultaneously recording the applied force. From the curves generated, the mechanical properties of the film can be determined. This widely used technique works well on materials at least a 0.5 µm thick, which poses a problem for the 32 nm node, where thicknesses are expected to be on the order of 100 nm. Instead of measuring the properties of thin film, those of the underlying substrate will likely influence the measurement.

1. Flowable coating chemistries are characterized on a spin-coating system at the Dow Corning lab in Midland, Mich. (Source: Dow Corning)

No other method is currently available for measuring very thin films, although other approaches are being studied. According to Rudi Cartuyvels, department director of interconnect technology and technology options at IMEC (Leuven, Belgium), a possible alternative is surface acoustic waves, a technology originally developed to measure the density of metal films. The technique applies a weak pulse to the film with a laser, producing a wave, which propagates along the surface of the film. A second laser records the propagation speed of that signal, and material properties, such as Young's modulus and hardness, are extracted.

Once mechanical soundness is established, the value of the dielectric constant (k) is determined by looking at its chemical bonds. This can be done with Fourier transform infrared spectroscopy (FTIR), where the peaks in the spectrum correlate to the types of chemical bonds in the film. When characterizing low-k films, carbon, carbon methyl groups (CH3) and hydroxy groups (OH) are the bonds of interest. OH groups are considered polar bonds that, when added to the material, can increase the low-k value. In contrast, when non-polar bonds such as carbon and CH3 are added, the relative amount of polar bonds in the structure is reduced and the k value lowered.

Nuclear magnetic resonance (NMR) can provide further bonding information by quantifying the various types of chemical bonds in the film. The technique itself is a difficult one, requiring special sample preparation — removing the thin film from the wafer and pulverizing it. The number of bonds in the film can then be determined with NMR.

Ideally, materials with good mechanical properties with a low-k value are sought, although in reality, the lowest k value will have the poorest mechanical characteristic. Current films have k values in the 2.0-2.5 range with fairly good mechanical properties, a Young's modulus of 10 GPa. But when the k value is reduced to 2.1 or 2.2, Young's modulus drops to below 5 GPa. According to Cartuyvels, to obtain an optimal balance between mechanical properties and k value, porosity may be introduced into the films.

Low-k films can become difficult to integrate if the number and size of pores get too large. These parameters can be measured by an ellipsometric method pioneered by IMEC (Fig. 2 ). A wafer with a deposited low-k film is placed in a low-pressure chamber and toluene is leaked in. As the film absorbs the toluene vapor, the optical properties of the film change. An ellipsometer records the change and determines the pore size, size distribution and porosity of the film.

2. TEM can be used to evaluate continued improvement in research, like those seen in ionized PVD (i-PVD) resputtered barrier and i-PVD copper seed. (Source: IMEC)

Reliability tests

Having passed fundamental testing, reliability tests look at the ruggedness of the final material. This can be achieved with four-point bending or a dual cantilever beam technique. Both tests measure the distance a crack can propagate in a dielectric film. Four-point bending begins by creating a dent in a multi-layered stack and pushing the surface film in such a way so that a crack forms and propagates through the layers. At the weakest interface, the crack will abruptly change directions and propagate along that interface. This occurrence determines whether the failure is of a cohesive or adhesive nature. If, however, the film becomes very weak, the crack will then begin to propagate through the bulk of the film.

Dual cantilever beam can also characterize failures and test the effect of different environments on the material. In the presence of moisture, for example, if the crack propagation accelerates, it is likely that moisture is detrimental to the material. In this manner, the number of films and types of stacks can be screened to evaluate compatibility.

It is well known that device performance can be improved when mechanical strain is generated in the channel regions. The strain has the effect of increasing carrier mobility, leading to a reduction in device delays and higher device frequencies. Measurements can be global or local, with local being of greater value to chipmakers. For non-patterned films, strain can be added by growing strained silicon on silicon germanium (SiGe) layers and characterized with Raman spectroscopy, X-ray diffraction (XRD) or ellipsometry. However, local strain measurements require spatial resolutions <50 nm. Thus far, Raman spectroscopy has the greatest potential.

Raman is a scattering technique that uses a coherent source — typically a laser. When measuring strain in thin films, shorter ultraviolet (UV) wavelengths are generally used to limit penetration into shallow ~50 µm layers. When scattering off a Si-Si or Si-Ge bond, a very sharp peak is generated. Strain is determined when the position of that peak shifts to longer or shorter wavelengths as strain is induced into the thin layer.

Researchers at AMD Saxony (Dresden, Germany) used Raman spectroscopy and a continuum-mechanical model to estimate the distribution of strain in channel structures. They derived a simple relationship among Raman shifts between two photon models, composition and stress, adding further credibility to Raman as a tool for measuring strain.¹

XRD is generally used for measuring crystallographic properties. However, Bruker AXS (Madison, Wis.) has developed equipment that measures local strain with resolutions of 10-3 Å, according to Uwe Preckwinkel, product manager at Bruker AXS. XRD measures the distance between the lattice plains in a crystal based on Bragg's law. Strain is determined from the difference between the lattice spacing as measured by XRD and that of the same material unstrained. No modeling or complex calculations are required. The availability of new brilliant X-ray sources has made the analysis of areas down to 10-100 µm possible, increasing the capability of XRD to measure local strain, claimed Preckwinkel. In-house tests performed with their high-resolution system successfully demonstrated that strain in thin films — 60 nm silicon on SiGe layers — can be detected and separated from signals coming from the substrate, noted Preckwinkel.

Once strain is induced, stability and the presence of strain gradients in silicon and Si/SiGe layers can be determined. Strain gradients can be mapped by moving across the surface of the film and measuring shifts in peak position with Raman (Fig. 3 ) or high-resolution XRD.

3. Raman spectroscopy’s excellent lateral resolution can effectively determine stress and strain in 300 mm wafers, and is the forerunner for characterizing local strain.

Similarly, thermal stability can be tested by putting the film through a series of thermal cycles and measuring changes in peak intensities and linewidths. Gotts noted that if recrystallization occurs, the peak may narrow. And if there is a decrease in peak intensity or area, this may indicate a loss in some of the thin film or that the stoichiometry of the film is changing.

Stability studies of strained Si/SiGe layers measured under various thermal annealing conditions demonstrated the durability of the induced strain and revealed thermal problems with germanium.² High-resolution XRD was used to determine the strain in the layers prior to thermal treatment and Raman spectroscopy, and secondary ion mass spectroscopy (SIMS) measured the impact of thermal processing on strain relaxation and germanium diffusion. Test results showed no indication of strain relaxation in silicon with temperatures up to 1000°C for a duration of 5 min. Germanium, however, did not fair as well. Raman and SIMS both indicated diffusion into the strained silicon layer.

The need for greater sensitivity and resolution is making some techniques obsolete, forcing equipment makers to redesign current tools or develop new ones. Electron dispersive spectroscopy (EDS) is one example. With generated X-ray spectra and density information, EDS is used to identify elements present on a wafer surface or in a layered stack. Used for over a decade in defect review tools, EDS is reaching its resolution limit as critical particles sizes get smaller.

The resolution of current EDS systems cannot sufficiently resolve the silicon peak from adjacent tungsten and tantalum peaks, each less than 30 eV apart. One solution used by EDAX Inc. (Paoli, Pa.) replaces the detector used in its current EDS equipment with a parallel beam X-ray spectrometer. Resolution is improved 8× to 6-8 eV, stated Del Redfern, product marketing manager at EDAX.

The company further configured its system to overcome previous time-consuming positioning problems with a software-controlled positioner and, in essence, created a “new” product for the semiconductor market, a user-friendly, fast wavelength dispersive spectrometer (WDS).

EDAX is now in the process of extending WDS to monitor layer thickness in the micron and nanometer ranges. According to Redfern, “Results from our internal studies look promising. We should be able to determine, for example, the composition of particles and thickness of layered materials, such as tantalum compounds on a silicon substrate, with a single piece of equipment.”

Material suppliers have long been faced with an all-or-nothing proposition. No sale occurs if the proposed material is not chosen, time and research notwithstanding. Developing new materials with lower and higher k values and materials with controllable stress have increasingly required that suppliers become experts in virtually all facets of the material, including the use of the material for its intended application, said Phil Dembowski, global marketing manager of device fabrication materials at Dow Corning (Midland, Mich.). Fortunately, the environment is changing. Sharing the cost and burden of technology development is growing, driven in part by challenging requirements and constraints of innovative materials.

The materials and services provided by material suppliers can vary widely, depending, in part, on the complexity of the material under development. Some suppliers, such as Dow Corning, have in place the capability of fully characterizing thin films, including chemical, electrical and mechanical properties. Therefore, when a low-k material with a particular dielectric constant and properties is needed, they can give them the necessary chemicals, a recipe for coating the material onto the wafer, and curing it. They supply the material and process. However, developing strained layers is not as straightforward. Process driven, it depends on a right deposition recipe, generally requiring close collaboration with equipment manufacturers.

It is difficult for a single materials supplier to have all of the latest deposition equipment necessary to support new technology development, especially in the chemical vapor deposition (CVD) and atomic layer deposition (ALD) areas used for strained silicon. To be effective, equipment and materials suppliers need to collaborate to develop new technologies. “We found that using a collaboration model helps to minimize cost while also improving the probability of success. It allows everyone to focus on what they are good at,” Dembowski said.

The value of a collaborative environment is most notable in research centers.

The recently announced joint effort between Micron Technology (Boise, Idaho) and the University of Washington (Seattle) is one example. The newly formed Micron Laboratory Combination Materials Exploration is intended to make the material screening process more efficient and cost-effective. The center combines the expertise within the University of Washington and aims to work with other institutions, ultimately building closer relationships within the semiconductor community.

Joint efforts are not new to the industry. There is just more need for them. But whatever the course, the industry will find a way to greater efficiency and economy without compromising technical advancement.

New materials, such as high-k, strained silicon and ultralow-k, have allowed continued scaling to smaller dimensions, and characterization tools are evolving to meet the demands. According to Vandana V. Mukherjee, IBM's project manager for technology analysis (East Fishkill, N.Y.), as fabs lay the groundwork for the 45 nm technology node, characterization is at the forefront. At IBM, this means characterizing eSiGe, porous SiCOH, silicides and undoped silicate glass.

Cartuyvels suggests that there are greater trials ahead. As scaling decreases the amount of bulk in the film relative to the interface, surface engineering becomes ever more dominant in the whole integration process, and characterization will see tougher challenges. With material properties changing at the interface, it is very difficult to characterize.

Technical issues aside, Mukherjee contends that extending scaling by novel integration schemes and the introduction of new materials may ultimately have monetary advantages, that is, if it means the continued use of existing tools in the fabs.