Metrology: EFTEM provides elemental mapping at nanometer resolution
03/01/2000
SPECIAL SECTION: EUROPEAN TECHNOLOGY
Ferdinand Hofer,* Research Institute for Electron Microscopy, Graz University of Technology, Graz, Austria
Othmar Leitner, Austria Mikro Systeme International AG, Unterpremstätten, Austria
Energy-filtering transmission electron microscopy is becoming an important nanoanalysis technique in materials science. Its main advantage comes from the two-dimensional chemical information or "elemental maps" at nanometer resolution extracted quickly from large sample areas. Work with a post-column imaging filter on a 200kV microscope illustrates the value of this technique for semiconductor device research.
Over the last decade, we have seen a large increase in the application of transmission electron microscopy (TEM) to semiconductor device analysis, driven by several factors:
- Higher density ICs and decreasing device dimensions have led to structures that are frequently in the nanometer range. Electron microscopes are ideally suited for imaging these structures, making them invaluable for routine constructional analysis. However, most work has been devoted to morphological investigations and analysis of crystallography and crystal defects.
- During the last few years, techniques have emerged that enable preparation of thin electron transparent samples for rapid and routine plan-view and cross-section analysis.
- Advanced development of analytical techniques, such as energy-dispersive x-ray spectrometry (EDXS) and electron energy-loss spectrometry (EELS), now provides chemical information from nanometer regions. A major advance is the recent availability of energy-filtering TEMs (EFTEM) that confers great advantages for semiconductor device analysis.
EFTEM is a very rapid method for both overview and nanoscale characterization of thin samples, applicable to most chemical elements and especially powerful for the light elements ranging from lithium to zinc [1]. In addition, EFTEM gives unique information that cannot be provided by any other method.
EFTEM configuration
Conventional TEM cannot use the information content of inelastically scattered electrons that have lost energy in the interaction with a specimen (see Fig. 1 insert). However, these electrons are very useful for micro- and nano-analytical purposes because they form an electron energy-loss spectrum (EELS) containing information on chemical composition, electronic properties, and chemical bonding. The rich information content of the EELS can be used only if an EEL-spectrometer is added to a TEM. The most versatile EEL-spectrometer, EFTEM [2], uses the spatially resolved nature of EELS. Within the last 2 years, 2 main approaches for energy filters have become available. One approach integrates an energy filter consisting of 4 magnetic prisms into the TEM's projector lens (so-called LEO OMEGA [3]). The other mounts an energy filter below the TEM column's camera chamber. This instrument, the Gatan imaging filter (GIF), is manufactured by Gatan Inc. and can be attached to practically any 100-400kV TEM [4].
For our work we used a Philips cm20 scanning TEM (TEM/STEM, 200kV, LaB6-cathode) equipped with a GIF (Fig. 1). When operated in imaging mode, this instrument is able to form EFTEM images and in spectroscopy mode EEL spectra can be acquired from small specimen regions. Images and spectra can be recorded with an integrated slow-scan CCD camera (YAG scintillator crystal, 1024 x 1024 pixel array). Acquisition times are typically 3-60 sec/image. In our work, we prepared specimens using standard cross-sectional preparation techniques with final argon ion beam milling at a low angle.
 |
Figure 1. Schematic of EFTEM equipped with a post-column imaging filter. The resulting EEL-spectrum shows an element-specific ionization edge. The positions of the energy-filtered images that are required for background subtraction to get the net element map are indicated as colored bars.
Any EELS feature can be the basis for imaging; for example, ionization edges caused by inner-shell excitation of specimen atoms by the high energy-electrons. Ionization edges occur at characteristic energy losses, enabling the identification of chemical elements. Consequently, energy-filtered images recorded at the ionization edge energy can be used to derive two-dimensional element distributions. Since the ionization edge sits on a non-element specific background, however, elemental mapping is used to remove the background contribution to image intensity for each pixel.
 |
Figure 2. TEM cross-section through a semiconductor device: a) TEM image, b) silicon map (Si K edge), c) titanium map (Ti L2, 3 edge), d) aluminum map (Al K edge), e) nitrogen map (N K edge), and f) oxygen map (O K edge), with g) a RGB image (red = Ti, green = N, blue = O).
We obtained elemental maps using two energy-filtered background images in front of the edge and one image at the ionization edge of the element of interest (this is depicted in the EELS schematic in Fig. 1). Then an extrapolated background image is subtracted from the ionization edge image, yielding a net image — an elemental map. Alternatively, so called jump-ratio images can be calculated by dividing the ionization edge image by a pre-edge image, giving elemental maps with minimum added noise, and which are less affected by diffraction effects arising in elemental mapping of crystalline materials [5].
Basic EFTEM application
EFTEMs are ideally suited for characterizing semiconductor devices, making them invaluable for routine constructional analysis. Figure 2 illustrates how the overview of structures provided by a standard TEM image (Fig. 2a) can be greatly enhanced with EFTEM elemental maps (Figs. 2b-2f), providing rapid defect identification.
For example, the elemental map in Fig. 2b reveals bright regions showing the silicon substrate and a thin polysilicon layer. The gray regions show silicon dioxide, silicon oxynitride, and silicon nitride layers; it is even possible to detect a small gradient in silicon concentration in the silicon dioxide layer that was introduced by doping with boron and phosphorus. In addition, a silicon-rich inclusion in the conductive layer (Al-Si-Cu alloy, see arrow in Fig. 2b) can be detected, which would have been overlooked by any other method.
 |
Figure 3. TEM cross-section of a titanium nitride layer: a) TEM image, b) aluminum map (Al K edge), c) titanium map (Ti L2, 3 edge), d) nitrogen map (N K edge), e) silicon map (Si K edge), and f) oxygen map (O K edge).
The titanium map (Fig. 2c) clearly shows the bright Ti-nitride layer and some spikes into the conductive layer (see arrows). The aluminum map (Fig. 2d) shows the conductive layer with silicon-rich defects and titanium-rich spikes visible as dark regions (see arrows). The nitrogen map (Fig. 2e) shows the thin titanium nitride layer, silicon oxynitride, and silicon nitride on top of the structure. Finally, the oxygen map reveals silicon oxide and the silicon oxynitride phases (Fig. 2f). Often an analysis stops at this point by combining all elemental distribution images into a collage (like Figs. 2a-2f). However, we consider it more important to show the spatial relationship between the elemental distribution maps.
Almost every image processing software package enables the combination of two or three images to form a color image by assigning each elemental map a red, green, or blue (RGB) color. While RGB-images are easily computed, they are limited to three elements. They often give rapid information on phase distribution and look beautiful (see Fig. 2g), but they are sometimes difficult to interpret because mixed colors occur.
Visualization of interdiffusion
As device dimensions decrease, the structure and composition of thin interfacial layers become increasingly important to device performance. Our work with a 90nm thick titanium nitride layer (Fig. 3a) and its interface, acquired at a higher magnification than the example above, illustrates that the ability to examine these interfaces in actual devices is extremely important and that EFTEM and EELS are uniquely suited to the task. The subsequent elemental maps show the Al-Si-Cu alloy conductive layer with highlights revealing defective regions with lower aluminum concentration (Al map, Fig. 3b), the titanium layer with bright spikes penetrating into the conductive layer (Ti map, Fig. 3c), and the thinner (70nm) titanium nitride layer (N map, Fig. 3d). A comparison of the silicon (Fig. 3e) and oxygen (Fig. 3f) maps shows that the oxygen rich zone extends further into the titanium layer, suggesting a titanium oxynitride layer in the interface region. This can be even more clearly seen in the RGB image (Fig. 4a), where we have superimposed the titanium map (red), the nitrogen map (green) and the oxygen map (blue) [6].
 |
Figure 4. a) Constructed RGB image of the titanium nitride layer in Fig. 3 and b) associated quantitative line profiles (red = Ti, green = N, blue = O).
A powerful feature of EFTEM is the possibility of quantification; EFTEM elemental maps can be quantified by calculating atomic ratio maps [6, 7]. We derived a quantitative line profile across the titanium nitride layer to reveal concentration profiles for Al, N, O, Si, and Ti (Fig. 4b). These line profiles clearly show the formation of an interface layer, consisting of Ti and Al, between Al and TiN. On the other hand, the interface toward the SiO2 also exhibits an intermediate layer containing mainly Ti and O. The composition of the SiO2 layer is very close to the nominal values (e.g., 33% Si when 66% O), whereas the concentrations within the titanium nitride layer are ~60% Ti when 40% N.
Understanding the technique
Application of TEM and EFTEM to semiconductor devices has not been, and may never be, as easy as that of scanning electron microscopy (SEM). However, we believe these techniques are not as difficult or unreachable as many think. Although very detailed knowledge is necessary for its application, we believe that the advantages of EFTEM are so striking that it will soon be an essential part of equipment in advanced research laboratories. To review, the advantages include the ability to record nano-resolution elemental distribution maps for lithium through uranium, to acquire many pixels in one exposure (which is an important factor when analyzing an extended area of the sample), and to convert elemental maps to concentration maps.
One must, however, be aware of the limits to spatial resolution and detection sensitivity. These strongly depend on optimizing imaging parameters and on sample thickness and energy-loss and ionization edge type. The limits can only be reached with very thin and uniformly thinned specimens. In addition, spatial resolution is limited by several other effects, such as lens aberrations of the microscope, residual chromatic aberrations due to the width of the energy window, spatial delocalization at low energy-losses, and by the point spread function of the CCD camera. Using a microscope with a LaB6 gun, elemental maps can be recorded with 1-5nm spatial resolution, depending on the element, its concentration, and specimen thickness [5]. Consequently, when using an EFTEM with a field emission gun, resolution below 1nm should be accessible.
Understanding achievable detection limits of EFTEM elemental mapping, which is largely determined by the noise level in recorded images, is also important. Improving detection limits inevitably means careful optimization of the signal-to-noise ratio because the relevant signals of the inner-shell edges are often relatively small compared to the background signal [8].
For silicon-based materials, the minimum detectable atomic fraction may be as low as 0.1at% in a 1nm sample region, but this largely depends on specimen thickness and the combination of elements occurring in the sample. This is far above the detection sensitivity necessary for mapping doping elements in silicon, for example. Indeed, we believe that EFTEM is the ideal technique to map relatively high local concentrations of elements distributed over large sample areas.
On the other hand, for measuring chemical information from small specimen regions, another approach may have spatial resolution and detection-limit advantages. For example, a scanning transmission electron microscope (STEM) equipped with a field emission source and an EEL spectrometer can record sequentially an elemental map pixel by pixel. Recently, this technique has been able to record EEL-spectra from single atomic columns [9]. However, the disadvantage for elemental mapping is that the process is relatively slow.
Conclusion
We have demonstrated that EFTEM elemental maps area very powerful method for characterizing the chemical composition of semiconductor devices in the form of thinned samples. Two-dimensional distributions of the elements present in a sample can be shown in a low magnification overview to reveal inhomogeneities and defects especially in and near, in our example work, a titanium nitride film. With higher magnification, we found interfacial layers on both sides of the titanium layer and determined quantitative profiles.
Acknowledgments
Additional authors include Peter Warbichler and Werner Grogger, Research Institute for Electron Microscopy, Graz University of Technology, Graz, Austria.
References
- R.F. Egerton, Electron Energy-Loss Spectron Energy-Loss Spectroscopy in the Electron Microscope, Plenum Press New York, 1996.
- L. Reimer, (Ed.), Energy-Filtering Transmission Electron Microscopy, Springer Verlag, Berlin-Heidelberg, 1995.
- W. Probst, et al., Adv. Mater. 5, 297-300, 1993.
- O.L. Krivanek, et al., Microsc. Microanal. Microstruct., 3, 187-199, 1992.
- F. Hofer, et al., Ultramicroscopy, 67, 83-103, 1997.
- W. Grogger, F. Hofer, G. Kothleitner, Micron, 29, 43-52, 1998.
- F. Hofer et al., Ultramicroscopy, 67, 83-103, 1997.
- G. Kothleitner, F. Hofer, Micron, 29, 349-357, 1998.
- N.D. Browning, et al., Micron, 28, 333-348, 1997.
Ferdinand Hofer received his PhD in technical chemistry at the Graz University of Technology, Austria. He is professor and section head of analytical electron microscopy at the Research Institute for Electron Microscopy, Graz University of Technology, Steyrergasse 17, Graz, Austria, A-8010; ph 0043/316-873-8346, fax 0043/316-811-596, e-mail [email protected].
Othmar Leitner received his PhD in technical physics at the Graz University of Technology. He is section head of process transfer at Austria Mikro Systeme International AG.