Failure Analysis

Adopting low-voltage STEM and automated sample prep to perform IC failure analysis

Bryan Tracy and Kirstin Alberi, Spansion; and Shams Tabrez, SELA USA

Shrinking semiconductor device sizes have pushed many routine measurement and analysis tasks beyond the resolution capability of scanning electron microscopy (SEM). Transmission electron microscopy (TEM) can provide higher resolution than SEM, but it is more expensive and more difficult to perform. Scanning transmission electron microscopy (STEM) is finding growing acceptance in semiconductor failure analysis laboratories as a viable alternative, providing image resolution approaching that of a TEM but with a cost and ease of use more like that of a SEM. However, analysts are reluctant to take on the burden of preparing the types of thin samples associated with STEM analysis. Automated sample preparation techniques developed for SEM and TEM address this concern.

This article compares the ability of both TEM and STEM to analyze typical semiconductor samples and describes an approach to performing automated sample preparation that makes the process easy, fast, and reliable.

SEM/TEM/STEM Alphabet Soup

A SEM such as that pictured in Figure 1a forms an image by scanning a finely focused beam of electrons over the sample surface and mapping, point by point, the intensity of various signals emitted by the sample as a result of interactions between the beam electrons and the sample atoms. If the spot formed by the beam on the sample surface drops below a certain size, SEM resolution is limited by the size of the region from which the mapped signal emanates. That region is any position where beam
electrons scatter when they enter the sample. The size of this interaction volume depends on many imaging conditions, foremost among which are the particular signal chosen and the energy of the incident electron beam (measured in kV). Most SEMs operate with beam energies of 20 to 30 kV. At those energies, the interaction volume for some signal types can be a cubic micrometer or more. While the use of low voltage when operating with beam energies on the order of 1 kV reduces the size of the interaction volume, low-energy electrons are more difficult to focus into a small beam, limiting resolution.

A TEM such as that shown in Figure 1b is analogous to a film projector as it projects a magnified film image onto a screen. A broad beam of electrons floods the imaged region of a thin sample, while magnetic lenses form a real image from transmitted electrons and project the image onto a fluorescent screen or other viewing device. Image contrast can derive from a number of different interactions between beam electrons and sample atoms, making image interpretation complex. Since the TEM sample is very thin—typically a tenth of a micron or less—the beam electrons have little opportunity to spread. TEM resolution is limited primarily by the performance of the lenses in the electron optical system. The higher beam energies of TEMs compared with SEMs—in the 200- to 300-kV range—yield better resolution.

The STEM (shown in Figure 1c) developed as a hybrid technique performed on modified TEMs or SEMs. As in a SEM, the beam focuses on a small spot that scans over the sample. The image is formed by mapping some signal intensity synchronously with the scan. As in TEM, image information is extracted from electrons that are transmitted through a thin sample. STEM has many of the advantages of SEM: TV rate imaging, very-low-magnification (wide-field) imaging, and easy electronic image rotation. However, it outperforms SEM in that it offers stronger materials contrast, longer sample lifetime in the beam, and lower effective contamination rates. Moreover, it does not require charging or special handling of conductive coatings. Because STEM samples are thin, beam spreading in the sample is minimal, and resolution is determined primarily by the spot size formed by the beam on the sample. While STEM’s resolution can approach that of a TEM, in practice some of its resolution is often traded for increased beam current and signal strength.

The term “low kV” as applied to STEM can be a little confusing. Low-voltage STEM as used here refers to the 20–30-kV regime, i.e., low relative to typical TEM beam energies. It is near the lower limit of energies that will provide sufficient transmission through the sample. STEM can also be performed at full TEM beam energies.

STEM can be performed on a dedicated instrument specifically designed to perform the technique, but more often it is added onto a SEM or TEM. In addition to being able to focus the beam on a spot and scan it over the sample, it must have an appropriate detector on the transmission side of the sample. STEM is becoming increasingly appealing because it can be adapted to a SEM platform and achieves high-resolution imaging comparable to that of a TEM. Since a SEM is designed to scan a focused beam over a sample, the only modification required of it to function as a STEM is the introduction of a transmitted-electron detector. Several such devices are commercially available.

The electrons transmitted at any point that are illuminated by the scanning beam may be divided into three regimes based on their interactions with the specimen. These regimes can be visualized as three coaxial cones of increasing angularity whose common vertices are at the point of transmission, as illustrated in Figure 2.

Figure 2: Diagram of the bright-field, annular dark-field, and high-angle annular dark-field functions of a STEM.

The central cone contains electrons that have passed largely unaffected through the sample. The intensity of the signal generated here is a function of the mass thickness of the sample. The thicker or more massive the sample, the more electrons are absorbed or scattered out of the cone and the weaker the signal becomes. When using this signal, which is called the bright-field signal, more-massive features are darker.

The intermediate cone contains electrons that have been scattered by interactions with electrons of the sample atoms, and, if the sample is crystalline, electrons that have been diffracted by its periodic structure. (In a conventional TEM, the pattern formed by the diffracted electrons is an important source of information about crystalline structure.) TEM images formed by these electrons are called dark-field images and are roughly the complement of the corresponding bright-field image. In other words, most electrons that are diffracted or scattered out of the bright-field signal end up in the dark field signal.

The third and widest cone contains electrons that have been scattered through high angles by interactions with the atomic nuclei of the sample (the same process that creates backscattered electrons in a SEM). The intensity of this signal is primarily a function of the atomic number of the sample. Images formed from it are known as Z-contrast images, and STEM detectors designed to capture this signal are called high-angle annular dark-field (HAADF) detectors.

Sample Preparation

Perhaps the most important reason that SEM has been the preferred technique in failure analysis labs is that it requires only minimal sample preparation. Especially since the appearance of variable-pressure and environmental SEMs, a sample probably can be imaged if it fits in the chamber. TEM and STEM, on the other hand, require small, extremely thin samples that can be very difficult to create. Whole careers have been built around TEM sample-preparation techniques. Failure analysts, for whom time is money, have been reluctant to take on the burden of using a TEM or STEM if a SEM could do the job.

In addition to the requirement that it be thin enough to transmit electrons, the typical failure analysis sample must be extracted from a particular wafer location (identified by an inspection tool) with sufficient accuracy to ensure that it includes the targeted feature or defect. Conventional preparation techniques consist of manually cleaving the wafer to a tiny sliver containing the target and then painstakingly polishing the sliver to remove material from both sides to reveal the target. This process of sizing and polishing the wafer is typically performed manually. However, manual processing can contaminate, damage, or even destroy a sample entirely.

Figure 3: Schematic diagrams of the MC600 automated sample-preparation process: (a) a coarsely cleaved sample, (b) microcleaving of coarse sample, (c) automatic scribing of the wafer edge, (d) shock wave induced from the opposite edge of the wafer piece before cleaving, and (e) final cross-sectioning of the target.

An automated sample-preparation process, presented schematically in Figure 3, avoids the risks associated with the manual procedure. First, a coarsely cleaved sample (Figure 3a) is microcleaved (Figure 3b) using an MC600 cross-sectioning system from SELA (Upper Yokneam, Israel; and Sunnyvale, CA). The purpose of the cleaving step is to achieve a wafer piece measuring several millimeters per side with the target feature located approximately 10 µm from one edge. Then the system automatically navigates to wafer defect coordinates provided by an inspection tool, after which the operator uses an optical microscope to refine the position of the desired cleave. Available software enables the operator to plan the cleaving process and yield multiple samples from a single wafer. Once the target has been pinpointed, the system automatically proceeds to scribe the wafer edge (Figure 3c) and initiates each cleave with a carefully controlled shock wave induced from the opposite edge of the wafer piece (Figure 3d). Finally, final cross-sectioning of the target is performed (Figure 3e). The MC600 is pictured in Figure 4.

Figure 4: The MC600 cross-sectioning tool.

After microcleaving, the sample is thinned in a SELA TEMpro tool, where it is automatically glued to a specially designed stub. As illustrated schematically in Figure 5a, the system uses a precision diamond saw to remove the bulk of the wafer piece and the unneeded portions of the stub. Glued to a TEM grid and gripped in a clamp (Figures 5b-h), the final sample is approximately 25 µm thick in the region of the target and ready for final processing. Although several techniques have been developed that substitute various forms of ion milling for polishing in the final stages of sample preparation, many semiconductor failure analysis labs use focused ion beam (FIB) systems to perform a final “polish.” FIB uses a finely focused beam of relatively heavy ions, much like a tiny sandblaster, to remove material with great accuracy and precision. However, although FIB systems offer very precise control over material removal, they are inherently slow and expensive.

Automated sample preparation is faster, easier, and more reliable than its manual equivalent. The process can prepare a FIB-ready sample in less than an hour with 90% reliability. Automated systems can be operated by technician-level personnel, freeing them for other tasks during much of
the process.

Examples

STEM, TEM, and SEM micrographs of a transistor are presented in Figures 6a–6c, respectively. The STEM image (at 30 kV) and the TEM image (at 200 kV) are of comparable quality. The individual components of the transistor can be distinguished in both, and the contrast differential between oxide and nitride is roughly equal. The resolution of the STEM image is not quite as good as that of the TEM image, but it could have been improved by reducing the spot size. However, doing so would have lowered the beam current, degrading the signal-to-noise ratio and increasing image acquisition time. Nevertheless, the STEM image has higher resolution and materials contrast than the SEM image. Additionally, it is not obscured by the gold coating required in SEM sample analysis to prevent charging artifacts.

Figure 6: (a) STEM, (b) TEM, and (c) SEM micrographs of a transistor. The STEM image (at 30 kV) and the TEM image (at 200 kV) are of comparable quality.

Shallow-trench-isolation structures are shown in the 30-kV STEM and 200-kV TEM micrographs in Figures 7a and 7b, respectively. The images show equal levels of detail and contrast. Figure 8a presents a high-voltage (300 kV) STEM image captured using a HAADF detector, while Figure 8b presents a low-voltage (30 kV) STEM dark-field image. The images show a planar view of the same thin aluminum film. The lower-voltage image provides better differentiation of the grain structure in the film than the higher-voltage one.

Figure 7: (a) a 30-kV STEM micrograph and (b) a 200-kV TEM micrograph of shallow-trench-isolation structures. The images show equal levels of detail and contrast.

Figure 8: (a) a high-voltage (300 kV) STEM image captured using a HAADF detector and (b) a low-voltage (30 kV) STEM dark-field image. The images show a planar view of the same thin aluminum film.

Low-voltage STEM bright-field and dark-field images of polymer-resist structures are illustrated in Figures 9a and 9b, respectively. Materials with low atomic weight show poor contrast at higher beam energies. Since the mean free path of an electron decreases with beam energy, lower beam energies provide better mass thickness contrast, eliminating the need to stain lighter materials.

Conclusion

STEM, together with automated sample preparation, promises to provide significant analytical cost savings by offering high throughput, relatively low capital expenditures, and personnel savings. Automation can cut average sample preparation time by a factor of two—from more than two hours to approximately one hour. Equally important, it may increase reliability by nearly the same amount. In addition, because STEM imaging is faster and easier than TEM imaging, the improvement in preparation throughput can be realized as an improvement in overall analytical throughput.

Figure 9: Comparison between (a) low-voltage STEM bright-field and (b) dark-field images of polymer-resist structures.

A new state-of-the-art TEM can easily cost more than $2 million and must be operated by top-level engineering personnel. A low-kV STEM, on the other hand, can be added to an existing field-emission SEM for $50,000 and can be operated by technician-level operators. Even when the additional cost of automated sample preparation is taken into consideration, the STEM capital expenditure is less than half that of TEM. While these analytical cost savings are significant in their own right, the true value of STEM analysis lies in the impact of rapid analytical turnaround times on semiconductor manufacturing operations. Fast recovery from process yield excursions reduces yield loss, and fast development cycles for new products and processes ultimately translate into greater market share and higher price margins.

Low-voltage STEM provides TEM-like resolution and material contrast at a fraction of the cost of TEM. Automated sample preparation eliminates much of the time and difficulty associated with conventional thin-sectioning techniques and improves the overall reliability of the sample preparation process. Together, low-voltage STEM and automated sample preparation offer a viable, cost-effective solution for high-resolution imaging with the capability to support semiconductor failure analysis well into the foreseeable future.

Bryan Tracy, PhD, has managed the materials characterization laboratory at Spansion (Sunnyvale, CA) since 1996. The group performs materials analysis to support the fab’s advanced process development for flash memory devices. In 1984 he entered the semiconductor industry and established TEM as a regular analytical technique at Intel’s facility in Santa Clara, CA. In 1991, he joined AMD as section manager in charge of SEM, TEM, and FIB instrumentation. Tracy holds 10 U.S. patents and has published more than 25 technical papers. His interests focus on the use of electron microscopy to characterize semiconductor materials. In 1984 he received a PhD in material engineering from Rensselaer Polytechnic Institute in Troy, NY. (Tracy can be reached at 408/749-4819 or bryan.tracy@spansion.com.)

Kirstin Alberi joined Spansion as a summer intern in 2003. She worked with the materials characterization lab to investigate the application of electron microscopy techniques in device failure analysis. She is a graduate student in the materials science and engineering department at the University of California, Berkeley, where she is researching thin-film processing and nanostructure fabrication. In 2003 she received a BS in materials science and engineering from the Massachusetts Institute of Technology in Cambridge.

Shams Tabrez is national sales and marketing manager for SELA USA (Sunnyvale, CA). With more than 15 years of experience in semiconductor device manufacturing and capital equipment, he has worked at Lam Research, National Semiconductor, and other companies. Tabrez has published nearly 20 technical and commercial papers and holds a U.S. patent. He has a BS in chemistry from the University of Manchester (UK) and a master’s degree in chemical engineering and economics from the Imperial College, University of London. He has also performed executive MBA work at the University of Chicago Graduate School of Business. (Tabrez can be reached at 408/736-3700 or shams@sela.com.)