Defect/Yield Analysis and Metrology

Using an in-line defect-analysis tool to provide real-time root-cause knowledge of yield loss

Stéphanie Blanc-Coquand, Benoit Hinschberger, and Eric Rouchouze, STMicroelectronics; Emmanuel Sicurani, CEA-LETI; and Janet Teshima, Marc Castagna, Matt Weschler, Larry Dworkin, and Didier Renard, FEI

The direct benefits of in-line defect analysis have never been in question: Faster recovery from yield excursions in high-volume production and faster ramp-ups for new processes and products contribute directly to improved yield and profitability. However, the net benefit of shortening the sampling-to-results loop by placing analytical equipment in-line has been questioned because the analysis process has been perceived as slow, difficult, and unreliable. The latest generation of automated defect-analysis tools may help change that perception. Based on the use of focused ion-beam (FIB) and scanning electron microscopy (SEM) technology to cross-section and image buried defects, surface defects, and 3-D structures, the new tools offer fast, easy, clean, and reliable analysis in the fab environment. Moving the process in-line can make real-time, 3-D defect analysis a practical reality.

The new systems could not have appeared at a better time. Today's process engineers must determine the root cause of a vast array of defects, many of which are unfamiliar. To meet the requirements of advanced technology nodes, semiconductor operations must continually incorporate new processes, new materials, smaller features, larger wafers, and complex 3-D structures—each of which creates opportunities for new defectivity modes. Many of these new defect types are difficult or impossible to characterize using traditional top-down inspection and review techniques, which focus on surface irregularities. The new defects are either buried by subsequent process layers or intrinsically 3-D, thus requiring cross-sectioning for complete characterization.

Conventional Defect Detection and Analysis

Traditionally, defect inspection, review and classification, and root-cause analysis have been regarded as distinct functions organized in a hierarchy of decreasing volume and increasing information. The initial phase offers high throughput but little information, while the final phase offers low throughput but detailed analytical information. Inspection is usually performed using optical imaging systems, although the use of electron-beam inspection tools has increased as feature sizes have decreased and more buried defects have appeared with the integration of copper processes. While electrical testing is also used to detect defects, it is typically performed only at the end of the fabrication process. Review and classification are usually accomplished with an E-beam system in advanced fabs because the structures reviewed demand the high resolution such tools provide. Root-cause analysis requires both cross-sectioning and imaging capabilities and is usually performed in a laboratory setting using a variety of failure-analysis techniques.

Because of its relatively high speed, optical inspection is, and will likely remain, the primary technique for detecting defects. Typical optical systems accumulate huge quantities of data and provide both statistical analysis and preliminary defect classifications. They are fast enough to support real-time corrective and loss containment activities, but too slow and too expensive to permit complete inspection at every process step. Thus, when they do find a defect, it is likely to have occurred one or more process steps prior to its detection. In addition, optical inspection is predominantly focused on surface irregularities and has limited success on subsurface defects, such as underetched vias and buried copper voids.

Optical inspection suffers another, less obvious drawback: strictly speaking, it detects not defects but anomalies—that is, differences from some acceptable reference. Many anomalies are functionally benign. In the best case, such nuisance defects consume analytical bandwidth that can be better applied elsewhere, and in the worst case, they may actually interfere with the discovery of underlying killer defects.

Detecting defects electrically at the end of the fabrication process has the advantage of identifying only those defects that actually affect the performance of the device. Once detected, these functional defects can be traced back to a specific process step, and further analysis can provide an understanding of the failure mechanism. One disadvantage of electrical detection is that it takes place well after the physical defect occurs. Hence, it cannot support real-time corrective action. It also does not provide a direct link to optical defectivity data. More importantly, the delay in achieving analytical results, which may be several weeks, puts a substantial amount of in-process product at risk.

Because electrical defects are always buried, there has long been a need for cross-sectioning to expose them for analysis. Until recently, cross-sectional analysis was incompatible with the fab environment and was performed off-line in an analytical laboratory, resulting in wafer scrap. It involved manual wafer cleaving and sample polishing, laborious and dirty procedures that could take several days. In addition to being time-consuming and tedious, it was difficult to cross-section a specific defect, and it was not unusual to destroy critical information in the very attempt to reveal it.

Laboratory-based analytical tools such as the DualBeam system from FEI (Hillsboro, OR) solved many of the problems of manual cross-sectioning by using a finely focused beam of relatively massive ions to cut a cross section and a beam of low-mass electrons to image the revealed surface nondestructively. When used in conjunction with an x-ray spectrometer, the E-beam technology can also provide a spatially resolved compositional analysis. In addition to being faster than manual polishing, a dual-beam system's precise navigational capabilities permit the analyst to reliably locate and cross-section specific defects.

Although they are a significant improvement over manual techniques, such lab-based systems are incompatible with the fab environment. They are not designed to meet in-line cleanliness requirements and typically require a highly trained, dedicated operator. Owing to their off-line location, they cannot provide the immediate feedback required for real-time process control. Additionally, removing wafers from the manufacturing line for basic defect analysis is becoming cost-prohibitive as more fabs move to 300-mm wafers.

Combining E-Beam Inspection with Dual-Beam Analysis

A new approach to defect detection and analysis combines E-beam inspection with in-line dual-beam analysis to address the shortcomings of all three conventional techniques: (1) optical inspection, which has limited ability to detect subsurface killer defects, (2) end-of-line electrical detection, which offers specificity but introduces long delays in critical feedback, and (3) laboratory-based cross-sectioning, which is too slow to support real-time process control.

Figure 1: An E-beam voltage-contrast image showing a floating metal line. The right-hand line of the circled pair is dark, rather than bright as in all neighboring pairs.

E-beam inspection is a rapidly growing technique that offers higher resolution, but lower throughput, than optical inspection. One of the primary drivers for the growth of the technology has been the need for better spatial resolution as feature sizes and critical dimensions have decreased. Using voltage-contrast techniques, E-beam inspection systems can provide access to electrical information before the completion of the fabrication process. Voltage contrast enables differentiation of open, or short, circuits by studying the signature of defects' charge-dissipation characteristics. Grounded elements show different contrast levels than do floating elements. Using voltage-contrast inspection techniques, the electrical connectivity of each line and contact can be evaluated as it is created. By comparing voltage-contrast images to reference images, E-beam inspection systems can quickly identify electrical anomalies, such as that presented in Figure 1.

Once an electrical defect is detected, the affected area can be cross-sectioned in-line for root-cause analysis, as shown in Figure 2. In the FEI DualBeam 300 defect analyzer used in the applications described in this article, the electron beam and the ion beam intersect at the sample surface. The system typically operates with the ion beam normal to the surface and the electron beam at an angle. Thus, the electron beam looks at the face of the cross section created by the ion beam. Switching between cutting and imaging requires only a few seconds and does not require repositioning of the sample. The system accepts full wafers and can navigate automatically to defects detected by other tools. The ion beam can also be used for imaging and can produce voltage-contrast effects, making it easy to recognize electrical defects detected previously by an E-beam inspection tool.

Figure 2: Micrographic image of the cross section through the open via. This area is responsible for the defect in Figure 1.

The combination of automatic defect redetection (ADR) and automatic defect cross-sectioning (ADX) of voltage-contrast defects creates a fully automated process that requires no operator assistance. ADX also can be used to cross-section and image defects at multiple sites across the wafer, enabling defectivity engineers to collect enough data to build a statistically representative model of the defect population without increasing the number of inspection steps.

In-Line Defect Analysis

The transition from laboratory instrument to in-line detection and analysis tool required that the dual-beam system's developers meet several critical goals.

Cleanroom Compatibility. For in-line use, a system must be compatible with the cleanroom environment of the fab. In the case of this tool, this requirement applies to the imaging and milling processes as well as to the overall mechanics of the tool and its enclosure. Related to cleanliness is the requirement in most advanced fabs that automated wafer handling involving the use of front-opening unified pods or standard mechanical interfaces be implemented.

Throughput. The value of an in-line analytical tool is a direct function of its ability to return results quickly and repeatedly. A number of considerations must be addressed to achieve high throughput. For example, using this tool, switching time between ion milling and E-beam imaging had to be minimized. The coincident-beam configuration accomplished this by eliminating the need to move, reposition, and reregister the sample when changing from mode to mode.

Another throughput consideration was the design of the tool's gas-assisted milling feature. The material removal rate of an ion milling operation can be greatly improved by introducing a small quantity of an appropriately reactive gas into the vicinity of the ion beam. Different gases work better for different materials. Gases may also be used to deposit insulating or conducting materials as part of the analysis procedure. For example, a protective material such as tungsten may be deposited to preserve the integrity of delicate surface structures or very small foreign particles during the milling operation. A complete analysis procedure may involve the use of several gases at different times. Typically, each gas is introduced through a small, needlelike apparatus in the sample chamber. The in-line use of the tool's multiple retractable needles, each dedicated to a single gas, improves analytical throughput by eliminating the time required to purge residual gas from a shared manifold design.

Ease of Operation. Process engineers in the fab must constantly juggle multiple priorities. Moreover, their access to samples and tools is constrained by cleanroom protocols and apparatuses. Ease of use and, if possible, unattended operation are thus highly desirable features in an in-line tool. To achieve these goals, the developers of the analysis tool included a number of automated capabilities. One such capability, automated defect redetection, uses die-to-die comparison and sophisticated algorithms to determine the precise location of a defect after navigating to a rough position based on coordinates provided by a previous inspection tool. Another, automated defect cross-sectioning, allows unattended cross-sectioning and imaging of multiple defects located anywhere on the wafer, and an automated slice-and-view feature permits the acquisition of a sequence of images progressing through the target, thereby creating a virtual 3-D reconstruction of the defect and its associated structures.

Wafer Integrity. One of the primary concerns raised when the dual-beam system was first proposed was the possibility of contaminating or damaging the entire wafer. This included the potential for diffusion of the gallium ions that compose the ion beam, redeposition of material sputtered during the milling process, frontside and backside contamination resulting from handling, transfer of contamination to adjacent wafers during subsequent thermal processing, and the effects of topography introduced by the analysis tool on subsequent processing steps. Extensive testing has shown that appropriate operating procedures can successfully mitigate all of these effects. Damage is typically confined to the analyzed die, which allows the rest of the sampled wafer to continue in the manufacturing process. In contrast, laboratory-based analysis requires that an entire wafer be removed from the fab, never to return to the production line. The in-line tool's potential reduction of scrap costs has thus become an important consideration in its adoption.

Figure 3 demonstrates the effects of in-line analysis on subsequent process layers. The initial cross section (Figure 3a) was made through a defect in level 1 copper. This location was then inspected at via etch 1 to observe the effects of the cross section on the interlayer dielectric and via etch (Figure 3b). Finally, the site was inspected again at copper 2 CMP to verify the planarity of the copper films (Figure 3c). No planarity issues were observed, verifying that the impact of the cross-sectional topography was limited vertically to a few process layers and laterally to the close vicinity of the sample.

In-Line Analysis Applications

The STMicroelectronics facility in Crolles, France, a mixed manufacturing environment that includes both volume production and R&D, has found the in-line analytical tool helpful in bringing new processes on-line quickly and maintaining mature processes at optimal yield. In the examples described below, the system was used to redetect and analyze defects that had been identified by other techniques.

Figure 4: Micrograph showing a particle embedded in an aluminum layer, which was revealed by an E-beam defect review tool.

Optically Detected Defects in Aluminum. In one instance, the fab had detected defects in the aluminum interconnect layer during optical inspection of a wafer. Subsequent review and classification of the site using an E-beam defect review tool revealed large particles embedded in the aluminum, as shown in Figure 4. Yield loss clearly correlated with the observed defectivity, but it was unclear from the top-down data alone at what point during the aluminum deposition process the particles had been introduced. This determination was the specific aim of the cross-sectioning analysis.

The defects were relocated by the in-line tool using the coordinated information from the defect file generated during the initial inspection. The defect was then cross-sectioned using the tool's multiple sampling capability, as illustrated in Figure 5. Slicing was stopped approximately halfway through the particle in order to preserve enough of it for compositional analysis. Within the image sequence, a single slice showed that the particle contacted the substrate layer with no intervening aluminum (Figure 5b). In contrast, Figures 5a and 5c show aluminum between the particle and the substrate. This confirmed that the particle must have appeared before the aluminum deposition process began, and probably before the wafer was introduced into the deposition chamber. This example demonstrates the value of producing multiple cross-sectional images, since a single, randomly located cross section through a particle would be unlikely to reveal the critical point of contact between the particle and the prealuminum substrate.

Once the contaminating chamber had been identified, chemical analysis of the particle quickly led to the identification of the defective process-tool parts. Without in-line analysis, this problem would have required an extensive partitioning experiment involving several inspection setups and defect reviews at each stage of the deposition process. Such an experiment typically takes days to perform, while the in-line tool provided a statistically significant number of cross sections in a few hours. The problem was fixed rapidly, and its impact was confined to a limited number of wafers.

E-Beam-Detected Defects in Metal. In two other cases, defects in tungsten and copper interconnect modules were initially detected using an E-beam inspection tool optimized to find voltage-contrast defects. The coordinates of the defects were then transferred from the E-beam tool to the in-line system, which automatically navigated to the designated locations, redetected the defects, and performed cross-sectioning.

Blocked or incompletely etched contact holes, pull-outs, and chemical vapor deposition fill problems are common failure mechanisms for tungsten contacts. Such defects are detectable by an E-beam inspection tool that compares the voltage-contrast images of the same region within multiple dies. Differences in the images are attributed to differences in the electrical connectivity of imaged features. This differential analysis permits relatively high throughput and accuracy.

Figure 6: Micrographs illustrating the ion-beam voltage image-contrast technique: (a) a redetected defect in the tungsten interconnect layer, (b) the reference image with a good contact used for comparison, and (c) the defective contact outlined by automatic defect relocation. (Images in this figure show a portion of the original field of view, magnified and cropped for clarity.)

Although an E-beam is included in the in-line system, the defects in the tungsten layer were redetected using the tool's ion beam for two reasons. First, the ion beam typically creates better voltage contrast than the E-beam because ions scatter more secondary electrons when they hit the sample surface than E-beams do. Second, when the ion beam is used to perform defect redetection, cross-sectioning can follow immediately. The redetection software automatically located the defect within the field of view (Figure 6a) by comparison with a reference image of the same region (Figure 6b). Once redetected, the defect was cross-sectioned to reveal its physical cause, which was an incomplete contact etch, as shown in Figure 7.

Figure 7: Micrograph of the incomplete contact etch responsible for the defect in Figure 6.

Copper interconnect layer defects were also analyzed using the voltage-contrast technique. In this case, an E-beam inspection tool had identified open elements—that is, electrically floating metal lines that should have been grounded. The defect file was then transferred to the in-line system for cross-sectioning. Although many of the defects were similar in appearance in the E-beam voltage-contrast image, cross-sectioning revealed at least five different root-cause mechanisms: bumps in the dielectric (Figure 8a), bad metal filling during first-level copper deposition (Figure 8b), missing metal at the bottom of the interconnect (Figure 8c), oxide underetch (Figure 8d), and bad metal filling during second-level deposition (Figure 8e). The redetection software enabled the rapid, unattended collection of a statistically representative sample of the defect population, while Pareto analysis allowed the defectivity engineer to direct corrective attention to the most important problems first.

Figure 8: Cross-sectional images of a copper interconnect layer showing five types of defects: (a) a bump in the dielectric, (b) bad metal filling during first-level copper deposition, (c) missing metal at the bottom of the interconnect, (d) oxide underetch, and (e) bad metal filling during second-level deposition.

Conclusion

In-line analysis may become an important part of defectivity control efforts during process development and high-volume manufacturing of advanced semiconductors, ultimately improving productivity and profitability. The technology is fast, easy to use, and cleanroom compatible. Moreover, it provides detailed 3-D images for use in the root-cause analysis of defect excursions. Its automation capabilities can free defectivity engineers for other tasks while permitting the collection of statistically representative data. In addition, the cost of acquiring defect data may decrease because there is no need to scrap analyzed wafers. Finally, in-line tools provide results in minutes, while laboratory analysis is too slow to support real-time process control.

Stéphanie Blanc-Coquand is lead engineer in the defectivity group at the STMicroelectronics fab in Crolles, France, where she is responsible for joint development projects with defectivity tool suppliers. She received a degree in materials science from the Institut National des Sciences Appliquées de Lyon in 1998. (Blanc-Coquand can be reached at +33 476 925684 or stephanie.blanc-coquand@st.com.)

Benoit Hinschberger is a defectivity engineering team leader in the device and yield engineering department at STMicroelectronics in Crolles. He joined the company in 1998 after working as a service and applications engineer in the field of defect review SEM in the United States and Europe. Hinschberger holds a degree in electrical engineering from the Institut National Polytechnique de Grenoble, having graduated in 1994. (Hinschberger can be reached at +33 476 926341 or benoit.hinschberger@st.com.)

Eric Rouchouze is the defectivity and process-transfer manager at STMicroelectronics in Crolles, a position he has filled since 2002. In 1992, he joined the company as a lithography process engineer in charge of litho clusters ramp-up and alignment strategies. In 1997, he became photo process section manager. He received a degree in chemical engineering from the Ecole Supérieure de Chimie Industrielle de Lyon, France, in 1987. He also received a degree in physical chemistry from Rice University in Houston. (Rouchouze can be reached at +33 476 926373 or eric.rouchouze@st.com.)

Emmanuel Sicurani is responsible for in-line FIB and E-beam inspection in the metrology defectivity group at the Commissariat à l'Energie Atomique–Laboratoire d'Electronique de Technologie de l'Information (CEA-LETI). He joined CEA-LETI in 1993 and worked as a CMP process engineer for several years. In 2001, he joined the Crolles2 Alliance in Crolles, France. He received a degree in metallurgy engineering at the Conservatoire National des Arts et Métiers de Grenoble in 2001. (Sicurani can be reached at +33 43 8922404 or emmanuel.sicurani@st.com.)

Janet Teshima is vice president of yield enhancement systems at FEI (Hillsboro, OR), and has held positions in applications and marketing since joining the firm in 1993. She has a BA in earth science from the University of California, Berkeley, and a masters in geology from Arizona State University in Tempe. (Teshima can be reached at 503/726-7623 or jmt@feico.com.)

Marc Castagna joined FEI as an applications development engineer in 2000 and is currently assigned to Crolles, France, as an applications engineer assisting in the joint development of an automated in-line system for semiconductor defectivity applications. He graduated from Arizona State University, receiving a BS in physics in 1997. (Castagna can be reached at +33 680 489117 or mcastagna@feico.com.)

Matt Weschler is also on temporary assignment from FEI in Crolles, France, where he works closely with STMicroelectronics. He holds an MS in organic chemistry from Florida State University (Tallahassee), where his studies were focused in the areas of photochemistry and laser-based chemistry. (Weschler can be reached at +33 677 663233 or mweschler@feico.com.)

Larry Dworkin joined FEI in 2002 and is currently involved in developing applications for the firm's dual-beam yield enhancement systems. He was previously a defect reduction engineer working on copper/low-k integration. Dworkin received an MS in physics from UCLA in 1997. (Dworkin can be reached at 503/726-7658 or ldworkin@feico.com.)

Didier Renard is a field application specialist at FEI; in this position, he provides FIB support for semiconductor manufacturers in Europe. He has a master's degree in solid-state chemistry from the University of Paris, Orsay. (Renard can be reached at +33 67 1578508 or didier.renard@feico.com.)