RequestLink
MICRO
Advertiser and
Product
Information

Buyer's Guide
Buyers Guide

tom
Chip Shots blog

Greatest Hits of 2005
Greatest Hits of 2005

Featured Series
Featured Series


Web Sightings

Media Kit

Comments? Suggestions? Send us your feedback.

 

MicroMagazine.com

Defect/Yield Analysis and Metrology

Detecting resistive shorts and opens using voltage contrast inspection

Oliver D. Patterson and Horatio Wildman, IBM; and
Doron Gal and Kevin Wu, KLA-Tencor

In-line voltage contrast inspection is a well-developed method for detecting hard shorts and opens. The technique is quite commonly used in the semiconductor industry to detect contact and via opens as well as metal, gate-level, and active-area shorts.1–9 These inspections provide timely and high-quality feedback to achieve yield improvements and line health monitoring.

ICs are also susceptible to softer defects, such as vias and contacts with elevated resistance, and resistive shorts between metal runners. While inspection scanning electron microscopy (SEM) can detect via chains with soft opens such as that illustrated in Figure 1, which was caused by voiding, SEM’s ability to detect these types of failures is not well established. In this case, probe data indicated that via chain resistances were elevated from tenfold to a thousandfold.

Figure 1: Figure 1: Via chain with soft opens as it appears in an inspection SEM image.

Knowing exactly what resistances can be detected and how the signal varies with changes in resistance would be of great value. This article highlights an effort by IBM (Hopewell Junction, NY) and KLA-Tencor (San Jose) to quantify such an inspection method.

Programmed defects (PDs) provide an effective way to calibrate an inspection system. Because the metal runners illustrated in Figures 2 and 3 were alternately grounded or floating, they appear in the images as bright or dark, respectively. By creating an open, as shown in Figure 2, a normally bright line becomes dark. By creating a bridge, as shown in Figure 3, a normally dark line becomes bright. Inspection conditions can be tuned so that nearly all PDs can be detected, while most nuisance defects are ignored. That concept was applied in this study.

Figure 2: Open PD on a metal comb test structure.
Figure 3: Short PD on a metal comb test structure.

To establish the sensitivity of an inspection SEM to resistive opens and shorts, an experiment was performed in which structures with resistive defects were designed using integrated polysilicon resistors. The experiment showed that voltage contrast inspection is excellent for detecting resistive shorts but not as good at detecting resistive opens. In fact, the inspection SEM was unable to detect any resistive open PDs automatically. Consequently, a model was developed to clarify these results.

Experimental Setup

Equipment. The primary inspection tool used to perform the work described in this article was an eS31 inspection SEM from KLA-Tencor, although an eS32 was employed in one case.

Structures were generated using IBM’s 65-nm bulk silicon technology, which uses copper/low k–dielectric interconnect. These structures contain a polysilicon resistor module, which was used to generate the PDs.

Methodology. Metal µLoop test structures from KLA-Tencor such as those presented in Figures 2 and 3 were used in this study.6–9 As shown in the schematic diagram in Figure 4, resistive open PDs were created by placing resistant structures in series with the metal runners. Resistive short PDs were created, as shown in the schematic diagram in Figure 5, by bridging a floating metal element with one that was grounded with a resistor. The defect types, created using integrated polysilicon resistors, are listed in Table I along with their resistance, location, and detection status. The largest defect, with a resistance of 4000 kΩ, occupied an area 100 × 50 µm in size. Resistances were selected so that one group of defects would be detected easily, another group would be detected only under the best E-beam conditions, and a third group would not be detected at all. Defect types with each level of resistance were inserted three times per reticle field. A wafer was inspected after the metal1 CMP step.

Figure 4: Schematic diagram of a metal open test structure with a resistive PD in series.
Figure 5: Schematic diagram of a metal comb PD with a resistive PD short.

Experimental Results

All short PDs were detected easily. Even the 400-kΩ short PD produced a very strong signal under a wide range of beam conditions. However, none of the open PDs were detected, regardless of the conditions. Even when a 4000-kΩ test structure was used, the inspection system was not sensitive to 4000-kΩ opens.

Table I: List of PDs, including their type, resistance, location on the structure, and detection status.

To investigate inspection SEM’s sensitivity to resistive opens further, the wafer was subsequently inspected at KLA-Tencor’s San Jose facility on an eS32 system, which can lower landing energies. With a landing energy of 300 eV, a beam current of 212 nA, and a high-extraction slow scan, the tool detected a faint signal from the 4000-kΩ resistance open. However, the signal disappeared when a standard eS31 scan was performed. The slow scan of the eS32 derives an average signal over a longer time period, which may provide a higher signal-to-noise ratio than the eS31. In fact, the swath analysis presented in Figure 6 shows only a four-gray-level range for the resistive open line, which is substantially below the noise floor for regular inspection.

Figure 6: Signal intensity versus pixel from a swath analysis. It shows only a four-gray-level difference for a resistive open line.

Model

To interpret the resistive open result, a basic discussion of E-beam inspection detection mechanisms is in order. The minimum detectable voltage difference in an image produced using E-beam-induced voltage contrast depends on several variables: the signal-to-noise in the image, the energy distribution of the secondary electrons, and the local wafer surface geometry. Depending on the design of the detector, the signal is a mix of secondary and backscattered electrons, although only the secondary electrons vary with the induced voltage. The brightness of a grounded conductor (ground potential is considered to be at whatever voltage the entire wafer is held) is determined by the landing energy of the primary electrons on the secondary electron yield. Electron yield curves for copper and tungsten are shown in Figure 7.

Figure 7: Electron yield curves for copper and tungsten based on peak and crossover values from reference 11.

Secondary electron yield is the ratio of emitted electrons to impinging primary electrons at normal incidence as a function of landing energy. It depends on the material and on the material’s crystal orientation and surface roughness. For most conductors other than carbon, the yield is greater than 1 for a range of energies near the maximum. For those energies, an electrically floating conductor would quickly charge to a positive voltage.

Optionally, the ultimate yield of electrons can be modulated by utilizing an extracting or retarding electric field above the wafer. A retarding field repels low-energy electrons back to the surface, effectively reducing the ultimate yield but also potentially offering preferential collection to specific electron energies. An extracting field can enable efficient collection of all available electrons. In either case, the brightness or pixel intensity of grounded conductors depends on the number of electrons emitted as a result of the primary electrons that hit the conductor during the pixel time. With a large extraction field and an efficient detector, more electrons can be detected than are in the beam.

Signal to noise cannot be better than the signal to noise in the E-beam itself. In the case of the Schottky field emission sources used in modern electron columns, the signal to noise is limited only by Poisson counting statistics, where the expected standard deviation is equal to the square root of the number of electrons detected. The number of electrons collected at the detector will be substantially less than the number in the beam for two reasons: electrons are actually emitted in all directions so that only a small percentage hit the detector, and the detector is not completely efficient. For a 212-nA E-beam at a standard inspection speed, the number of electrons in the E-beam is approximately 300–500. The signal to noise is 4.5 to 5.7%, or ~6–10 grey levels within a difference image depending on the nuisance rates desired.

The minimum detectable voltage difference depends inversely on how rapidly signal intensity varies with local surface voltage. Voltage contrast results primarily from the effect of the local electric field on the secondary electrons. Landing energy changes with conductor voltage, but the resulting change of intensity caused by a change in landing energy—at most 0.03% per volt—is insignificant. Voltages from local charging modulate brightness by creating potential barriers for secondary electron emission. Secondary electrons must have sufficient energy, with respect to Fermi energy, to overcome not only the potential barrier resulting from the work function, but also the retarding potential set up by local charging. A small change in surface voltage causes a large change in intensity because most of the secondary electrons have relatively little energy. Estimating the rate of intensity change with voltage requires knowledge of the potential field above the surface of the conductor.

In the space above the conductor, the potential must satisfy Laplace’s equation. For simple geometries, this equation can be solved analytically. Figures 8 and 9 show the solution in a scenario similar to that shown in Figures 2, 3, 4, and 5, where alternating metal lines are grounded. To simplify the problem, it is assumed that the rapidly scanning beam induces an equal voltage on the floating lines, that there is negligible insulating space between the lines, and that there is an extracting field above the surface. The topographical map illustrated in Figure 8 indicates that a potential barrier exists over lines with an induced voltage. The vertical axis is the potential energy, and x and z are the coordinates within a cross-sectional slice that includes both the metal lines and the Wehnelt plate. Figures 9a and 9b show how the potential energy changes in lines directly above and normal to the floating and ground electrodes, respectively.

Figure 8: Topographical plot of potential energy above a comb structure with alternating grounded and floating lines.
Figure 9: Diagrams showing how potential energy changes in lines (a) normal to the ground electrodes and (b) directly above a floating electrode.

The potential energy of an electron as it leaves the surface is just the electron charge times the voltage at that distance above the wafer. In the absence of an extracting field, the voltage just above the surface (a few linewidths in the z direction) is equal to the average local voltage, Vav. For an electron emitted from a conductor at the most positive local voltage, Vc, the potential barrier is equal to e( Vc– Vav).

For the sake of comparison, consider a comb structure in which only one tooth floats (and, therefore, charges), while the others are grounded. In this scenario, Vav would be close to 0, and the potential barrier would therefore be close to eVc. The end result would be greater contrast between the floating and grounded lines. The exact effect this extra potential barrier has on the measured intensity depends on the shape of the energy distribution curve of emitted electrons.

The maximum positive charge that develops on floating lines was estimated in the following manner. Two parameters were multiplied together to obtain the distribution of escaping secondary electrons as a function of energy: an appropriate electron energy distribution and transmission function for copper.11,12 A typical work function of 4 V and a secondary electron emission angle of 90° were assumed. The transmission function is a function of the potential barrier, which changes as the floating teeth charge. Therefore, multiple secondary electron emission plots were obtained, as shown in Figure 10. These plots were integrated to obtain the total emitted electrons for the given potential barrier, and then they were used to create the effective secondary yield plot shown in Figure 11. A constant was added to account for the backscattered electrons unaffected by charge voltage. Values of 1.3 for the maximum secondary electron yield, δ, and 0.34 for the backscattered electron coefficient, η, were used for copper.11 The curve was normalized to the sum of these numbers (s = δ + η), an estimate of the maximum number of electrons emitted for copper.

Figure 10: Secondary electron emission for several different potential barriers.

As a floating conductor charges, the operating point travels down the curve in Figure 11 to the right. Equilibrium, where “electrons in” equal “electrons out”, is reached at the unity crossover. This is the maximum positive voltage that can be induced on the floating conductor. Traces for two different scenarios are shown in Figure 11. The trace with equilibrium at 3.1 V is for the scenario in which the retarding potential is half of the induced voltage. This occurs with the comb pattern assumed for this model and when the Wehnelt voltage is 0 V. The retarding potential decreases with Wehnelt voltage, as shown in Figure 9. The second trace with equilibrium at 1.5 V represents the scenario in which the retarding potential is equal in magnitude to the induced voltage. This occurs when there is only one floating structure (e.g., the comb structure in which all the teeth are grounded except the one with the resistive open mentioned earlier). The true induced voltage on these experimental structures was <3.5 V as a Wehnelt voltage was applied, but more than the 1.5 V expected for a single floating tooth.

Figure 11: Chart showing how the total electron yield changes as a positive electric charge builds up on a floating line.

Assuming a maximum charge of 3 V and a starting total electron yield of 1.64 of the incident beam, the average rate of change of total electron yield is about –0.21, or 13% per volt. Given shot noise of ~5%, this amounts to a detectable voltage of 5%/(13%/V) = 0.38 V.

For floating lines, the voltage increases until the number of emitted electrons equals the number of incident electrons. For a line with a resistive open, equilibrium is reached when the current flow from incident electrons equals the current from electron emission plus the leakage current through the resistive open. For resistances at the detection limit, the electron emission is negligible.

The voltage of a conductor with a resistive defect can be modeled using Ohm’s law:

V = IdefectRdefect

where Rdefect is the resistance of the defect and Idefect is the current through the defect. The current is a function of the SEM beam current, ISEM, the total effective yield (s = δ + η), and the percent of this current that actually hits the conductor of interest (P). Therefore, the first equation can be rewritten as follows:

V = ISEM(σ – 1)PRdefect

The maximum beam current for the eS31 is 212 nA. The largest programmed defect used in experiment 1 was 4000 kΩ. The peak σ on the electron yield curve is less than 1.64. Assuming a P of 0.5, the induced voltage obtained by solving the second equation is 0.27 V. This value is slightly below the estimated minimum detectable voltage, which makes sense since only a four-gray-level difference was observed.

Conclusion

The study reported in this article verifies that resistive opens and shorts can be detected using voltage contrast inspection. Sensitivity to resistive shorts from 0 Ω to >4000 kΩ and resistive opens from millions of ohms down to 4000 kΩ was demonstrated for copper interconnect. For the resistive opens, the 4000-kΩ one was too faint to be detected automatically. A model was presented to help explain these results.

The sensitivity of the inspection tool at the extremes of these ranges can likely be modulated by tuning the inspection conditions. New wafers are being created with larger resistances to demonstrate this principle and to establish the threshold at which resistive opens can be detected automatically. A follow-up study with resistances in the megohm range is in progress. Based on the experimental and modeling work described here, it is anticipated that the detection threshold for resistive opens in copper interconnect will be between 10 and 20 MΩ. This information will be useful for understanding how best to utilize inspection SEM technology in order to tackle yield problems resulting from resistive defects.

Acknowledgments

This article is an edited version of a paper that was presented at the IEEE/SEMI Advanced Semiconductor Manufacturing Conference, held May 22–24, 2006, in Boston. The authors thank Yi Feng and Loren Hahn of IBM and Chris Kang of FEI for their assistance in attempting to create high-resistance defects using an in-line focused ion-beam tool. They would also like to thank Jan Lauber and Indranil De of KLA-Tencor for their suggestions on voltage contrast detection theory.

References

1. OD Patterson et al., “Real Time Fault Site Isolation of Front-End Defects in ULSI-ESRAM Utilizing In-Line Passive Voltage Contrast Analysis,” in Proceedings of the International Symposium for Testing and Failure Analysis (Materials Park, OH: ASM International, 2002), 591–599.

2. V Liang, H Sur, and S Bothra, “Passive Voltage Contrast Technique for Rapid In-Line Characterization and Failure Isolation During Development of Deep-Submicron ASIC CMOS Technology,” in Proceedings of the International Symposium for Testing and Failure Analysis (Materials Park, OH: ASM International, 1998), 221–225.

3. JL Baltzinger et al., “E-Beam Inspection of Dislocations: Product Monitoring and Process Change Validation,” in Proceedings of the Advanced Semiconductor Manufacturing Conference (Piscataway, NJ: IEEE, 2004), 359–366.

4. A Nishikawa et al., “An Application of Passive Voltage Contrast (PVC) to Failure Analysis of CMOS LSI Using Secondary Electron Collection,” in Proceedings of the International Symposium for Testing and Failure Analysis (Materials Park, OH: ASM International, 1999), 239–243.

5. AVS Satya, “Microelectronic Test Structures for Rapid Automated Contactless Inline Defect Inspection,” IEEE Transactions on Semiconductor Manufacturing 10, no. 3 (1997): 384–389.

6. OD Patterson et al., “Rapid Reduction of Poly-Silicon Electrical D0 using µLoop Test Structures,” in Proceedings of the Advanced Semiconductor Manufacturing Conference (Piscataway, NJ: IEEE, 2003), 266–272.

7. K Weiner et al., “Defect Management for 300mm and 130mm Technologies Part 3: Another Day, Another Yield Learning Cycle,” Yield Management Solutions 4, no. 1 (2002): 15–27.

8. R Guldi et al., “Characterization of Copper Voids in Dual Damascene Processes,” in Proceedings of the Advanced Semiconductor Manufacturing Conference (Piscataway, NJ: IEEE, 2002), 351–355.

9. A Shimada et al., “Application of µLoop Method to Killer Defect Detection and In-Line Monitoring for FEOL Process of 90nm-Node Logic Device,” in Proceedings of the International Symposium for Semiconductor Manufacturing (Piscataway, NJ: IEEE, 2004).

10. CH Wang et al., “The Study and Methodology of Defects Isolation for Contacts of Non-Isolated Active Regions on New Logic Designs,” in Proceedings of the International Symposium for Testing and Failure Analysis (Materials Park, OH: ASM International, 2005), 479–483.

11. H Seiler, “Secondary Electron Emission in the Scanning Electron Microscope,” Journal of Applied Physics 54, no. 11 (1983): R1–R18.

12. ZJ Ding, XD Tang, and R Shimizu, “Monte Carlo Study of Secondary Electron Emission,” Journal of Applied Physics 89, no.1 (2001): 718–726.


Oliver D. Patterson, PhD, develops and applies tools and methodologies for improving the yields of cutting-edge technologies at IBM’s 300-mm fab in East Fishkill, NY. A current focus of his work is the development of methods for using inspection SEM technology for in-line detection of yield-limiting defects. Previously, he held a similar position at Agere Systems in Orlando, FL. He received an SB degree from the Massachusetts Institute of Technology in Cambridge, MA, an MS from the University of Wisconsin–Madison, and a PhD from the University of Michigan in Ann Arbor, all in electrical engineering. (Patterson can be reached at 845/894-3159 or opatters@us.ibm.com.)

Horatio Wildman applies in-line inspection, including E-beam voltage contrast inspections, to achieve advanced yield learning in semiconductor development at IBM in East Fishkill, NY. He has spent the major part of his career in the field of surface analysis, especially Auger electron spectroscopy. He received a BS in physics from Georgetown University in Washington, DC. (Wildman can be reached at 845/297-0787 or hwildman@us.ibm.com.)

Doron Gal, PhD, is a senior product marketing manager in KLA-Tencor’s E-beam inspection division in San Jose. He joined after performing postdoctoral research at the University of Texas in Austin. He received a PhD in physical chemistry from the Weizmann Institute of Science in Israel. (Gal can be reached at 408/368-8129 or doron.gal@kla-tencor.com.)

Kevin Wu works in the wafer inspection group at KLA-Tencor, where he is assigned to the IBM business unit. He worked in the photolithography and yield enhancement departments at Samsung Semiconductor in Austin, TX. He received BS and MS degrees in materials science and engineering from the State University of New York in Stony Brook. (Wu can be reached at 845/897-1714 or kevin.wu@kla-tencor.com.)


MicroHome | Search | Current Issue | MicroArchives
Buyers Guide | Media Kit

Questions/comments about MICRO Magazine? E-mail us at cheynman@gmail.com.

© 2007 Tom Cheyney
All rights reserved.