Analysis and Metrology

Using partial-pressure analysis to detect contamination in an oxygen gas supply

Thomas P. Schneider, Ferran Scientific; and Kelly J. Taylor, David A. Rothenbury, Mark Chavis, Timothy Hoff, and Craig H. Huffman, Texas Instruments

Early detection of process gas contamination is critical to preventing misprocessing of semiconductor product wafers, which usually results in scrap. Waiting for catastrophic contamination to occur and then troubleshooting the problem—which is common practice in the industry—is not only costly, it is shortsighted, especially since low-cost monitoring technologies are available.

The ratio of the cost to develop and maintain gas-purity-monitoring instruments to the cost of lost revenue from scrapped product approaches zero within a short time (weeks to months). Currently, 300-mm-diam bare silicon wafers cost as much as $1190 each, and increase in value with each process step. A gas monitor such as the one used in the study presented in this article can be purchased for $6000. Assuming, for simplicity, a case of front-end single-wafer processing in which each wafer takes an average of 3 minutes to process (including setup time and chucking/dechucking), the gas purity monitor will pay for itself in 23 minutes of scrap prevention. Making the further assumptions that the end product profit per device is $100 and there are 250 devices per wafer, the scrap produced in 23 minutes affects approximately $190,000 of gross revenue.

One useful instrument for monitoring process gas purity is the partial-pressure gas analyzer (PPA), also known as a residual gas analyzer (RGA). A PPA reports the partial pressure (in torr) of almost all species present in the gas being sampled and the atomic masses of the detected species. Species identification is then usually straightforward, using either the periodic table or stored libraries of PPA data. The price of most PPAs versus the amount and value of information they can provide makes their use quite cost-effective. Literature is available describing the successful use of PPAs for process gas—purity monitoring and diagnosis, as well as suggesting that there is a clear need for the use of such monitors in current and next-generation tools.^1—3 This article presents a case study that illustrates the usefulness of PPA monitoring following an instance of catastrophic gross contamination.

The Problem

Over the course of several hours, various events occurred in a wafer fab that were symptomatic of gas contamination. The chronology of the indications that a problem existed was as follows: The first sign of trouble was when the ashers that remove photoresist from wafers became ineffective. Three hours later, some of the furnace torches turned off, and the ozonators for silicon oxide deposition tools stopped working. Then, after another 2 hours had passed, it was discovered that there was black char on the quartz reactor walls after a standard furnace process and there was no silicon oxide on the processed wafers. At approximately the same time, off-color plasma was detected in a plasma processing system.

Given this sequence of events, it was not immediately obvious that they were linked. Fortunately, process knowledge came into play once the fourth symptom occurred and engineers interpreted the problem as the presence of nitrogen in the oxygen gas supply. However, at this point, 5 hours after the first symptom had occurred, it could not be discerned if the problem was a continuous flow or a one-time burst of N₂ into the O₂ main supply, or an atmospheric leak into the O₂ main supply.

Equipment and Experimental Setup

To follow up on the problem, a miniature PPA, the MicroPole analyzer (Ferran Scientific, San Diego) was brought in to help determine the source of the contamination and to monitor the gas supply during the recovery period. The design and operating principles of this PPA are described in detail elsewhere.⁴ Its small size permits the unit to operate at pressures up to 15 millitorr without differential pumping. The detector also features a low profile and a modular design, which make it convenient for process gas troubleshooting, when it is necessary to move from tool to tool and to collect data rapidly.

During the evaluation, the PPA was placed in two locations (in two separate submains) to monitor oxygen purity. The first location was in the central wafer handler of a deposition system; the second was in the main chamber of a plasma etcher. In each case, the O₂ gas mass-flow controllers (MFCs) were opened fully and the vacuum systems were pumped turbomolecularly. In this configuration, the PPA did not need differential pumping and the pressures during gas-purity sampling ranged between 5 x 10^—4 and 1.5 x 10^—2 torr. It took approximately 20 minutes to install the analyzer and to begin collecting data. The PPA was operated in two data-display modes: analog spectra and partial-pressure trend. The analog spectra data were used for species detection and the partial-pressure trend data were used to monitor changes in contamination levels in real time.

Figure 1: Analog spectra data from the central wafer handler of a deposition tool. The dashed curve is a 100x amplification of the solid curve.

Results and Discussion

The initial analog spectra data collected from the central wafer handler of a deposition tool are displayed in Figure 1. The dashed curve is an amplification (100x) of the solid curve. Thus, both curves represent data collected while the O₂ MFC was fully open and gas was flowing through the process chamber and into the central handler. The peak at 2—3 amu represents hydrogen (H₂+), the peak centered on 14 amu is nitrogen (N^{+), and the small peak at 18 amu is background moisture (H₂O+) in the system. The largest feature, the peak that extends out of the figure at 28 amu, is a combination of N₂+ and carbon monoxide (CO+), and the peak centered over 44 amu is carbon dioxide (CO₂+). It is known that the ratio of the peak intensity of 28 amu to that of 44 amu is 0.114 for CO₂ and that similar ratios for peaks at 14 and 28 amu are 0.006 for CO and 0.072 for N₂, respectively.5 For the data displayed in Figure 1, the respective peak intensity ratios for 14 to 28 amu and 28 to 44 amu are 0.05 and 0.005. After comparing these ratios with the known ratios from the reference source,5 it was concluded that the dominant ion species at 28 amu was N₂+. Similar logic was followed in evaluating all of the analog spectra data that were collected during this case study.}

Figure 2: Partial-pressure trend data from the wafer handler for 28- (N₂+, dashed curve) and 32-amu (O₂+, solid curve) species. The data were collected during tests to determine the contamination source and which bulk gases were affected.

Figure 2 displays the partial-pressure data trends for the 28- and 32-amu species (N₂+ and O₂+, respectively) that were collected during a series of tests that were run to determine the source and extent of the contamination. During the first test, the O₂ MFC was activated and the resulting data indicated there was a small change in O₂+ partial pressure and a large change in N₂+ partial pressure. These trends were interpreted as a clear indication that the O₂ process gas was contaminated with N₂. The second test was similar to the first except the argon MFC was activated. In this case, the N₂+ and O₂+ levels did not change appreciably, which was interpreted as indicating that there was no N₂ contamination in the Ar process gas. These two tests were then repeated, yielding similar results. The last test shown in Figure 2 involved the activation of the N₂ MFC. In this case, the O₂+ level did not change appreciably, compared with the N₂+ level in first test, which indicated that the N₂ process gas was not contaminated with O₂.

Figure 3: Partial-pressure trend data from the wafer handler for 18- (H₂O+, solid curve), 28- (N₂+, dotted curve), and 32-amu (O₂+, dashed curve) species. The dominance of the H₂O+ signal immediately following collection of a gas sample was a definite sign that the O₂ submain had been exposed to atmosphere.

Other analog spectra and trend data not illustrated here initially indicated there was no atmospheric contamination. However, during the PPA monitoring in the central wafer handler, the submain was exposed to atmosphere when samples were collected for off-line spectral analysis of the O₂ gas. The results of this event are seen in Figure 3, which displays the partial-pressure trends of the 18- (H₂O^{+), 28- (N₂+), and 32-amu (O₂+) species. The region labeled "sampling from O₂ submain" represents a continuation of the initial monitoring of the system gases with the O₂ MFC activated. Immediately following the collection of the gas sample for off-line analysis, the partial-pressure trends changed precipitously; H₂O+ became the species with the highest partial pressure followed by N₂+ and O₂+. The ratio of N₂+ to O₂+ became ~6, which is close to the known ratio for these species in air (~4—5). That the N₂+ concentration remained slightly higher than O₂ was reasonable since N₂ had previously been the dominant species in the O₂ process gas. As these PPA results indicated, an atmosphere leak had exposed the internal chamber surfaces to H₂O during sample collection, and the chamber and gas lines had to be dried down before any processing could occur. The fact that the PPA immediately detected the atmospheric exposure of the O₂ submain is substantial evidence of the potential usefulness of installing PPAs on process tools for continuous gas monitoring.}

Figure 4: Background analog spectra data from the main chamber of a plasma etcher. The dashed curve is a 10x amplification of the solid curve. The 18-amu (H₂O+) peak is strongest because the chamber had been exposed to atmosphere during installation of the partial-pressure analyzer.

After measurements from the deposition-tool wafer handler were completed, the PPA was moved to the main chamber of a plasma etcher served by a different O₂ submain. The background analog spectra data for the etch chamber are displayed in Figure 4. It is important to fingerprint background residual gases desorbing from internal chamber surfaces in order to reduce system-to-system variation. As in Figure 1, the dashed curve is an amplification of the solid curve; the species also appear in the same locations. In this case, the H₂O^{+ peak was the largest because of the presence of moisture on the chamber walls, and all measured species levels were slightly higher than normal as a result of the chamber's exposure to atmospheric pressure during installation of the PPA. The high H₂O+ peak decreased quickly in subsequent readings, followed by a slower decrease over time, as would be expected. (The evolution of H₂O+ immediately following the exposure of a vacuum processing chamber to atmosphere will be the subject of another study.)}

The first test run on the plasma etcher verified that N₂ was flowing through the O₂ MFC. The N₂ MFC was then activated to determine if doing so would change the intensity of the 32-amu (O₂+) peak, and the resulting data indicated that it did not. These tests confirmed that the PPA was responding correctly to changes in gas species and thus was suitable for monitoring the recovery of the purity of the O₂ process gas main supply.

Figure 5: Partial-pressure trend data from the etch chamber for 28- (N₂+, dashed curve) and 32-amu (O₂+, solid curve) species. The recovery from the N₂ contamination problem can be observed on the right-hand side of the figure.

Figure 5 displays the partial-pressure trend data for N₂+ and O₂+ collected during testing and recovery. The "O₂ flow test" regions indicate that N₂ was continuing to flow into the reactor when the O₂ MFC was activated. At about 23 minutes on the horizontal axis, however, there is an early indication that the O₂+ concentration has begun increasing while the N₂+ level decreases. Then at approximately 32 minutes there is a crossover in the concentrations, indicating recovery from the N₂ contamination event is definitely occurring, and this switch is followed abruptly by what appears to be a rapid and full recovery to nearly pure O₂ flowing through the O₂ MFC. This result was expected because the entire O₂ gas line system was being purged with O₂ from the source. In this case, the PPA monitored the recovery of the purity of the gas supply submain in real time.

Conclusion

In the case study described above, a PPA proved useful for troubleshooting and monitoring the recovery of a bulk gas system following an accidental contamination event. Employing such PPAs on process tools in a full-time monitoring mode would have resulted in earlier detection and diagnosis of the gas contamination problem, thereby preventing costly misprocessing and product loss. Because it is not common practice in the semiconductor industry to have purity monitors along the bulk gas distribution system, it would make good sense to install PPAs on process tools for gas-purity sampling. Full-time PPA monitoring has the potential to decrease wet clean recovery times, to detect cross-chamber contamination in cluster tools when used on the central robot chamber, and to speed up process development, as well as to prevent misprocessing, the importance of which increases with each technology node.

Acknowledgments

The authors appreciate the support provided by R. Gale, R. A. Bowling, and J. Abernathy during the course of this study. In addition, S. Bushman was instrumental in setting up the project. We also appreciate the efforts of Tammy Fletcher and Dennis Shipley, who assisted in the installation of the PPA data collection system.

References

1. Schneider TP, Krocak P, and Van Eck B, "Real-Time In Situ Residual Gas Monitoring during Ion Implantation in High Volume Semiconductor Manufacturing," Future Fab International, 4:237—239, 1997.

2. Haider AM, Fu TTH, and Rosenberg RW, "Investigating Use of an In Situ RGA for Process Monitoring and Diagnostics," MICRO, 13(4):35—41, 64, 1995.

3. Peters L (ed), "Residual Gas Analysis," Semiconductor International, 20(11):94—101, 1997.

4. Ferran RJ, and Boumsellek S, "High-Pressure Effects in Miniature Arrays of Quadrupole Analyzers for Residual Gas Analysis from 10^—9 to 10^—2 Torr," Journal of Vacuum Science and Technology A, 14:1258, 1996.

5. Drinkwine MJ, and Lichtman D, Partial Pressure Analyzers and Analysis, part of the American Vacuum Society Monograph Series, Whetten NR and Long R Jr. (eds), New York, American Vacuum Society, 1978.

Thomas P. Schneider, PhD, works for Ferran Scientific, where his responsibilities include sensor applications for fault detection, classification, and process control. Before joining Ferran, he worked as a postdoctoral fellow at Sematech and as a member of the technical staff at Texas Instruments in the plasma etch R&D group. Schneider has a BS in physics from Moravian College, an MS in physics in applied quantum theory from Saint Bonaventure University, and a PhD in physics from North Carolina State University, where he studied hydrogen plasma interactions with silicon surfaces. He has several publications correlating real-time in situ metrology to the processing of surfaces. (Schneider can be reached at 919/773-2579, or via e-mail, tomaps@earthlink.net.)

Kelly J. Taylor, PhD, is a senior member of the technical staff at Texas Instruments (Dallas), which he joined in 1990. He is active in R&D of dielectric thin films needed for sub-0.15-µm technologies in development at TI's Kilby Center. Taylor has a BS in physics from Brigham Young University and a PhD in physics from Rice University, where he studied under Nobel Laureate Richard Smalley. He has published many works on low-dielectric-constant thin films and holds five U.S. patents.

David A. Rothenbury is a thin-films equipment engineer at TI's DMOS 5 manufacturing facility. His work includes troubleshooting and providing rapid corrective actions on semiconductor process tools, including the application of in situ sensors for diagnosis of tool health. Before joining TI, he was a thin-film development engineer for vacuum roll-coater applications at Vadeko International (Ottawa, ON).

Mark Chavis is an engineering technician in TI's etch equipment module. With experience in plasma etch and thin-film deposition tools, he has been involved with real-time sensor studies throughout his career. His interest in sensor-based control systems has resulted in contributions to applications development for partial-pressure analyzers, RF sensors, and optical emission spectroscopy systems. Chavis has also used tool data collection systems for troubleshooting and rapidly returning process tools to production status. He has a BS in electrical engineering technology from the University of Southern Mississippi.

Timothy Hoff is a working foreman at TI, overseeing the etch equipment engineering technicians. He has experience in plasma etch tools and ion implantation systems and has been involved in partial-pressure analysis studies throughout his career. His efforts have resulted in the use of PPAs for diagnosing ion implantation vacuum system integrity. Before assuming his current position, Hoff worked on the repair and maintenance staff in TI's R&D groups. He holds an AA in electronic engineering from the National Education Center.

Craig H. Huffman is a member of the technical staff and etch equipment module manager at TI. He was a TI assignee to Sematech as the etch manufacturing tools program manager, where he coordinated process tool improvement testing and performance evaluations at national laboratories and member companies of the consortium. Huffman championed process equipment characterization studies conducted at Sandia National Labs that led to process, tool, and data management improvements benefiting the semiconductor industry. He has also been involved in several studies where sensors were used to improve manufacturing methods. A member of AVS and Sigma Pi Sigma, Huffman holds a BS in engineering physics from Southwestern Oklahoma State University.

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