Facilities Technologies

Maintaining an on-site analytical chemistry lab to monitor air and site water quality

T. Paul Adl, Lawrence S. Turek, Judy A. Montag, and Terrence C. Leslie, Dominion Semiconductor

Analyses of organic contaminants in fab air and reactive silica in DI water demonstrate that an on-site chemistry lab can provide round-the-clock support for the quality control of air and water.

In the 1990s an industrywide trend to outsource support services emerged. Among the casualties of this trend was the maintenance of full-fledged analytical laboratory services for in-house quality control of incoming chemicals and real-time analytical support for the root-cause analysis of process problems. The quality assurance of incoming chemicals, which had previously been subjected to total in-house inspection, migrated to outside vendors, while fabs confined themselves to performing internal spot-test analyses. As a result of this migration, tests to determine chemical-based failure mechanisms, such as foreign materials in the fab environment or particles in deionized (DI) water, are generally performed after the fact by external laboratories.

From its inception, Dominion Semiconductor (Manassas, VA) decided to establish a state-of-the-art on-site analytical chemistry support laboratory with trace instrumental and traditional wet chemical analytical capabilities. This Class 100 lab serves as a competence center for the quality assurance of incoming critical materials, chemical failure analysis, fab environmental monitoring analysis, and facilities support.

An on-site analytical support laboratory offers a fab important advantages. It can provide rapid turnaround times that cannot be matched by external laboratories, even at a high premium. The continual availability of analysts in emergency situations also offers the fab around-the-clock support to qualify chemicals and tools after disruptions and line shutdowns occur.

This article describes two functions of Dominion's on-site chemistry laboratory. Presented first is the lab's method for sampling and measuring organic contaminants in the fab atmosphere. The article then explains how the laboratory applies a traditional wet chemical method to analyze the fab's DI water for parts-per-trillion levels of reactive silica in order to assure water quality and track the longevity of the facility's exchange beds.

Testing for Organic Contaminants in Fab Air

The detrimental effects on wafer integrity of charged mobile species in the form of cations and anions from metallic and nonmetallic sources are well understood in the semiconductor manufacturing industry. Not as well understood are the potential organic molecular contaminants that are present in resists, strippers, and solvents. The degassing and aging of fab construction materials such as paints, filters, and adhesives also may constitute possible sources of airborne contamination.¹

One of the adverse effects of organic particulates and films on wafers is the formation of carbides at high process temperatures.² It has been suggested that because certain organic molecules are hydrophobic, a property they share with the silicon wafer surface, they adhere to the wafer surface and react with it to form molecular species in the form of carbides.³ Hydrophobic organic films that affect wetability in subsequent manufacturing steps are normally difficult to remove because of their water-repelling nature.^4,5 The chemically "reductive" nature of organic substances, together with the physical surface alterations they induce, influences the chemical kinetics of oxidation processes.⁶ In addition, reliability problems associated with gate oxide integrity have been attributed to organic foreign materials.⁷ Analyzing and controlling these materials is a prime function of a fab's analytical laboratory. Analysis and control, in turn, are dependent on the collection of samples in the various process bays, the development of a proper analytical methodology, and the analysis of data from total organic carbon (TOC) tests.

Testing Steps. To trap organic contaminants in various manufacturing bays of the fab, laboratory engineers fill Teflon containers to the top with DI water and place the containers near process tools. The containers, which do not contain impingers, traps, or pumps, are left in the fab without supervision and collected a week later for analysis. For continuous monitoring, fresh containers are placed in different areas of each fab zone when the previous week's samples are collected. After the water samples are collected, organic deposits that collected in the water in the form of particulates, vapors, and films from various sources are analyzed for TOC content. Raw data obtained at the analysis level from different bays are normalized to account for nonuniform evaporation rates and expressed as ng/50 cm²/wk of carbon. The area dimension in this formula denotes the exposed surface area of the containers, which are 8 cm in diameter. This simple sampling method is thought to generate data that accurately reflect contaminant levels in the fab over a prolonged period of time. In fact, contaminants collected in this absorbent "wet" water matrix are believed to better represent contaminants in the overall fab environment than those collected on solid wafer surfaces.⁸

After collection, the samples are analyzed with the total organic carbon analyzer Model TOC-5000 from Shimadzu (Kyoto, Japan) equipped with a nondispersive infrared spectroscopy gas analyzer. Standards prepared from potassium hydrogen phthalate (KHP) are used for instrument calibration, and sodium oxalate primary standards prepared in the lab are the source for spike recovery determinations.

Concerns about using Teflon containers for trapping organic contaminants were addressed at the outset of the testing procedure by running samples from Teflon and quartz containers in the analytical laboratory. Although the samples collected in Teflon containers in different areas of the laboratory showed an average 4% upside bias in TOC content, the decision was made to continue their use. These containers are used for monitoring anions and cations in the fab, and it is practical to use only one set of containers for all contaminants. Control samples in Teflon containers were prepared exactly as the field samples but were kept capped for the duration of the test. These control samples had minimal amounts of TOC in comparison with the field samples.

Test Results. Data obtained for error analysis are presented in Tables I and II. The TOC data in Table I are from three containers that were placed next to one another in the laboratory. Three samples each from two of the containers and one sample from the third container were analyzed. Table II presents data from six samples that were obtained from two containers placed in a carpeted office environment. The statistical analysis results for the seven samples in the laboratory and the six samples in the office area are based on total analytical data and thus disregard between-container variations. The data in Tables I and II demonstrate that in the laboratory, where practically no organic solvents are used, statistically significant lower levels of TOC are collected than in the office area, where paper products abound.

 

Sample TOC (ng/50 cm2/wk) Mean (ng/50 cm2/wk) Sigma Lab 1A 74,700 76,200 3211 Lab 1B 71,800 — — Lab 1C 72,500 — — Lab 2A 79,100 — — Lab 2B 77,000 — — Lab 2C 79,400 — — Lab 3A 78,900 — —

Table I: TOC data for error analysis from three containers placed next to one another in the laboratory.

Sample TOC (ng/50 cm2/wk) Mean (ng/50 cm2/wk) Sigma Office 1A 514,000 529,667 14,459 Office 1B 539,000 — — Office 1C 552,000 — — Office 2A 525,000 — — Office 2B 532,000 — — Office 2C 516,000 — —

Table II: TOC data for error analysis from two containers placed next to each other in a carpeted office area where many paper products are used.

The remainder of the sample in the third lab container was used for a "spike-recovery" experiment, the results of which are shown in Table III. The sample was spiked with a known amount of carbon prepared from a primary oxalate carbon source and analyzed twice against a KHP-based calibration curve. The spike recovery results are well within the acceptable 75–125% recovery limits.

 

Sample Recovery (%) Lab 3A + spike (1) 86 Lab 3A + spike (2) 85 Known carbon from oxalate 95–100

Table III: Spike recovery data from the remainder of the sample in the third container placed in the laboratory.

Figure 1 compares organic levels in a manufacturing zone (X) where organic chemicals are used in significant quantities with those in another zone (Y) where no organic chemicals are used. Student's t statistics run between the two data sets confirm that at a 95% confidence level the difference between the two zones is real and significant, and thus beyond the analytical or experimental error.

Figure 1: TOC content of a fab zone that uses organic chemicals (X) compared with a zone that does not (Y). The data represent weekly samples collected over a 6-month period.

Figure 2 presents TOC data obtained from samples from 10 manufacturing zones (A–J). The data point for each zone is the average of the weekly samples for that zone over a 6-month period. Analysis of variance statistics with a 95% confidence level performed among the 60-point database in Figure 2 indicate a statistically significant overall difference between actual and critical F-factors. More specifically, a Pair–Wise Tuckey's test analysis performed on 45 pairs in the database indicates a significant difference in TOC levels among 13 pairs of process bays.

Figure 2: Data from average weekly TOC samples collected in 10 manufacturing bays over a 6-month period.

Testing for Reactive Silica in DI Water

Because silicon in the form of silica and silicates in the earth's crust is naturally abundant, it is ubiquitous in the waters that flow into semiconductor facilities. The removal of various species of silica in the form of granules, colloids, and dissolved matter presents a continual challenge to water filtration and purification facilities. In semiconductor manufacturing processes, minute amounts of particulate matter, even in the form of submicron colloidal silica suspensions or films resulting from dissolved silica, are harmful. Improvements in filtration and purification technologies have lowered silica levels in process DI water to sub-parts-per-billion levels.⁹ A sampling procedure, sample preparation scheme, and wet chemical method have been adopted at Dominion to detect low-parts-per-trillion levels of dissolved silica. These analyses of low levels of silica in DI water are used to measure water quality and monitor the longevity of purification resin beds.

Testing Steps. DI water is analyzed for parts-per-trillion-level reactive silica after fiftyfold concentrations of the material have been generated. This is accomplished by collecting 2 L of DI water and concentrating it to 40 ml in Teflon containers. The concentrated samples are then treated with ammonium molybdate, after which the resulting molybdosilicic acid is analyzed in a 10-cm quartz cell using a Model 8453 UV/VIS spectrophotometer (Agilent Technologies, Palo Alto, CA) equipped with a diode array detector.¹⁰ Because of the samples' low signal-to-noise ratio at the analysis level, the optical density of the 815-nm absorption band is measured over an extended integration time of 25 seconds. Data for standards and samples are read three times, and regression fit calibration curves are used in all analyses.

To ensure quality control of the analyses, the calibration curves are used to measure known standards that are prepared using "chain" and "dilution" series that differed from the ones used in constructing working curves. Because of the samples' low signal-to-noise ratio and to gain adequate statistical confidence, tenfold analyses are performed. Figure 3 presents the "unprocessed" 815-nm absorption spectra of 3.0-, 5.0-, and 10.0-ppb SiO^₂ standards in a 10-cm cell, which are used to construct a working calibration curve. A typical calibration curve based on these three SiO^₂ measurements and using background wavelengths at 700 and 900 nm for subtraction is presented in Figure 4.

 

Figure 3: Absorption spectra of 3.0-, 5.0-, and 10.0-ppb SiO2 standards in a 10-cm cell.

Figure 4: Calibration curve used to measure reactive silica concentrations. Data points are from three measurements for each SiO2 concentration.

Test Results. The data in Table IV are the raw results of the standards analysis together with the inherent biases and statistical parameters at four levels. These data clearly indicate that the measurements deteriorate from the 10- to the 1-ppb level. This conclusion is based on the calculation of the relative standard deviation (RSD), which rises from 3% at the 10-ppb level to 20% at the 1-ppb level. A similar trend is observed in the range of data relative to the mean. Any bias observed in measured versus known concentration values in Table IV is not considered significant.

If the reactive silica content in the as-received DI-water samples is as low as the site DI-water specification limit of 100 ppt, this would translate into 5 ppb at the analysis level in the test measurements. At that level, the RSD based on the data in Table IV is 5%, from which it can readily be determined whether the water system meets the specification. Furthermore, as a result of the detection-level studies near the 1-ppb level at the analysis stage, 20 ppt can be detected to within 20% for the as-received samples. In routine analyses, the data for the as-received samples are conservatively reported as <50 ppt. The site DI-water specification of 100 ppt is the attainable target specified in Balazs Laboratories' Ultra Pure Water Monitoring Guidelines 2000 for Facility and Fabrication Engineers.¹¹ The acceptable level of dissolved silica in fab DI water for the 0.18-µm technology is 200 ppt while the critical limit is between 500 and 1000 ppt. Table V summarizes the data at the analysis level obtained from the site's high-purity DI-water supply. In all cases except for the Loop Y sample collected on date E, all data for the as-received samples were <50 ppt. For the Loop Y sample collected on date E, the data were 60 ppt.

 

Run No. 1.0 ppb 3.0 ppb 5.0 ppb 10.0 ppb 1 0.89 2.61 6.20 9.22 2 1.44 4.28 5.77 9.99 3 1.10 3.21 5.55 9.77 4 1.52 3.57 5.85 9.69 5 1.31 3.03 5.42 9.72 6 1.11 3.02 5.31 9.29 7 0.89 3.49 5.47 9.19 8 1.19 3.49 5.48 9.39 9 1.11 2.44 5.74 9.99 10 0.84 3.50 5.15 9.48 Mean 1.14 3.26 5.59 9.57 Sigma 0.23 0.50 0.29 0.29

Table IV: Data from four known standards versus a working calibration curve.

Date Sample Reactive Silica (ppb) A Loop X Loop Y 1.8 0.6 B Loop X Loop Y None 1.4 C Loop X Loop Y 0.5 0.3 D Loop X Loop Y 0.7 None E Loop X Loop Y 2.0 3.0 F Loop X Loop Y 0.8 0.7

Table V: Data showing reactive silica (parts per billion at analyses level) in the high-purity DI-water supply.

Conclusion

Dominion Semiconductor, like other high-volume DRAM chip producers, continues to push the leading edge of manufacturing toward finer linewidths and denser packaging technology. Leading-edge technology requires diligent quality control of the fab environment and DI-water purity. Effective methods for collecting, preparing, and analyzing samples at low levels is crucial to maintaining quality control standards.

In addition to the advantage a fab enjoys from having around-the-clock analtyical support services on-site, an in-house analytical laboratory enables analysts and process or manufacturing engineers to jointly resolve problems. Whereas the exchange of information between fab line personnel and external laboratories is necessarily restricted because of the proprietary and confidential nature of semiconductor materials and processes, an on-site laboratory fosters the free flow of information about the nature of a problem, sample histories, and the design of the analytical approach, enhancing the potential for root-cause discovery.

Acknowledgments

The authors wish to acknowledge Henry F. "Buddy" Stehmeyer for his contributions in starting up the laboratory at the site. We also wish to thank Bert Woods for contributing to the graphical presentation of some of the data in this article.

References

1. Balazs pamphlet (Sunnyvale, CA: Balazs Laboratories).
2. RC Henderson, "Silicon Cleaning with Hydrogen Peroxide Solutions: A High Energy Electron Diffraction and Auger Electron Spectroscopy Study," Journal of the Electrochemical Society 119, no. 6 (1972): 772–775.
3. P Parimi and V Sundarsingh, "Auger Characterization of Silicon Surfaces Cleaned with H^₂O^₂ Based Solutions," in Proceedings of the Electrochemical Society 90, no. 9 (Pennington, NJ: The Electrochemical Society, 1990), 260–265.
4. D Burkman, "Optimizing the Cleaning Procedure for Silicon Wafers Prior to High Temperature Operations," Semiconductor International (July 1981): 103–107.
5. MG Yang, KM Koliwad, and GE McGuire, "Auger Electron Spectroscopy of Cleanup-Related Contamination of Silicon Surfaces," Journal of the Electrochemical Society 122, no. 5 (1975): 675–678.
6. A Licciardello, O Puglisi, and S Pignataro, "Effect of Organic Contaminants on Oxidation Kinetics of Silicon at Room Temperature," Applied Physics Letters 48, no. 1 (1986): 41–43.
7. K Hashimoto et al., "Gate Oxide Deterioration Caused by Organic Contamination onto the Oxide," in Extended Abstracts of the 1991 International Conference on Solid State Devices and Materials, 243.
8. P Sun, M Adams, and T Bridges, "Monitoring Organics on Wafer Surfaces Using Thermal Desorption GC-MSD/AED," MICRO 18, no. 3 (2000): 59–71.
9. M Wu and G Carr, "High-Purity Water Sample Concentrator Extends On-Line Silica Analysis to 6 ppt," Ultrapure Water 17, no. 2 (2000): 30–35.
10. Standard Methods for the Examination of Water and Wastewater, 17th ed. (Washington, DC: American Public Health Association, 1989), 4:184–4:187.
11. Ultra Pure Water Monitoring Guidelines 2000 for Facilities and Fabrication Engineers, 2d ed. (Sunnyvale, CA: Balazs Laboratories, 1999), 7–8.

T. Paul Adl, PhD, is team leader and serves as the technical director of the chemistry laboratory at Dominion Semiconductor in Manassas, VA. Previously he worked at IBM's analytical laboratories. Adl received a BS in chemistry from the University of Illinois (Champaign-Urbana) and a PhD in physical chemistry from Pennsylvania State University in University Park. (Adl can be reached at 703/396-1161 or tadl@dominionsc.com.)

Lawrence S. Turek is an analyst in the laboratory at Dominion Semiconductor, where he is responsible for monitoring the reactive silicate levels in site DI water. Previously he worked for IBM and Lockheed Martin. He received a BA degree in biology from the State University of New York in Oswego. (Turek can be reached at 703/396-1601 or lturek@dominionsc.com.)

Judy A. Montag is an engineer in the laboratory at Dominion Semiconductor, where she is responsible for analyzing trace contaminants in site DI water and the fab's air samples for anions, cations, and organic contaminents. For many years she worked in the University Environmental Science Laboratory at the University of Virginia in Charlottesville. She received a BS in science (with major in chemistry) from James Madison University in Harrisonburg, VA. (Montag can be reached at 703/396-1581 or jmontag@dominionsc.com.)

Terrence C. Leslie is the director of quality assurance at Dominion Semiconductor. Before joining the company, he spent several years as a process and development engineer and in management positions at IBM. He received a BS in electrical engineering from the University of Illinois (Champaign-Urbana), an MSEE from the University of Vermont in Burlington, and an MS in the management of technology from the Sloan School of Management at MIT in Cambridge, MA. (He can be reached at 703/396-1202 or tleslie@dominionsc.com.)

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