Critical Materials Delivery

Performing ICP-MS characterization of high-purity chemicals in fab delivery systems

Robert M. Woods, T. Paul Adl, and Terrence C. Leslie, Micron Technology

In advanced semiconductor fabrication, large volumes of ultra-high-purity chemicals are delivered to automated wafer-processing tools. While it is critical to verify chemical integrity, it is difficult to take samples in fabs because of double-walled piping, automated tool interlocks, and safety concerns. These obstacles can be overcome by sampling chemicals when they arrive at the fab and at various points in the chemical-distribution system.

As semiconductor devices shrink, roadmaps continue to drive down tolerable contamination levels.^1,2 Given this trend, quality control of chemicals and the determination of their integrity by analytical means is crucial. The reagent chemicals used in semiconductor manufacturing processes, however, can be a significant source of IC device failure.³ Dielectric breakdowns, stress effects, and yield issues can pose additional reliability problems.⁴

This article presents trace-level contamination data that were collected from virgin (unfiltered) and filtered high-purity chemicals in the high-use delivery system at Micron Technology's fab in Manassas, VA. In the interests of safety, the fab was built with a closed, double-walled chemical-delivery system that does not provide access ports for taking chemical samples. For quality assurance purposes, a compromise protocol has been established in which inductively coupled plasma¡Vmass spectrometry (ICP-MS) is used to analyze as-received, unfiltered chemicals drawn from the fab's chemical-distribution units (CDUs) against filtered samples from the day tank. At parts-per-billion and sub-parts-per-billion levels (the trace levels of interest), the delivery system maintains its integrity while preserving the purity of the incoming materials for point-of-use applications.

This article analyzes the data from this analysis and evaluates the quality control issues they raise.

Experimental Procedure

To test the chemical-delivery system, samples were taken from hydrochloric acid (HCl), hydrofluoric acid (HF), sulfuric acid (H₂SO₄), nitric acid (HNO₃), phosphoric acid (H₃PO₄), a 25:1 mixture of HNO₃ and HF, hydrogen peroxide (H₂O₂), choline, tetramethylammonium hydroxide (TMAH) developer, and various buffered oxide etchants. The sample analysis was done in Micron's chemistry laboratory at its Manassas, VA, fab shown in Figure 1.

Four 250-ml PFA bottles from Kurabo (Osaka, Japan) were cleaned for at least 12 hours using warm 10% nitric acid and then rinsed at least 10 times with DI water. The PFA tubing used to transfer the samples from the CDU port was soaked in a 25:1 mixture of HNO₃ and HF and rinsed several times with DI water. Before samples were collected, the chemicals were allowed to flow at a high rate for at least 2 minutes, after which the lines were purged.

All samples were analyzed using a 4500 ICP-MS from Agilent (Palo Alto, CA). The instrument was equipped with a radio-frequency (RF) shield plate, which, under low power, enables low-level (parts-per-billion and sub-parts-per-billion) analysis of potassium, calcium, and iron without forcing users to resort to alternative analytical methods. With the exception of zinc, barium, and cadmium, elements were analyzed under low RF power and cool plasma conditions (600–800 W).5 Zinc, barium, and cadmium were analyzed under hot plasma conditions (1200–1600 W).

The method of standard additions was used for all chemicals except HCl. The software used by the ICP-MS generates a calibration curve that is converted into an external calibration curve. This feature simplifies multisample analysis for each chemical by eliminating the need to prepare individual standard additions. HCl was performed using the traditional evaporation-to-dryness method, followed by resolvating the residue in 2% HNO₃. Optimum analytical conditions for detection limits and relative standard deviation were achieved using an inert kit and Agilent's Micro Flow nebulizer.

Results and Discussion

Figures 2–4 present ICP-MS data for 49% HF, TMAH, and H₃PO₄, respectively. In these charts, only the trace contaminants that were detected above the detection limits appear. Detailed data for these chemicals, together with detection limits, are shown in Tables I–III. The detection limits were normalized for the dilution factors whenever the "dilute and shoot" approach was used before standard additions.

Figure 2: ICP-MS analysis results showing trace contaminant levels of 49% HF.

Figure 3: ICP-MS analysis results showing trace contaminant levels of TMAH.

As the charts illustrate, the frequency of sampling and analysis was not identical for all chemicals. A judicious sampling scheme based on the consumption volume and nature of each chemical was developed. Chemically aggressive materials such as HF, H₂SO₄, HNO₃, and HF/HNO₃ acids were tested frequently. In its postfiltered state, highly corrosive HF was found to have a very high iron content. After root-cause analysis determined that the source of the iron contamination was a damaged coating in a stainless-steel component, the part was repaired and the integrity of the delivery line was restored.

The technique of analyzing pre- and postfiltration samples was used to discover other CDU-related problems, including proprietary etchant contamination and an HCl leak. It was found that the proprietary etchant system was contributing potassium, iron, nickel, and chromium to the chemical-distribution system. A root-cause analysis showed that the etchant was being contaminated by a variety of CDU components, such as valves, pumps, and conductivity meters. By performing pre- and postfiltration sampling, the problem was identified and addressed before the contaminated chemical could be delivered to the manufacturing floor.

The HCl leak was discovered in a sample port valve. After sampling the HCl, the lab noticed that HCl assays were considerably lower than the specification. Root-cause analysis indicated that an eroded valve was allowing water to seep into the HCl samples, generating false results.

While the data in Tables I–III indicate that all as-received and postfiltration chemicals were within specifications and that chemical integrity was maintained, contaminant levels varied. However, a statistical analysis of the data involving a paired Student's t-test and inspection of the probability values (P-values) for the 95% confidence limit invalidated the null hypothesis that there was a significant difference between pre- and postfiltration chemicals. The P-values for the t-test are provided in Table IV.

Figure 4: ICP-MS analysis results showing trace contaminant levels of H3PO4.

Although the t-test results suggested that the pre- and postfiltration chemicals did not differ from each other, it was important to probe further whether postfiltration samples showed signs of significant degradation. The ability to detect differences between pre- and postfiltration materials depends on the size of the difference that must be detected, the risk researchers are willing to take in not detecting it, the number of samples taken for each material, and the sample variation. These variables determine what is called the power of the t-test: the probability that a difference will be detected when it occurs. The power of each t-test was evaluated to calculate the size of the difference that each test had an 80% chance of detecting. The percentage of this difference versus the upper specification limit for a range of analytes in HF, TMAH, and H₃PO₄ are presented in Table V. Small percentages substantiate the conclusion that there was no difference between pre- and postfiltration chemicals.⁶

After the data had been collected, a retrospective study was performed to help the investigators understand the sensitivity of the tests. A power value of 80% was chosen as a reasonable minimum detection level that the sample sizes and error variances would be able to differentiate. By specifying the power value and sample size, statistical software release 13.20 from Minitab (State College, PA) was able to predict the minimum detectable difference.

The minimum detectable difference was then normalized to the upper specification limit for each analyte in each chemical under evaluation. Table V shows this normalization as a percentage of the upper specification limit for each element. As Table V illustrates, the sample size and error variance were sensitive enough to detect differences that approached the upper specification levels for the materials under evaluation.

Figure 5: Trace amounts of aluminum in 49% HF before versus after filtration.

The rationale for using the paired t-test is illustrated in Figure 5, which correlates pre- and postfiltration data for one trace element (aluminum) in one chemical (HF). Because relationships such as that shown in Figure 5 indicate that there are large amounts of lot-to-lot and sample-to-sample variability for the same chemical, and because the samples discussed here were not taken independently, the paired t-test was chosen for this statistical analysis.

Conclusion

The objective of this investigation was to evaluate an alternative sampling protocol for the quality assurance of process chemicals. The approach discussed here used samples from a fab's chemical-delivery system and overcame the problem posed by that system's inaccessible point-of-use ports.

The investigation found that the quality of incoming material could be determined without compromising the system; that the integrity of the chemical-distribution system components, including the day tank for storage of large quantities of chemicals, could be analyzed using the alternative sampling plan; and that the application of the alternative sampling protocol could uncover sources of critical contaminants in the delivery system and provide a means for performing root-cause analysis of potential product-reliability issues.

References

1. H Kitajima and Y Shiramizu, "Requirements for Contamination Control in the Gigabit Era," IEEE Transactions on Semiconductor Manufacturing 10, no. 2 (1997): 267–272.

2. The International Technology Roadmap for Semiconductors, Yield Enhancement section (San Jose: Semiconductor Industry Association, 2003); available from Internet: http://public.itrs.net.

3. "Sulphuric Acid for Semiconductors" (Geneva: Agilent Technologies Europe, 2001 [cited 1 March 2004]); available from Internet: http://www.laboratorytalk.com/news/agi/agi106.html.

4. WW Abadeer et al., "Key Measurements of Ultrathin Gate Dielectric Reliability and In-Line Monitoring," IBM Journal of Research and Development 43, no. 3 (1999): 407–416.

5. SD Tanner et al., "The Application of Cold Plasma Conditions for the Determination of Trace Levels of Fe, Ca, K, Na, and Li by ICP-MS," Atomic Spectroscopy 16, no. 1 (1995): 16–18.

6. R Kenett and S Zacks, Modern Industrial Statistics: Design and Control of Quality and Reliability (Pacific Grove, CA: Duxbury Press, 1998).

Robert M. Woods is a staff chemist at Micron Technology's facility in Manassas, VA. He has been the trace metals and ICP-MS expert at the facility for seven years. He received a BS in chemistry from Radford University (Radford, VA) in 1997. (Woods can be reached at 703/396-1229 or rwoods@micron.com.)

T. Paul Adl, PhD, is the manager of the chemistry laboratory at Micron Technology's Manassas, VA, fab. Previously he worked at IBM's analytical laboratories. He 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@micron.com.)

Terrence C. Leslie is the department manager of wafer final test (probe), quality, and reliability assurance and the chemistry laboratory at Micron Technology's fab in Manassas, VA. Before joining the company, he was a senior engineer and manager in semiconductor development and manufacturing at IBM and vp of quality assurance at Dominion Semiconductor. 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 the Massachusetts Institute of Technology in Cambridge. (Leslie can be reached at 703/396-1202 or tleslie@micron.com.)