SURFACE CHEMISTRIES

Characterizing organic impurities in semiconductor-grade hydrogen peroxide

Stefan Huber, DOC-Labor; and Dietmar Oeter, Merck

As the semiconductor industry moves toward the 256-Mb chip generation with 0.15-µm feature sizes on 300-mm wafers, wet chemical cleaning remains essential to the removal of particles and metallic and organic impurities from the wafer surface. Since semiconductor-grade hydrogen peroxide is the process chemical most commonly used in wet cleaning systems, the most prevalent cleaning method, the hydrogen peroxide—based RCA standard clean, will continue to be the state of the art.¹

While there is much information about the impact of cationic impurities on yield, little is known about the effect of organic impurities in high-purity hydrogen peroxide used in wet cleaning processes. The amount of organic contaminants in hydrogen peroxide is well within the parts-per-million concentration range, and total organic carbon (TOC) values around 15 ppm are considered acceptable by many users. Organic components, in particular hydrophobic organic components, are assumed to have adverse effects on yield,² but systematic investigations on this subject have not yet been published. In one experimental study, a "special quality" hydrogen peroxide with a TOC content of <3 ppm was used, and no detrimental effect to oxide film formation was observed.³ No results were disclosed for standard semiconductor qualities.

This article examines the amount and composition of organic components that can be found in semiconductor-grade hydrogen peroxide. In the study presented here, one customized and two standard analytical techniques were used to analyze the TOC content in products of four different suppliers. Identical lots were measured over a 20-month period to assess the impact of sample aging. In an earlier study, it was determined that TOC in hydrogen peroxide is a heterogeneous composition of very different organic components.⁴ These components originate either from refinery processes of raw products or from the working solution used in the production process (autoxidation process).⁵

TOC Measurement Methods

Several sample pretreatment approaches may be used to identify and measure TOC content in hydrogen peroxide. The classical method involves catalytic decomposition of hydrogen peroxide in a platinum vessel, followed by high-temperature combustion of the organic residues and nondispersive infrared detection of the oxidation product carbon dioxide. A common instrument is the Model 5000 TOC analyzer from Shimadzu (Kyoto, Japan), which was used in these experiments by researchers at Merck (Darmstadt, Germany). Another standard oxidation method involves thermal (50°—100°C) or catalytic (UV) wet oxidation with sodium persulfate as oxidant.^2,6 Other sample pretreatment techniques, including sample dilution to a hydrogen peroxide concentration of 5% or direct injection of the sample, may be used.⁶

A relatively new chromatographic method has been developed by DOC-Labor (Karlsruhe, Germany), which involves connecting a liquid chromatograph (LC) to a custom-made organic carbon detector (OCD).^4,7 In addition to determining TOC content, this LC-OCD technique provides information on the qualitative composition of TOC. Original samples of hydrogen peroxide are injected onto a gel chromatographic column and analyzed by UV detection and organic carbon detection. The OCD is an infrared detector connected to a special thin-film reactor in which all treatment steps required for TOC analysis (acidification, first purge, oxidation, second purge) are performed on-line and quasi-simultaneously. Unlike the aforementioned standard methods, this customized approach involves neither hot combustion nor catalytic wet combustion. Instead, the organic material is oxidized by radicals produced by the radiolytic decomposition of water in a nitrogen atmosphere.

Since the LC-OCD approach cannot identify organic components, the question, What is in the sample? cannot be answered. With this method, however, the question, Is this or that component in my sample? can be answered with a standard chromatogram of the component. Basically, all organic components in the sample can be quantified on the basis of carbon masses. After calibration of the LC-OCD system, the area covered by the chromatogram refers to the TOC value. The specific retention times of the components, or the classes of the components, express important chemical properties used for interpretation (e.g., molecular weight, hydrophobicity, acidity). The following classes of organic components can be quantified in hydrogen peroxide: high-molecular-weight (HMW) components (>20,000 g/mol, semiquantitative), acids, hydrophilic components, and hydrophobic components.

The hydrophobic fraction is the portion of TOC that elutes beyond the total volume of the stationary phase (a polymer). Increasing retention time reflects increasing hydrophobicity. The matrix hydrogen peroxide is separated from most of the TOC during the chromatographic run. Cross-sensitivity may occur only for coeluting components, which make up a minute fraction of TOC. Ablution of organic components from the polymer column ("column bleeding") as a potential source of error in the LC-OCD method can be considered negligible. Hydrogen peroxide samples of a special quality were analyzed and were found to be almost free of organic impurities. No indications of column bleeding were seen in these samples.

Different analytical techniques should lead to identical results if the techniques are formally correct, yet various studies have shown that this goal cannot be achieved when determining TOC content in hydrogen peroxide.^2,6 For example, the hot combustion method is very powerful with respect to the oxidation step, but large volumes of water vapor and oxygen are produced that influence the residence time, hence the concentration of the analyte (carbon dioxide) in the infrared cuvette. This systematic error has to be leveled out in the calibration function. Little is known about the systematic loss of TOC induced in high-temperature combustion methods. The addition of acids (formic, oxalic) to hydrogen peroxide was seen to lead to incomplete recoveries. Such acids may be oxidized by hydrogen peroxide at the entrance of the combustion cell before the proper oxidation process on the catalyst surface starts. Other systematic errors occur in sample pretreatment when the peroxide is thermally decomposed. This will be discussed further in this article. Overall, the LC-OCD method was found to provide the most accurate results, because the chromatographic process separates the problematic hydrogen peroxide matrix prior to analysis. Also, samples are directly injected and analysis is performed in an on-line system in which no changes in temperature, pressure, and volume take place.

Experimental Design and Procedures

In June 1996 we conducted our initial experiment, Investigation I. Four representative products (A—D) of semiconductor-grade hydrogen peroxide (31%) from four suppliers were analyzed, using the catalytic decomposition method and the LC-OCD method. The same products were analyzed again in October 1996 to assess the influence of product aging. All products originally had been packed in polyethylene bottles. For the aging experiments, samples were stored in glass flasks in the dark at 4°C. In July 1997 we began Investigation II, using new samples of fresh lots of products A—C. In addition to the measurements taken in Investigation I, diluted samples of hydrogen peroxide (5%) were analyzed using the TOC analyzer and LC-OCD system. These measurements were repeated in February 1998 to assess sample aging effects.

For measurements taken using the catalytic decomposition method, samples of 50 ml were decomposed in a platinum vessel in a hermetically sealed system for 2 hours. The temperature was kept below 5°C. The injection volume was 30 µl. As described above, the TOC analyzer that was used is based on the high-temperature combustion technique. For the dilution method, samples were diluted with ultrapure water to a hydrogen peroxide concentration of 5%. The injection volume was 30 µl. The same TOC analyzer was used. For both methods, a relative standard deviation of 2.0% was determined. For the LC-OCD method, 200 µl in Investigation I (400 µl in Investigation II) of the original samples were injected into the chromatograph. The relative standard deviation for this method was determined once (five measurements) and a value of 2.8% was found.

Results and Discussion

As shown in Table I, the TOC values determined in this study were well within the parts-per-million concentration range. TOC content for the catalytic decomposition method ranged from 6 to 25 ppm (Investigation I, fresh lots of June 1996). Regarding TOC content, product A yielded the best results, followed by C, D, and B. Results obtained with the LC-OCD method showed the same ranking, but values were significantly higher, ranging from 11 to 27 ppm. These differences in TOC values were highly product specific. The TOC content of product A was about twice as high in the LC-OCD analysis.

Product	LC-OCD	TOC Analyzer
		Dilution to 5%	Catalytic Decomposition
Investigation I:
June 1996
A	11.9	—	5.9
B	32.9	—	25.3
C	17.7	—	13.8
D	26.7	—	18.4
October 1996
A	11.2	—	6.0
B	24.8	—	24.2
C	13.4	—	13.3
D	19.6	—	20.7
Investigation II:
July 1997
A	12.9	10.3	7.5
B	24.4	19.3	18.0
C	19.1	17.1	14.3
February 1998
A	11.2	9.9	5.9
B	22.5	19.2	20.3
C	16.3	17.7	13.6

As for the differences in TOC values, the aged products measured in October 1996 showed a similar behavior for product A. For products B, C, and D, the differences are no longer significant. Obviously, the qualitative composition of TOC must have changed during storage, and hence the recovery of TOC in at least one of the analyzers. Fresh products B—D might have contained unstable organic components that were eliminated in the course of catalytic sample pretreatment. In aging processes, these unstable components were eliminated by slow autoxidation of the peroxide, and thus, there were no significant differences between the TOC analysis methods. In the case of product A, these unstable components may have been specific acids that were formed and destroyed to varying extents during sample pretreatment and aging.

Similar trends were found in Investigation II of fresh lots of products A—C, but LC-OCD gave slightly higher values for the aged samples. Again, product A was the exception, exhibiting much lower TOC values for the hot combustion method. The dilution method, which was only applied in this investigation, showed TOC values that were higher than those obtained with the catalytic decomposition method, but that were still lower than the comparative LC-OCD results. If LC-OCD is assumed to be the more precise method, then data obtained with the dilution method were also too low, but to a lesser extent. A straightforward explanation for lower values for the dilution method is cross-sensitivity of activated oxygen in the infrared cuvette, which leads to a depression of the baseline and hence to lower readings. This depression is observed constantly in LC-OCD chromatograms (see Figure 1). This adverse effect of hydrogen peroxide in samples most likely would be even more pronounced if nondiluted samples were injected.

Figure 1: Chromatograms of products A—D of June 1996.

LC-OCD analysis shows that aging consistently leads to slightly lower TOC values. The other two TOC analysis methods do not show a clear trend, in fact, some products even show an increase in TOC with aging. Obviously the accuracy of these methods is not sufficient to identify small changes in TOC content. Significant differences between lots can be observed too. In the lots used for Investigation II, products A and C had slightly higher TOC content, but product B was much lower in TOC. Variability in TOC content between different lots thus can be quite significant.

Qualitative TOC Analysis Using LC-OCD

The impact of sample pretreatment relates to differences in the chemical composition of TOC. This can be visualized with the LC-OCD technique. In Figure 1, the chromatograms of products A—D are shown for the investigation conducted in June 1996. The results are summarized in Table II. The following classes of chemical components can be quantified: high-molecular-weight components (>20,000 g/mol), hydrophilic neutrals (e.g., alcohols, esters), carboxylic acids (monoprotic to triprotic), and hydrophobic components. In all products A—D, the hydrophobic fraction is the dominant fraction. Product B shows the highest amount of hydrophobic material. As most of this material elutes as a prominent, late-eluting peak at about 125 minutes, the hydrophobic fraction is stronger hydrophobic (>C6 alcanol) compared to the other products. Product A has the lowest amount of hydrophobic components, both from quantitative and qualitative points of view. It is certainly no coincidence that this product showed the strongest discrepancies in the different TOC methods. Since less oxygen is required to oxidize hydrophilic components to acids, more oxygen was available during catalytic pretreatment for the complete oxidation of some of the TOC to carbon dioxide.

Total TOC	Product	HMW	Hydrophilic Neutrals	Acids	Hydrophobics	Hydrophobics as % of TOC
A	11.9	0.002	4.1	0.6	7.8	65
B	32.9	0.011	4.5	1.2	28.4	86
C	17.7	0.014	5.9	1.9	11.8	66
D	26.7	0.072	6.8	3.3	19.9	74

The impact of catalytic pretreatment on the change in TOC content is evident in LC-OCD chromatograms. For example, the chromatogram of Product B before and after catalytic decomposition is shown in Figure 2 (October 1996). The hydrophobic compound at 125 minutes was destroyed to a large extent (—8 ppm) while the increase in acids was minimal (+0.5 ppm). As shown by the extreme loss in TOC, most acids formed in the process must have been oxidized to carbon dioxide, while some material was transformed into other hydrophilic components.

Figure 2: Impact of catalytic decomposition of hydrogen peroxide as pretreatment on TOC (Product B, October 1996).

In general, aging has a positive impact on the qualitative composition of organic impurities, as shown in Figure 3 (Investigation I). Organic acids have been formed (+1 ppm) and the dominant hydrophobic fraction at 125 minutes has decreased significantly (—5 ppm). Autocatalytic decomposition of minimal amounts of hydrogen peroxide is assumed to oxidize TOC at a very slow rate, leading to the formation of organic acids (e.g., oxalic, formic).⁶ Depending on the amount and kind of acids, some of these acids will be totally oxidized to carbon dioxide. For the characterization of organic impurities in hydrogen peroxide, the determination of organic acids is of little value. The quantification of acids, rather, may serve as an indicator for the age of a given lot. It was also found that an impact of aging is highly product specific. In one sample quality, no effect of aging (for 6 months) was found.

Figure 3: Impact of aging on TOC (Product B, June and October 1996).

As shown in Figure 1, resolution of the chromatographic fractions in LC-OCD is still poor because baseline resolution is not achieved for all fractions. As a consequence, the heterogeneity of organic compounds in all samples must be high, much higher than the resolution power of the column used (the number of theoretical plates is around 5000). It was concluded that TOC in all products is composed of many different components, likely more than 20. On the other hand, the elution patterns are very specific. Therefore, LC-OCD chromatograms can be used as fingerprints to identify not only production processes but also producers.

Conclusion

Organic impurities in four representative products (A—D) of semiconductor-grade hydrogen peroxide (31%) from four different suppliers were analyzed for TOC content. TOC was determined with the hot combustion method, and samples were either diluted to a hydrogen peroxide concentration of 5% or the peroxide was eliminated by catalytic decomposition. Results from these standard analysis techniques were compared with results of the customized LC-OCD method, in which TOC is chromatographically separated and then detected by organic carbon detection. In contrast to the other methods, the LC-OCD technique provides additional information on the qualitative composition of TOC. Identical lots were measured over an 18-month period to assess the impact of sample aging.

All four products showed TOC content well within the parts-per-million concentration range. Results of the LC-OCD analysis indicated that catalytic decomposition of the peroxide prior to analysis, sample dilution, and other sample pretreatment may result in TOC values that are obviously too low. The prominence of this effect depends not only on the product-specific composition of the organic impurities in the samples, but also on their age.

LC-OCD analysis shows that the TOC content of all products is made up of a very heterogeneous mixture of likely more than 20 different components. Most TOC exhibits hydrophobic properties as indicated by the affinity with the chromatographic column. However, the amount and the degree of hydrophobicity varies greatly from product to product.

Acknowledgments

Part of this work was published previously.⁷ The authors gratefully acknowledge the permission of Balazs Analytical Laboratory (Sunnyvale, CA) for reproducing those materials.

References

1. Kern W, and Puotien DA, "Cleaning Solutions Based on Hydrogen Peroxide for Use in Semiconductor Technology," RCA Review, 31(2):187—206, 1970.

2. Camenzind MJ, and Balazs MK, "Analysis of Organic Impurities in Semiconductor Processing Chemicals," in Chemical Proceedings of the Semiconductor Pure Water and Chemical Conference (SPWCC), Sunnyvale, CA, Balazs Analytical Laboratory, pp 143—161, 1992.

3. Kimura K, Ogata Y, Tanaka F, et al., "Study on an Influence of TOC in Hydrogen Peroxide for Advanced Wet Chemical Processing," in Proceedings of the Symposium on Contamination Control and Defect Reduction in Semiconductor Manufacturing—I, Pennington, NJ, The Electrochemical Society, pp 347—360, 1992.

4. Huber SA, and Frimmel FH, "Flow Injection Analysis of Organic and Inorganic Carbon in the Low-ppb Range,"Analytical Chemistry, 63:2122—2130, 1991.

5. Ullmann's Encyclopedia of Industrial Chemistry, 5th ed, vol A13, Elvers B (ed), Weinheim, Germany, Vertagsgesellschaft, p 443, 1989.

6. Oberson SR, "Examining Organic Impurities in Ultrapure Hydrogen Peroxide," MICRO, 13(6):67—71, 1995.

7. Huber SA, Oeter D, Dusemund C, et al., "Quantification and Characterization of Organic Impurities in Semiconductor-Grade Hydrogen Peroxide with a Direct Chromatographic Method," in Chemical Proceedings of the SPWCC, Sunnyvale, CA, Balazs Analytical Laboratory, pp 15—30, 1997.

Stefan Huber, PhD, is president of DOC-Labor (Karlsruhe, Germany), which he founded in 1995 to offer TOC characterization to the water and power generation industries. He earned an MS in geology at the University of Heidelberg and completed his PhD at the Engler Bunte Institute at the University of Karlsruhe with the development of the LC-OCD system detailed in this article. Huber's LC-OCD system was honored by the federal state of Baden Württemberg, and he also received a PhD award from the Water Chemistry Association of the German Chemical Society. (Huber can be reached at +49 721 69 5940 or 100741.1715@compuserve.com.)

Dietmar Oeter, PhD, is a process manager at Merck (Darmstadt, Germany), where he has worked since 1995. He is responsible for the quality assurance of process chemicals including A-clean and highest-purity hydrogen peroxide. He earned an MS in physical chemistry from the Technical University of Karlsruhe and completed his PhD in chemistry at the University of Tübingen, specializing in investigating -oligothiophenes as model substances for semiconductor technology. (Oeter can be reached at +49 6151 72 4834 or dietmar.oeter@merck.de.)

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