ANALYSIS AND METROLOGY—SURFACE CONTAMINATION

Using VPD ICP-MS to monitor trace metals on unpatterned wafer surfaces

Greg Settembre, IBM Analytical Services; and Ebenezer Debrah, Perkin-Elmer SCIEX

Metallic contamination must be carefully monitored throughout the semiconductor manufacturing process because levels of 1 x 10¹⁰ atoms/cm² on a silicon wafer's surface can adversely affect manufacturing yields and cause device failures.^1—3 One of the most prevalent analytical methods used in such monitoring is vapor phase decomposition (VPD) sample preparation.^4—6

The VPD technique concentrates the distributed impurities on a wafer's surface into one liquid droplet. This is accomplished by exposing an unpatterned wafer (containing a native or thermal oxide layer) to hydrogen fluoride vapor in a closed, chemically inert container, which results in the decomposition of the oxide into HF/H^₂O droplets that contain the metallic contaminants, which are then distributed across the wafer's remaining hydrophobic surface. A small amount of ultra-high-purity scanning collection solution (SCS) is then used to accumulate all the droplets into one sample extract, with final volumes ranging from 500 to 1000 µl. Once prepared, the liquid sample is analyzed by instrumental techniques for identification and quantification of the sample constituents.

VPD sample preparation has been coupled with several different trace elemental analytical techniques, such as atomic absorption spectrometry (VPD-AAS), total reflection x-ray fluorescence (VPD-TXRF), and inductively coupled plasma mass spectrometry (VPD-ICP-MS). Sample preparation has traditionally been performed manually; this article presents results obtained from a semiautomated sampling and instrumental analysis methodology.

Sample Preparation

Manual collection of the droplets from the wafer's hydrophobic surface is one of the problems with the traditional sample preparation methods. A vacuum pencil holds the wafer, with the droplets collected by scanning the wafer with the collection solution. This scanning process consists of the systematic movement of the SCS droplet across the entire wafer surface. The analyst must be skillful in order to achieve proper collection and complete wafer surface coverage. Inherent problems include sample preparation precision, total surface coverage, wafer-edge contamination, and actual sample loss. With the advent of 300-mm wafers, manual scanning of the wafer surface will become more difficult. There are several commercially available scanners in a variety of configurations ranging from semiautomatic to fully automatic. The scanners are capable of the controlled movement of an SCS droplet across the entire wafer surface. The water droplets formed by VPD coalesce into one droplet of about 500—1000 µl, which is then transferred for instrumental analyses.

End-users must decide on the degree of automation they require. Another critical design feature is the use of chemically inert high-purity polymer construction materials, in order to prevent sample contamination and withstand the harsh chemical environments of laboratories or semiconductor fabs. Regardless of the scanner's configuration, there must be a mechanism for the quantitative and contamination-free transfer of the SCS into an autosampler vial for subsequent instrumental analysis. Although the sampling procedure is important, a sensitive and reliable analytical technique must also be available for the determination of the trace element contaminants.

Instrumental Analysis

ICP-MS is an important analytical tool for the semiconductor industry.⁷ This sensitive, rapid, multielement atomic spectroscopic technique offers ultralow detection capability, high sample throughput, wide dynamic range, wide elemental coverage, rapid semiquantitative analysis, and the ability to provide isotopic information.⁸ With its relatively simple spectrum, the spectrometer is rugged, reliable, and stable and can be used to run a wide variety of sample matrices. Traditionally, ICP-MS has not met the requirements for the determination of all critical elements found in semiconductor applications because of its inability to measure calcium, potassium, and iron at the required ultratrace concentration levels. This shortcoming is primarily caused by the formation of ⁴⁰Ar⁺, ³⁸Ar¹H⁺, ⁴⁰Ar¹⁶O⁺, and other plasma-derived interferences, which results in elevated background levels for these critical elements. It has been shown that the argon and argon polyatomic levels can be significantly reduced by operating the ICP-MS under cold plasma conditions (i.e., low ICP power and high nebulizer gas-flow rate). However, this technique can pose problems when a large number of elements must be determined in a limited amount of sample, such as in the case of VPD samples, where the typical size is 500—1000 µl. Cold plasma conditions lower the plasma temperature, which drastically reduces the formation of argon-based molecular species and allows the determination of elements that otherwise could not be determined at ultratrace levels with ICP-MS. An automated run can be set up seamlessly to determine both cold plasma elements (e.g., K, Ca, Fe) and normal plasma elements (e.g., W, Mo, Cu) in the same sample, using a microconcentric nebulizer (MCN) optimized for low sample volumes.⁹

However, with the very small volumes associated with a VPD extract, there is a fine balance between elemental coverage and detection capability. In addition, because the available ionization energy is much lower in a cold plasma, matrix interferences and polyatomic spectral interferences from the silicon matrix of the VPD sample can be a potential problem. For that reason, a combination of cold and normal plasma, even with a nebulizer optimized for small samples, may not be the best approach for VPD extracts. This is because the successful microsampling analysis of such extracts requires HF inertness, minimum matrix/polyatomic interferences, sub-parts-per-billion quantification, and expanded elemental coverage.

A membrane desolvator sampling accessory can be used to overcome the limitations of a conventional MCN. Since the desolvator removes most of the solvent, very little reaches the plasma, minimizing the formation of polyatomic species that adversely affect the determination of ultratrace elements. This system, shown in Figure 1, uses an MCN with a membrane desolvation system coupled to the ICP-MS. A sweep gas passed outside the microporous membrane removes the solvent in the sample. This approach not only leads to the elimination of the solvent but also to reduced polyatomic species formation, enabling lower detection limits to be obtained. In addition, the problematic silicon fluoride matrix of the VPD sample, when heated in the PTFE spray chamber, becomes volatile and passes through the membrane, never reaching the mass spectrometer.

Figure 1: Schematic of a microconcentric nebulizer attached to the ICP-MS.

The desolvator, used together with normal plasma operating conditions, reduces or eliminates the formation of many polyatomic spectral interferences that affect the determination of iron, calcium, potassium, and other critical semiconductor elements. The major benefit of combining microsampling membrane desolvation with ICP-MS, however, is that less sample is used because all elements are determined with a single set of ICP-MS plasma conditions. This means more time is available to either extend the elemental coverage or increase the measurement time to improve detection capability.

The remainder of this article presents data for the determination of a suite of elements using a commercially available MCN with a membrane desolvation system and a quadrupole ICP-MS. The optimization protocol is discussed as well as a demonstration of the use of a membrane desolvator in combination with normal plasma conditions as an alternative to other analyses of wafer surface extracts.

Experimental Procedures

All sample and standard preparation was done in a Class 100 cleanroom using high-purity acids. Sample and standard preparation, storage, and analysis containers were made of high-purity Teflon and were precleaned. Instrumental analysis was performed using the Elan 6000 ICP-MS (Perkin-Elmer SCIEX Instruments, Concord, ON, Canada), with the MCN-6000 microconcentric nebulizer (Cetac Technologies, Omaha, NE) as the sample introduction system. A small amount of nitrogen gas bled into the MCN system at a low flow rate enhanced sensitivity and reduced polyatomic formations. Table I lists the ICP-MS conditions used, while Table II shows those of the MCN.

RF power	1350 W
Nebulizer gas-flow rate	0.9 L/min
Plasma gas-flow rate	15 L/min
Ion lens voltage	6.5 V
Integration time	2625­3500 msec
Number of sweeps	75
Quadrupole dwell time	35­50 msec
Quadrupole settling time	200 µsec
MCA channels per peak	1
Quadrupole scan mode	Peak hopping

Table I: Operating conditions for the ICP-MS.

Spray chamber temperature	80°C
Desolvator temperature	165°C
Sweep gas-flow rate	3.5 l/min
Nitrogen flow rate	34 ml/min
Sample uptake	80­110 µl/min

Table II: Microconcentric nebulizer operating conditions.

A 500-µl sample lasts only 5 minutes when using an MCN sample flow rate of 100 µl/min. For this reason, the ICP-MS measurement protocol must be optimized to ensure that the maximum amount of time is spent quantifying the analyte peaks. In a separate experiment not discussed here, a dwell time of 35—50 milliseconds, depending on the element, and a quadrupole settling time of 200 microseconds using 1 point/peak were found to be optimum. The number of sweeps of the quadrupole depended on the integration time used for each mass. This measurement protocol guaranteed that the most efficient duty cycle was used to achieve the optimum detection limits and maximum elemental coverage in the 5-minute duration of the sample signal.

Following an initial warm-up of the MCN and ICP-MS prior to sample analysis, the membrane was conditioned with a prolonged nebulization of a dilute nitric acid solution. Because of its similar ionization properties to ⁵⁶Fe (one of the critical elements), a 1-ppb cobalt solution was used to optimize the system for acceptable performance criteria. Key performance indicators include intensity and precision measurements for ⁵⁹Co and background masses ³⁹K, ⁴⁴Ca, and ⁵⁶Fe, in addition to the mass 59:mass 56 ratio. An internal standard of ¹¹⁵In was used in the measurements.

Results and Discussion

After successful system optimization, calibration plots were generated using a blank as well as 50-, 100-, 500-, and 1000-ppt standards for 20 different elements. Figures 2 through 4 show typical calibration plots for ⁴⁴Ca, ⁵⁶Fe, and ³⁹K, while Table III lists typical calibration statistics for additional elements. As the calibration statistics show, the linearities of the plots are very good, indicating that the ICP-MS and the desolvation apparatus are both stable. These data demonstrate that the combination of the two systems can determine the critical elements in small-volume semiconductor samples using one set of operating conditions. Previous attempts to analyze such a small volume with a wide elemental coverage would have been difficult because the plasma conditions had to be changed to determine both cold and normal plasma elements.

Figure 2: Calibration plot for the determination of ⁴⁴Ca.

Figure 3: Calibration plot for the determination of ⁵⁶Fe.

Figure 4: Calibration plot for the determination of ³⁹K.

Element	Slope [A]	Intercept [B]	Sigma A	Sigma B	R2
Ca	2.0440 x 10—4	—4.8226 x 10—6	3.8717 x 10—7	2.1751 x 10—4	0.9999
Al	1.3905 x 10—4	—4.0038 x 10—3	6.5812 x 10—7	3.6973 x 10—4	0.9999
W	6.5194 x 10—5	—2.9683 x 10—4	2.9422 x 10—7	1.6529 x 10—4	0.9999
Fe	3.2000 x 10—4	—3.3337 x 10—4	3.4867 x 10—7	1.9588 x 10—4	0.9999
K	3.5090 x 10—4	—1.5180 x 10—4	2.9630 x 10—7	1.6646 x 10—4	0.9999
Co	2.0440 x 10—4	—4.8226 x 10—6	3.8717 x 10—7	2.1751 x 10—4	0.9999

Table III: Calibration statistics (y = Ax+B) for expanded element set.

Figure 5 shows data obtained over a two-month period for the mass 59:mass 56 ratio, which indicates the integrated system's performance. Typical relative standard deviation (RSD) values of <5% for ⁵⁶Fe and <1.5% for ⁵⁹Co were obtained. The day-to-day fluctuations for the 59:56 ratio were caused by instrumental system changes (changing of sample introduction systems, preventive maintenance, and so forth). Even with these changes, acceptable performance levels were obtained, demonstrating the system's overall ruggedness.

Figure 5: A plot of the ratio of 100-ppt ⁵⁹Co to mass 56 background over a 2-month period shows the system is in control. Also shown is a plot of the ratio of mass 59 to mass 56, which demonstrated long-term stability.

Figure 6: Stability for 100-ppt ⁵⁶Fe, ³⁹K, and ⁵⁹Co of the microconcentric nebulizer system on the ICP-MS over approximately 2 hours.

Table IV shows the short-term precision, represented by the percentage RSD values of three replicate measurements, for a selected group of elements using the integrated system. Figure 6 plots the intensities for ⁵⁶Fe, ³⁹K, and ⁵⁹Co from 10 replicates of a 100-ppt standard, taken over a period of approximately 2 hours, which indicate RSDs of 1.1%, 1.7%, and 3.6%, respectively. The data demonstrate the stability of the combined systems. Elements that previously required both normal and cold plasma conditions can now be determined with a single set of plasma conditions, a significant advantage for semiconductor applications with sample size limitations.

Element	% RSD
27Al	4.3
39K	3.5
56Fe	1.6
59Co	2.1
184W	1.7

Table IV: Short-term precision of three replicates per determination for 100 ppt of some selected elements.

The wafers submitted for analysis by VPD ICP-MS typically contain significantly different thicknesses of silicon oxide layers. Oxide layers may range in thickness from native (~50 Å) to thermally grown (up to ~5000 Å). Depending on the thickness, VPD extracts contain different concentrations of dissolved silicon, with typical levels somewhere between a few hundred and a few thousand parts per million of silicon. Table V details the results (percent recoveries) obtained from the analysis of VPD extracts. Each of the four wafers, varying in oxide thickness from ~800 to 2500 Å, were spiked with a multielement standard, resulting in the addition of 250 ppt to the extracts. The recoveries are the mean of three replicate measurements. The results show that the combined system can obtain very good recoveries, even when varying levels of silicon are found in the VPD extracts.

Element	Wafer 1	Wafer 2	Wafer 3	Wafer 4
23Na	85	91	85	94
64Zn	86	93	86	88
56Fe	98	94	96	97
39K	100	86	114	102
44Ca	97	80	127	96
59Co	100	98	96	97
48Ti	96	101	91	95
98Mo	99	100	94	97
27Al	87	79	74	84

Typically, for the most critical process steps, surface metallics levels are close to the reportable values of the analytical method. Table VI presents conservative reportable limits (RLs) as atoms per cubic centimeter for a 200-mm wafer surface and as parts per trillion (nanogram per liter) in solution. These RLs meet the data quality objectives of the test laboratory where the study was conducted and can be easily achieved on a daily basis. Parameters that affect RLs include ambient contamination levels, accuracy of calibration, elemental coverage, and final sample extract volume. However, these are not the absolute limits of detection achievable. Based on the equivalent LOQ in solution, the limit of detection will be approximately 5 to 50x better, depending on the element.

Element	RL (1 x 1010 atoms/cm2) for 200-mm wafer	Equivalent to RL in solution (ppt)
Cu	0.2	50
Na	0.8	100
Ni	0.3	100
Ca	0.9	200
K	0.5	100
Fe	0.3	100
Cr	0.4	100
Al	0.7	100
Mo	0.2	100
Zn	0.3	100
Ti	0.4	100
Co	0.3	100
W	0.1	100

Table VI: Reportable limits (RL) for a 200-mm wafer and solution equivalents.

Conclusion

The routine determination of trace metal contaminants on wafer surfaces by VPD ICP-MS without the need for cold plasma conditions has been demonstrated by the use of a membrane desolvator and a fast scanning quadrupole ICP-MS. By optimizing the instrument measurement protocol, one set of ICP-MS operating conditions allows the determination of up to 20 elements, including such critical semiconductor elements as iron, calcium, and potassium, in a VPD sample. Reportable limits of approximately 1 x 10⁹ atoms/cm² on a 200-mm wafer can be achieved with good accuracy and precision. In addition, because the silicon fluoride matrix volatilizes inside the heated chamber of the nebulizer, most of it passes through the membrane and never reaches the ICP-MS. For this reason, good spike recoveries can be expected, even when carrying out VPD analysis on wafers with oxide layers up to 2500 Å thick.

References

1. Shimazaki A, Hiratsuka H, et al., "Chemical Analysis of Ultrapure Impurities in SiO^₂ Films," in Extended Abstracts of 16th Conference on Solid State Devices and Materials, Tokyo, Japanese Society of Applied Physics, 281—284, 1984.

2. Shiraiwa T, Fujino N, Sumita S, et al., "Chemical Analysis of Metallic Impurity on the Surface of Silicon Wafers," Semiconductor Fabrications Technology and Metrology, ASTM STP 990, 1989.

3. Radshaw N, Hall EF, Sanderson NE, et al., "Inductively Coupled Plasma as an Ion Source for High-Resolution Mass Spectrometry," Journal of Analytical Atomic Spectrometry, 4:801, 1989.

4. Ohsawa A, Honda K, and Toyokura N, "Metal Impurities Near SiO^₂—Si Interface," Journal of the Electrochemical Society, 131:2964—2969, 1984.

5. Honda K, Nakanishi T, Ohsawa A, and Toyokura N, "Catastrophic Breakdown in Silicon Oxides: The Effect of Fe Impurities at the SiO^₂—Si Interface," Journal of Applied Physics, 62:1960—1963, 1987.

6. Shimono T, Tsuji M, Morita M, and Muramatu Y, "Device Degradation by Metallic Contamination, and Evaluation and Cleaning of Metallic Contaminants," in Proceedings of Microcontamination 91 Conference, Santa Monica, CA, Canon Communications, pp 544—551, 1991.

7. Balazs M, "Semiconductor Industry Benefits from ICP-MS," R&D, 51—56, 1995.

8. Houk RS, "Mass Spectrometry of Inductively Coupled Plasmas," Analytical Chemistry, 58:97A—104A, 1986.

9. Debrah E, Beres SA, Gluodenis TJ, and Thomas RJ, "Benefits of a Microconcentric Nebulizer for the Multielement Analysis of Small Volumes by Inductively Coupled Mass Spectrometry," Atomic Spectroscopy, 16:197—202, 1995.

Greg Settembre is a staff scientist for IBM Analytical Services in East Fishkill, NY. He is responsible for the development and implementation of ICP-MS analytical methods in support of contamination control investigations. He has more than 15 years of expertise in the area of trace organic and inorganic analyses. Settembre has a BS in environmental chemistry from Marist College (Poughkeepsie, NY). (Settembre can be reached at 914/894-7302 or settembr@us.ibm.com.)

Ebenezer Debrah, PhD, is a scientist at Perkin-Elmer SCIEX, Concord, Ontario, Canada. He is responsible for accessories support and is also involved in new product development for the ICP-MS line. Previously, he worked at Perkin-Elmer (Wilton, CT) in the ICP worldwide marketing department. Debrah has a BS from the University of Ghana, an MS from the University of Technology (Loughborough, UK), and a PhD in analytical chemistry from the University of Massachusetts (Amherst). He has authored more than 40 articles and technical presentations in the area of analytical atomic spectrometry.

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