Issue



Avoiding the pitfalls of surface analysis


10/01/1997







Scott E. Beck, Andrew G. Gilicinski, Air Products and Chemicals Inc., Allentown, Pennsylvania

A variety of surface analytical tools are available to the process engineer, but data from these tools may be misleading because of hidden pitfalls in the surface analysis technique. Examples from x-ray photoelectron spectroscopy (XPS), secondary ion mass spectrometry (SIMS), total reflection x-ray fluorescence (TXRF), and atomic force microscopy (AFM) are used to illustrate some of the problems encountered when interpreting data from surface analysis equipment. Suggestions are made to assist users of these techniques in avoiding these pitfalls.

IC process engineers can use many surface analytical tools to determine the chemical and physical state of a surface being processed. Surface analysis is used in areas such as yield improvement, defect reduction, and the development of contamination-free manufacturing. Typical techniques (see Table 1 at the end) such as XPS and SIMS have become standard analytical tools that have improved many processing steps. The fact that these techniques have been used for so many years, with many books and review articles written about them, may lull the novice into a false sense of security. The formation of national and international working committees on the standardization of these techniques attests to their dynamic nature [1]. Industry roadmaps such as the National Technology Roadmap for Semiconductors call for lower detection limits for metals on wafer surfaces, and have encouraged modifications to the more standard techniques and stimulated the development of techniques such as TXRF and AFM. We use the example of a vapor phase cleaning method [2, 3] to review the use of XPS, SIMS, TXRF, and AFM, and the pitfalls that may lead to misinterpretation of data obtained from these instruments.

Proper sample handling

Without proper sample handling, meaningful surface analysis is almost impossible. Three different handling issues can confuse the analysis:

  1. manipulation of the sample,
  2. the cleanroom environment, and
  3. the container used to store and ship the sample.

Instruments used to transport the wafer can add metallic species and particles. Most of the contributed contamination resides where the manipulation device has come into contact with the sample. The cleanroom, although clean from a particle point of view, contains many gas phase contaminants [4]. These contaminants include organic compounds (e.g., hydrocarbons, volatile organic compounds, chlorofluorocarbons, amines, organophosphates), inorganic boron, nitrogen-containing species, and sulfur-containing species. The fluoropolymer wafer carriers and storage boxes used in wafer processing can contribute fluorinated species to the wafer surface [5]. All of these contaminants can hinder analysis of the chemical state of the wafer surface. The following suggestions will improve the outcome of any surface analysis.

  1. Use control samples to establish background levels on the surface of interest.
  2. Keep physical contact with the sample to a minimum. For example, use a vacuum pickup on the side opposite to where the analysis will take place. If surface contact is inevitable, make sure the analysis is performed outside the contact areas.
  3. The ambient (both gaseous and particulate) around the sample is a potential source of contamination. Laboratory experience with TXRF and SIMS has shown that samples will typically pick up sulfur and carbon from environmental exposure. A high-efficiency particulate air (HEPA)-filtered area for unloading and transferring samples to the analytical equipment can be used to reduce particle contamination. If gaseous and/or particulate contamination is a problem, it may be worthwhile to use special carriers (glass, vacuum sealed, or with a continuous flow of inert purge gas).
  4. Choose an appropriate container. Products outgassing from a particular type of container may confound the analysis. If another type of container cannot be used, the use of a control sample to determine the effect of the container is required.

The pitfalls of surface analysis

Many analytical techniques are available today to help develop a new process or to troubleshoot an existing process. Due to the sheer numbers of techniques, it is not feasible to give a full description of all their possible setbacks. Our intention is to give the reader a solid basis from which to begin an analysis of a surface.

Chemical state of surface contaminants. In many instances, knowing the chemical state of a surface is important to properly process a device. The determination of the source of some unknown contaminant and post-etch wafer cleaning are examples. Unfortunately, there are very few options available to the researcher or process engineer.

By far the most common technique for determining the surface chemical state is XPS. This technique has many advantages over others in that it can also determine the surface atomic concentration of the elements detected. Unfortunately, it is not suitable for trace surface analysis since it has an absolute sensitivity between 0.01 and 0.3 atomic-% (depending on the element of interest).

The chemical state of the surface can be determined through a knowledge of electron-binding energies. Adventitious carbon is commonly used as a binding energy standard [6, 7]; the peak on the C(1s) portion of the XPS spectrum is typically related to C-H and C-C bonding. Air exposure of the sample can broaden the peak associated with the adventitious carbon by the formation of oxidized carbon bonds such as C-O-C, O=C-O, O=C=O, and C-OH. This broadening can lead to improper assignment of the carbon peak and thus affect the assignment of other peaks in the spectrum.

Additional problems can arise when trying to determine the nature of a reaction on a surface with only one technique, as illustrated by our work on vapor phase cleaning with 1,1,1,5,5,5-hexafluoro-2,4-pentanedione (H+hfac). Initial examination of cleaning effects revealed both carbon and fluorine on wafer surfaces exposed to this chemical. The F(1s) portion of the spectra always contained two peaks (fluoride and organic fluoride), while the C(1s) spectra always contained adventitious carbon and a peak related to CFx [8]. From these data, one might assume that the H+hfac molecule was adsorbed on the wafer surface. This was in part true. However, a secondary technique, time-of-flight secondary ion mass spectrometry (TOF-SIMS), revealed that the resultant surface species included both H+hfac and trifluoroacetic acid [9, 10]. Mass separation in SIMS is done in a magnetic field and relies on the mass-to-charge ratio. A quadrupole mass spectrometer is typically used. In TOF-SIMS, the secondary ions go through a drift tube to be separated. Ions having different masses take different times to reach the detector (with the lightest ions arriving first). This gives TOF-SIMS higher mass resolution and an extended mass range compared to SIMS. XPS was not able to provide definite information since the concentrations of the two species were low, and because both molecules contain O=C-CF3. TOF-SIMS also showed the formation of FeF3 and Fe3C on iron-containing surfaces. The complementary technique resulted in a better understanding of the cleaning process and overcame some of the limitations associated with XPS.

TOF-SIMS is not without its own problems. The major deficiencies of this technique are the lack of surface standards and a limited understanding of the sputter characteristics of most species. These problems can be surmounted by carefully choosing control samples for comparison with the sample of interest.

The SIMS conundrum. SIMS is another common surface analytical technique used in microelectronics processing. This technique has yielded many beneficial results due to its ability to profile low levels of dopants in semiconductors and to determine metal contamination levels on wafer surfaces [11, 12].

We evaluated the use of SIMS to determine the amount of metal contamination on wafer surfaces. Figure 1 depicts typical results for SiO2 surfaces that were intentionally contaminated with Fe via evaporation and cleaned with H+hfac. This type of SIMS profile is common for any contaminant found on a wafer surface. At first glance, one may conclude that Fe is distributed throughout the SiO2 film and the cleaning process actually removes Fe from within the SiO2 layer. The conundrum is that the Fe should only exist on the film surface since it was deposited by evaporation. Furthermore, cleaning is entirely a surface phenomenon and should not draw Fe from the bulk to the wafer surface. At the temperature of the clean, the Fe should only diffuse through a depth of 10 Å.


Figure 1. SIMS depth of profiles of a submonolayer of Fe as deposited via evaporation on a 1000-Å SiO2 layer and after a cleaning treatment with H+hfac. The analysis was done with a 6-keV O2+ primary ion beam in a quadrupole SIMS tool.
Click here to enlarge image

The most likely explanations for the results in Fig. 1 are ion beam mixing [13, 14] and Gibbsian segregation [14, 15]. Analysis of the contaminated samples at various sputter rates resulted in distinctly different profiles, a good indication that ion beam mixing occurred during the analysis [13, 14]. Ion beam mixing was also supported by a simulation of the ion beam-solid interaction using TRIM95 [16, 17] to examine recoil implantation of Fe into the substrate during the SIMS analysis. This simulation showed that the Fe was transported into the sample about 10 Å.

To determine the potential for Gibbsian segregation, we examined the heat of reaction (ΔHR) for the species that could be formed during SIMS analysis. We considered FeOOH and Fe2O3 to be the Fe species on the wafer surface. For the formation of SiO2:

SiO2 = Si + 2O, ΔHR = 7.3 eV/O atom

For the formation of FeOOH and Fe2O3:

FeOOH = Fe + H + 2O, ΔHR = 4.3 eV/O atom

Fe2O3 = 2Fe + 3O, ΔHR = 5.4 eV/O atom

Thus the thermodynamic tendency is for the Si to segregate and for some of the Fe to remain within the sample during analysis. These results indicate that during SIMS analysis a variety of physical and chemical changes are taking place at the sample surface. A knowledge of how the analysis will affect the sample is very important to proper interpretation of the results.

Interferences. Interferences abound in the many different analytical techniques available today. They can lead to some confusion if not accounted for during the analysis. Being able to distinguish between a metal and its various oxidation states via XPS has been important in the development of vapor phase cleans. If the photoelectron lines only are taken into account (e.g., Cuo (metal) and Cu+ in Cu2O [18]), distinguishing between chemically inequivalent atoms by XPS is sometimes impossible. Auger lines can help resolve interference problems. A survey scan of the surface can show the presence of these lines.

Mass spectrometry, including SIMS, suffers from mass interferences. Interferences are very common in SIMS due to the variety and intensity of molecular ions and multiply charged species generated during the analysis [19]. In the analysis of Fe shown in Fig. 1, the 54Fe++ ion was used instead of the more abundant 56Fe++ ion, since 56Fe++ interferes with 28Si+. Other sources of interference can come from the SIMS instrument itself. The ion source and extraction lens can contribute elemental contamination (e.g., H, C, N, O, Al, Cr, Mn, Fe, Ni, Cu, Ta, Na, and K). A variety of methods are available to eliminate mass interferences and background contamination, including high mass resolution spectroscopy, molecular ion subtraction, energy distribution comparisons, isotopic abundances, filters, and voltage offsets.

Interferences can also occur when using TXRF. The major source of these interferences is the anode used to produce the x-rays. Anodes of W and Mo are the most typical with Cu and Cr less frequently used on commercial equipment. Fluorescence interferences between two species are also possible (e.g., Al and Si, or S and Mo). It is common practice to supplement TXRF measurements with surface SIMS, allowing the user to obtain information on the interfering atoms.

TXRF Round-robin. The results of a round-robin experiment with TXRF illustrate the potential variability in this technique. A wafer intentionally contaminated by evaporating approximately 1015 Fe atoms/cm2 onto a SiO2 layer was analyzed by different laboratories, each with its own TXRF instrument. Table 2 shows the results. There is reasonable consistency between the listed concentrations of elements from different instruments. Tungsten was not detected with the TREX 610-T since a W anode was used for this analysis. Each instrument showed the presence of S (or Mo) on the surface. The presence of S and absence of Mo were confirmed by SIMS. The S contamination probably came from the atmosphere since the sample was exposed to air. The Cl surface concentrations noted in Table 2 are most likely due to the growth of the oxide substrate layer in a Cl-containing ambient. The Atomika instrument also detected Co, As, Br, and Au. Only the analyst using the TREX 610-T also noted that the high level of Fe interfered with the detection of Ti, Cr, and Mn, thus showing the importance of selecting an experienced analyst to perform surface analysis.

Click here to enlarge image

Surface roughness. Silicon surface microroughness correlates with a variety of performance problems, including the breakdown of thin gate oxides, the decrease of the effectiveness of wafer-cleaning processes, and the increase in copper deposition out of a buffered oxide etchant solution [20]. AFM is the most promising method for the determination of surface microroughness of silicon wafers at nanometer lateral scales. This technique is attractive because it does not require special sample preparation; it can examine both insulating and conducting surfaces; and it can conduct the measurement in air.

The most typical result of an AFM experiment is an image of the surface and a value describing the surface roughness. The two most commonly used figures of merit are the average roughness (Ra) and the root mean square roughness (Rrms). Ra measures the mean value of the surface to its center plane (i.e., a virtual plane that bisects the surface such that equal volumes are enclosed above and below it). Rrms is a derived parameter reflecting the spread of height values about the mean plane (an arithmetic plane intersecting the true arithmetic mean of the height data). Unfortunately, these figures are not always useful in describing surfaces due to their inherent averaging of all lateral roughness scales into one number. Images of two dissimilar-looking surfaces can have the same Ra or Rrms value. To overcome this shortcoming, the use of power spectral density (PSD) calculations has been proposed for discretely analyzing the magnitude of surface microroughness at a range of lateral distance scales (limited by the scales available in the data) [20].


Figure 2. An image of a calibration standard at two scan sizes: a) 10 µm and b) 1.5 µm.
Click here to enlarge image

A power spectral density calculation of a surface yields the frequency spectrum of the AFM data. That is, the PSD spectrum is generated by performing a Fourier transform on the z(x) or z(x,y) data sets obtained from the 2- or 3-D AFM images. The resulting spectrum for a 3-D set of AFM data is a plot of the PSD function (in units of nm4) vs. the lateral wavelength (in units of µm). Figure 2 shows an image of a calibration standard at two scan sizes (10 and 1.5 µm). Figure 3 shows PSD plots for the two images. The Ra and Rrms values for the two images are identical, since the z-axis variation is the same for both the scan sizes; however, the PSD curves reveal differences in roughness at the lateral wavelengths analyzed in the AFM images. Greater resolution of roughness at smaller scales is measured in the higher-magnification image.


Figure 3. PSD plots for the two images in Fig. 2.
Click here to enlarge image

Although PSD improves the analysis of experimental data, subtle issues in the experiment can cause problems in quantitative comparisons. A key experimental issue involves the size and shape of the AFM probe. AFM data are a complex function of the surface topography and its interaction with the probe tip. Tip effects become significant when the lateral scale of topography approaches the probe tip size, which is usually the case for silicon wafer microroughness at the nanometer scale. One way to account for this is by using a colloidal gold sphere AFM "tip characterizer" [20]. We suggest that this standard be used both to monitor possible tip blunting and to understand the extent of tip shape convolutions in the AFM data. In addition, PSD spectra can be used to identify lateral wavelengths where tip shape has influenced AFM data significantly. Figure 4 shows an example of variation due to tip effects for an as-received silicon wafer.


Figure 4. Images of the same surface with different AFM tips.
Click here to enlarge image

We propose the following set of procedures for determining surface microroughness:

  1. Obtain an AFM image of the tip characterizer with the tip to be used for sample analysis.
  2. Acquire one image of the sample of interest.
  3. Obtain a second AFM image of the tip characterizer using the same probe tip.
  4. Apply plane fitting to the AFM image. Do not use any other type of filtering. Report the Ra and Rrms values and PSD spectrum of the image. Include the tip diameter at 5 nm from the apex.
  5. Repeat steps 1–4 for at least two additional images of the surface.

The careful application of this methodology significantly improves the reproducibility of AFM microroughness data, especially for analysis of device-quality wafers (Ra <2 Å).

Additional ways to avoid the pitfalls

The previous examples illustrate ways to avoid some of the common pitfalls encountered when interpreting surface analysis data. Some other points that may save you time and money are as follows:

  • Use the same analyst for work that is ongoing in nature. This will eliminate subtle variations in the data that may be introduced from one analyst to another. In addition, an ongoing relationship with the analyst will let him or her become familiar with your particular needs and concerns.
  • Use the same operating conditions if comparisons are to be made between samples.
  • When using spectroscopy, always get a full survey spectrum of the surface. Although only a particular species (or portion of the spectrum) may be of interest, a full spectrum will give additional information that can help resolve a particular problem and eliminate any discrepancies.
  • Tell the analyst as much as you can about the sample and the processing it has undergone.
  • Learn as much about the particular analytical technique as possible.

Acknowledgment

This work was supported by the ARPA-NCAICM program (contract no. N0014-94-C-0076) and by internal research funds from Air Products and Chemicals Inc.

References

  1. C.J. Powell, "Activities of the ASTM Committee E-42 on Surface Analysis," Surf. Interface Anal., Vol. 19, pp. 237–240, 1992.
  2. S.E. Beck et al., "Chemical Vapor Cleaning Technologies for Dry Processing in Semiconductor Manufacturing," in Proceedings of the Third International Symposium on Cleaning Technology in Semiconductor Device Manufacturing, pp. 253–263, eds., J. Ruzyllo, R.E. Novak, Electrochem. Soc. Proc. Vol. 94–97, Pennington, NJ, 1994.
  3. S.E. Beck et al., "The Effects on Surfaces of Silicon and Silicon Dioxide Exposed to 1,1,1,5,5,5-hexafluoro-2,4-pentanedione," in Proceedings of the Third International Symposium on Ultra Clean Processing of Silicon Surfaces, pp. 264–271, ACCO, Leuven, Belgium, 1996.
  4. A.J. Muller, L.A. Psota-Kelty, H.W. Krautter, J.D. Sinclair, "Volatile Cleanroom Contaminants: Sources and Detection," Solid State Technology, pp. 61–72, Sept. 1994.
  5. J. Goodman, S. Andrews, "Fluoride Contamination from Fluoropolymers in Semiconductor Manufacture," Solid State Technology, pp. 65–68, July 1990.
  6. Practical Surface Analysis, 2d. ed., Vol. 1, eds., D. Briggs, M.P. Seah, John Wiley & Sons, Chichester, 1990.
  7. T.L. Barr, S. Seal, "Nature of the Use of Adventitious Carbon as a Binding Energy Standard," J. Vac. Sci. Technol. A, Vol. 13, pp. 1239–1246, 1995.
  8. S.E. Beck et al., "The Effects of Chemical Vapor Cleaning Chemistries on Silicon Surfaces," in Interface Control of Electrical, Chemical, and Mechanical Properties, pp. 263-268, eds., S.P. Murarka, K. Rose, T. Ohmi, T. Seidel, Mat. Res. Soc. Proc. Vol. 318, Pittsburgh, PA, 1994.
  9. S.E. Beck et al., "Submonolayer Iron and Copper Removal by Exposure to 1,1,1,5,5,5-hexafluoro-2,4-pentanedione (H+hfac)," in Proceedings of the Fourth International Symposium on Cleaning Technology in Semiconductor Device Manufacturing, pp. 166–174, eds., J. Ruzyllo, R.E. Novak, Electrochem. Soc. Proc. Vol. 95–20, Pennington, NJ, 1996.
  10. M.A. George et al., "Reaction of 1,1,1,5,5,5-hexafluoro-2,4-pentanedione (H+hfac) with Iron Oxide Thin Films," J. Electrochem. Soc., Vol. 143, pp.3257–3266, 1996.
  11. S.P. Smith, "Routine SIMS Measurement of Surface Metal Contamination on Silicon," in Secondary Ion Mass Spectrometry: SIMS IX, pp. 476–479, eds., A. Benninghoven, Y. Nihei, R. Shimuzu, H.W. Werner, John Wiley & Sons, Chichester, 1990.
  12. M.R. Frost, "On the Use of Quadrupole SIMS for the Measurement of Surface Metallic Contamination," in Contamination Control and Defect Reduction in Semiconductor Manufacturing III, pp. 339–348, eds., D.N. Schmidt, D. Reedy, R.L. Guldi, J.V. Martinez de Pinillos, Electrochem. Soc. Proc. Vol. 94–99, Pennington, NJ, 1994.
  13. P. Williams, J.E. Baker, "Implantation and Ion Beam Mixing in Thin Film Analysis," Nucl. Instrum. Meth., Vol. 182/183, pp. 15–24, 1981.
  14. K. Wittmaack, "Towards the Ultimate Limits of Depth Resolution in Sputter Profiling: Beam-induced Chemical Changes and the Importance of Sample Quality," Surf. Interface Anal., Vol. 21, pp. 323–335, 1994.
  15. V.R. Deline, W. Reuter, R. Kelly, "Gibbsian Segregation during Depth Profiling of Copper in Silicon," in Secondary Ion Mass Spectrometry: SIMS V, pp. 299–302, eds., A. Benninghoven, R.J. Colton, D.S. Simons, H.W. Werner, John Wiley & Sons, Chichester, 1986.
  16. A program based on J. Ziegler, The Stopping and Range of Ions in Matter, Pergammon Press, New York, 1985.
  17. J.P. Biersack, L. Haggmark, "A Monte Carlo Computer Program for the Transport of Energetic Ions in Amorphous Targets," Nucl. Instrum. Meth., Vol. 174, No. 1–2, pp. 257–269, 1980.
  18. P.E. Larson, "X-ray-induced Photoelectron and Auger Spectra of Cu, CuO, Cu2O, and Cu2S Thin Films," J. Electron Spectrosc., Vol

    Scott E. Beck received his PhD degree in physics from Lehigh University. From 1990 to 1992, he was a visiting assistant professor in the Department of Electrical and Computer Engineering at the University of Arizona. He is a principal research physicist with Air Products and Chemicals Inc., where he heads the chemical vapor cleaning effort. Air Products and Chemicals Inc., 7201 Hamilton Boulevard, Allentown, PA 18195-1501; ph 610/481-8138, e-mail [email protected].

    Andrew G. Gilicinski received his PhD degree in chemistry from the University of Wisconsin-Madison. He is a lead applications chemist with Air Products and Chemicals Inc., where he established an AFM lab in 1991. He has lectured internationally on AFM and pioneered improved methods for evaluating silicon wafer surfaces.

    Table 1. Typical surface analysis techniques in the semiconductor industry*

    X-ray photoelectron spectroscopy (XPS)

    Identifies the elements on a surface, their concentration, and their chemical state. Monoenergetic soft x-rays incident on a material produce ejected photoelectrons. The kinetic energies of the photoelectrons are analyzed to identify the elements and chemical state. This technique is also known as electron spectroscopy for chemical analysis (ESCA).

    Secondary ion mass spectrometry (SIMS)

    Measures the impurity profile. A sample is exposed in vacuo to a primary beam of ions that are energetic enough to eject (or sputter) atoms and clusters of atoms from the sample surface. Some of the sputtered species are ions (secondary ions), which are accelerated and analyzed via a mass spectrometer as a function of their mass-to-charge ratio. The intensities of the signals in the mass spectrum are converted to concentrations via comparison to standards. These intensities change as a function of sputter time, thus giving impurity concentration in the bulk material as a function of depth.

    Total reflection x-ray fluorescence (TXRF)

    Used to determine concentrations of transition metal contaminants on wafer surfaces. A wafer is exposed to an x-ray beam at a grazing angle, which is kept below the critical angle for total reflection. Thus, only a few nanometers of the surface are penetrated by the x-rays. The surface atoms fluoresce due to the x-ray excitation. The spectrum of this radiation contains quantitative information on the surface elements.

    Atomic force microscopy (AFM)

    High-resolution profilometry used to map a 3-D image of a surface. A sharp stylus on a microcantilever is scanned with angstrom precision over a sample surface. Interaction between the sample surface and probe tip causes deflection in the microcantilever, which allows height measurements at each point of the scan. Data are reconstructed to generate nanometer-scale topographic maps of the surface.

    *C.R. Brundle, C.A. Evans, S. Wilson, Encyclopedia of Materials Characterization, Butterworth-Heinemann, Stoneham, MA, 1992.