Stress Characterization of Si by Near-Field Raman MicroscopeUsing Resonant Scattering

MASANOBU YOSHIKAWA* and MASATAKA MURAKAMIToray Research Center Inc., Sonoyama 3-3-7, Otsu, Shiga 520-8567, Japan

We have developed a tapping-mode scanning near-field optical Raman

microscope (SNORM) with a caved and pyramidical probe, using

resonant Raman scattering, and have measured the stress distribution

of Si. The peak frequency shifts to a lower frequency by 0–0.5 cm�1 in the

area covered by silicon dioxide, whereas it shifts to a higher frequency by

0–0.3 cm�1 in the area uncovered by silicon dioxide, showing that the

areas covered and uncovered by silicon dioxide are under tensile and

compressive stresses, respectively. It has been found that compressive

stresses of about 0.69 GPa/cm2 are concentrated on the corner of the area

uncovered by silicon dioxide. The comparison of stress distributions

measured with and without the cantilever shows that the SNORM we

developed has a spatial resolution of at least less than 250 nm.

Index Headings: Near field; Raman spectroscopy; Resonant Raman

scattering; Nanometer; Si; Stress.

INTRODUCTION

There are few reports on Raman spectroscopy of the stressestimation of Si using scanning near-field optical microscope(SNOM),1–3 although there has been much research onmaterials that emit a very strong Raman signal such asRhodamine 6G,4 C60,5 and polydiacetylene6 and materials thatemit a strong luminescence signal such as quantum dots7 andpolymers.8–10 This is because of the very weak near-fieldRaman signal of Si.

A scanning near-field optical Raman microscope with anoptical fiber probe has been developed to break down therestriction of spatial resolution.1,6 However, the apparatus hasthe following problems: (1) obstruction of the weak Ramansignal of the measured sample by the strong luminescence andRaman line emitted from the optical fiber probe made of silicondioxide; (2) low transmittance of the optical fiber probe (10�3);and (3) weakness of the signal due to the optical penetrationdepth of the near-field light being less than 100 nm.

We have developed a scanning near-field optical Ramanmicroscope with the following specific characteristics in orderto overcome these problems: (1) adoption of a probe made ofsilicon nitride with lower luminescence and Raman line; (2)adoption of a caved and pyramidical probe with a hightransmittance of 10�2; and (3) adoption of an ultraviolet (UV)laser with an optical penetration depth of 5 nm.

The change of the excitation wavelength is one of severalmethods used to increase very weak near-field Raman signals.Sands et al.11 have detected very weak near-field Ramansignals of diamond films by use of a deep UV laser. ResonantRaman scattering occurs when the excitation energy corre-sponds to the electronic transition energy of the materials. Inthe case of Si, the direct band gap at the C point is about 3.4 eV(;363.8 nm).12,13 Indeed, the resonance in the Si leads to

a strong enhancement of the Raman scattering cross-section(about a hundred times) and the optical penetration depthbecomes very thin under the resonant condition. The opticalpenetration depth of Si at the excitation wavelength of 363.8nm is 5 nm.14

In this work, we report our results on the stresscharacterization of silicon employing a tapping-mode SNORMand using resonant Raman scattering with an ultraviolet (UV)Arþ laser.

EXPERIMENTAL

Sample. Commercially available VLSI (very-large-scaleintegration) standards with pitches of 1800 nm, which werecoated with a silicon dioxide layer with a thickness of 180 nm,on a Si(100) substrate were used for the UV Ramanmeasurement. As a result, there are periodic patterns witha step height of 180 nm formed by the silicon dioxide layer andwith a square of 1600 nm that is not covered by the silicondioxide layer in the VLSI standards.

Setup. The schematic diagram of the tapping-modeSNORM that we developed is shown in Fig. 1. We haveadopted the feedback system, which detects the signal to beproduced by a change in the amplitude of vibration ofa micro-tuning fork,15,16 mounted by a cantilever. A cavedand pyramidical probe with an aperture of 100 nm was usedfor UV Raman measurement. The laser beam was focused onthe aperture of the probe by a UV objective lens with an NA(numerical aperture) of 0.5 at a power of less than 0.3 mW toavoid the peak-frequency shift by thermal expansion, and thenear-field Raman scattered light was collected by the sameobjective lens through the aperture of the probe (illumination-collection mode). The UV Raman spectra were measuredusing a UV Raman microprobe with a focal length of 1 mand with a grating of 3600 grooves/mm, designed byPHOTON Design Co., with a 2048 channel UV-coatedcharge-coupled device (CCD) produced by Roper ScientificCo. All UV Raman spectra were measured at roomtemperature in the backscattering configuration with thepolarization direction of the incident light parallel to the,010. axis of Si, and with depolarized scattered light. An x-y piezoelectric scanner was purchased from NanonicsImaging Ltd. The 363.8 nm line of a UV Arþ laser wasused as the excitation source. UV Raman spectra of VLSIstandards were measured, moving the x-y piezoelectricscanner by steps of 250 nm, while controlling the separationbetween the probe and the surface of the VLSI standardsusing the feedback system. Each of the Raman spectra wasmeasured within an exposure time of 10 seconds. The peakfrequency was determined by fitting a Lorenz curve to thespectrum and an accuracy of about 60.05 cm�1 wasachieved.

Received 3 January 2006; accepted 14 March 2005.* Author to whom correspondence should be sent. E-mail:masanobu_yoshikawa@trc.toray.co.jp.

Volume 60, Number 5, 2006 APPLIED SPECTROSCOPY 4790003-7028/06/6005-0479$2.00/0

� 2006 Society for Applied Spectroscopy

RESULTS AND DISCUSSION

Figure 2 shows the topographic image in the scanning areaof 5000 nm 3 5000 nm of the VLSI standards, measured by theSNORM. This image was measured by a commerciallyavailable bent fiber probe with an aperture of 20 nm. FromFig. 2, an area with a width of 200 nm, covered by the silicondioxide layer, and the area of the Si(100) substrate witha square of 1600 nm uncovered by the silicon dioxide layer, arefound to be observed periodically.

Figure 3 shows a topographic image in the scanning area of5000 nm 3 5000 nm of the VLSI standards, measured bya caved and pyramidical probe with a diameter of 100 nm. InFig. 3, although the area of the Si(100) substrate covered by thesilicon dioxide layer and the area of the Si(100) substrateuncovered by the silicon dioxide layer are observed periodi-cally, the area covered by the silicon dioxide layer is found tohave spread in comparison with Fig. 2. This difference betweenFigs. 2 and 3 is considered to be caused by the difference of thetip diameter and the solid angle between the pyramidical probeand the bent fiber probe.

Figures 4a, 4b, 4c, and 4d show the typical Raman spectra ofthe VLSI standards, measured using the pyramidical probewith a diameter of 100 nm. Figure 4a shows the far-fieldRaman spectrum of a single crystalline Si(100) wafer forcomparison. The Raman line in the Si(100) wafer is observedat 520 cm�1. The intensity of the far-field Raman spectrum wasfound to be about 103–104 times larger than that of the near-field Raman spectra, measured using the same laser power. Wecould not detect the near-field Raman spectrum using the bentfiber probe. The peak frequency shifts to a lower frequency inthe area covered by the silicon dioxide, showing that the areascovered by silicon dioxide are under tensile stresses.

As seen in Figs. 4b, 4c, and 4d, the Raman line of Sibecomes asymmetric with a main peak around 520 cm�1 anda shoulder with some tailing toward a lower frequency. Theweak shoulder band might originate from the probe itself,which is made of silicon nitride, because a broad Raman bandhas been observed around 500 cm�1 in amorphous siliconnitride films.17 The observed spectrum was divided into themain peak and the weak shoulder component with some tailing

FIG. 1. The schematic diagram of the tapping-mode SNORM.

FIG. 2. The topographic image in the scanning area of 5000 nm 3 5000 nm ofthe VLSI standards, measured using the SNORM. This image is measuredusing a bent fiber probe with an aperture of 20 nm.

FIG. 3. The topographic image in the scanning area of 5000 nm 3 5000 nm ofthe VLSI standards, measured using a pyramidical cantilever with a diameter of100 nm.

480 Volume 60, Number 5, 2006

toward a lower frequency by fitting a Lorenz curve to thespectrum.

Figures 5a, 5b, and 5c show the images of the peak intensity,peak frequency, and peak-frequency shift, respectively, of themain peak in the VLSI standards, measured using thepyramidical probe. We could detect the Raman signal of Sionly when the incident light was focused on the aperture of theprobe by the UV objective lens. The hardness of silicon nitridefilms is about two times larger than that of silicon dioxidefilms.18 Although the probe separation in the area covered bythe silicon dioxide layer is 180 nm wide, the probe may beburied in the silicon dioxide layer because of the measurementby the tapping mode and the difference of hardness betweenthe probe, made of silicon nitride, and the silicon dioxide layer.Since the periodic patterns of the peak intensity, peak

frequency, and peak-frequency shift are observed in Figs. 5a,5b, and 5c, the aperture is considered to be not damaged andwidened during measurement.

Generally, the Raman line of Si shifts to higher and lowerfrequencies under compressive and tensile stresses, respective-ly.19–21 In Fig. 5c, the peak frequency shifts to a lowerfrequency by 0–0.5 cm�1 in the area covered by the silicondioxide, whereas it shifts to a higher frequency by 0–0.3 cm�1

in the area uncovered by the silicon dioxide. This result showsthat the areas covered and uncovered by the silicon dioxide areunder tensile and compressive stresses, respectively. Thetensile stresses are considered to be caused by a difference inthe thermal expansion coefficients between silicon and silicondioxide when the wafer was cooled from deposition temper-ature to room temperature. Furthermore, the compressivestresses are considered to be caused by a balance betweencompressive and tensile stresses. Images in the peak intensityand peak frequency of the decomposed shoulder componentwith some tailing toward lower frequency do not show periodicpatterns. This suggests that the decomposed shoulder compo-nent originates from the probe itself.

As seen in Fig. 5c, the compressive stresses concentrate onthe interface of the areas covered and uncovered by the silicondioxide. We can estimate the compressive stress from theobserved value of the peak-frequency shift using Eq. 11 of Ref.22. It has been found that the compressive stresses of about0.69 GPa/cm2 are concentrated on the interface of the areascovered and uncovered by the silicon dioxide. In Figs. 5a, 5b,and 5c, the areas covered by the silicon dioxide are recognizedto be about 200 nm. This suggests that the spatial resolution ofthe SNORM developed is at least less than 250 nm.

Figures 6a, 6b, and 6c show the images of the peak intensity,peak frequency, and peak-frequency shift in the VLSI stan-dards, respectively, measured without the pyramidical cantile-ver. As seen in Fig. 6a, the area covered by the silicon dioxidelayer seems broad. Furthermore, the distributions of the peakfrequency and peak-frequency shift in Figs. 6b and 6c arehomogeneous and out of focus in comparison to Figs. 5b and5c. This difference is considered to be caused by the differencein spatial resolution.

The up-shift of the peak frequency in the areas uncovered bythe silicon dioxide in Fig. 5c is larger than that in Fig. 6c. In the

FIG. 4. The typical Raman spectra of VLSI standards, measured using thepyramidical cantilever with a diameter of 100 nm: (a) far-field Raman spectrumof the single crystalline Si (100) wafer for comparison, (b) near-field Ramanspectrum of point 1 in Fig. 5c, (c) near-field Raman spectrum of point 2 in Fig.5c, and (d) near-field Raman spectrum of point 3 in Fig. 5c. The observedspectrum was divided into the main peak and the weak shoulder componentwith some tailing toward a lower frequency by fitting a Lorenz curve to thespectrum. The solid and dashed lines show the experimental and calculatedRaman spectrum, respectively.

FIG. 5. The images of (a) the peak intensity, (b) peak frequency, and (c) peak-frequency shift in the VLSI standards, measured using the pyramidical cantilever witha diameter of 100 nm. The peak frequency shift was measured with respect to the peak frequency (520 cm�1) of the Raman line in the Si(100) wafer. Each of theRaman spectra was measured within an exposure time of 10 seconds.

APPLIED SPECTROSCOPY 481

case of high excitation power density, the peak frequency ofthe Raman line in Si tends to shift to lower frequency. Theseresults show that the peak-frequency shift caused by the high

excitation power density in Fig. 5c is not significantly largecompared with that in Fig. 6c.

From a comparison of Figs. 3 and 5, the spatial resolution inFig. 5 seems to be higher than that in Fig. 3. The separationcontrol feedback works between the edge of the silicon dioxide

layer and the side slope of the pyramidical probe. Since thepyramidical probe has a large solid angle, the sample–aperture

separation changes gradually in the vicinity of the edge of thesilicon dioxide layer in Fig. 3. However, the near-field incident

light penetrates into the substrate and the near-field Ramanlight is detected through the aperture of the pyramidical probe.The difference of images between Figs. 3 and 5 might be

mainly caused by the penetration of the near-field incident lightinto the substrate.

CONCLUSION

We have developed a tapping-mode scanning near-fieldoptical Raman microscope, using resonant Raman scattering,

and have measured the stress distribution of VLSI standards.The peak frequency shifts to a lower frequency by 0–0.5 cm�1

in the area covered by the silicon dioxide, whereas it shifts to

a higher frequency by 0–0.3 cm�1 in the area uncovered by thesilicon dioxide, showing that the areas covered and uncovered

by the silicon dioxide are under tensile and compressivestresses, respectively. It has been found that compressive

stresses of about 0.69 GPa/cm2 are concentrated on the cornerof the area uncovered by the silicon dioxide. The comparisonof stress distributions measured with and without the cantilever

shows that the SNORM developed has a spatial resolution lessthan 250 nm.

ACKNOWLEDGMENT

This work was financially supported by The New Energy and IndustrialTechnology Development Organization (NEDO).

1. Near-Field Nano/Atom Optics and Technology, M. Ohtsu, Ed. (Springer-Verlag, Tokyo, 1998).

2. S. Webster, et al., Appl. Phys. Lett. 72, 1478 (1998).3. J. Grausem, et al., J. Raman Spectrosc. 30, 833 (1999).4. J. Prikulis, K. V. G. K. Murty, H. Olin, and M. Kall, J. Microsc. 210, 269

(2003).5. J. J. Wang, D. A. Smith, D. N. Batchelder, Y. Saito, J. Kirkham, C.

Robinson, K. Baldwin, G. Li, and B. Bennett, J. Microsc. 210, 330 (2003).6. Y. Narita, T. Tadokoro, T. Ikeda, T. Saiki, S. Mononobe, and M. Ohtsu,

Appl. Spectrosc. 52, 1141 (1998).7. S.-K. Eah, W. Jhe, and Y. Arakawa, Appl. Phys. Lett. 80, 2779 (2002).8. M. Micic, K. Radotic, M. Jeremic, D. Djikanovic, and S. B. Kammer,

Colloids Surf., B 34, 33 (2004).9. N. Hayazawa, Y. Inouye, and S. Kawata, J. Microsc. 194, 472 (1999).

10. R. Stevenson, M. Granstrom, and D. Richards, Appl. Phys. Lett. 75, 1574(1999).

11. H. S. Sands, et al., J. Raman Spectrosc. 33, 730 (2002).12. J. B. Renucci, R. N. Tyte, and M. Cardona, Phys. Rev. B 11, 3885 (1975).13. M. Yoshikawa, M. Murakami, K. Matsuda, T. Matsunobe, R. Sugie, K.

Okada, and H. Ishida, J. Appl. Phys. 98, 063531 (2005).14. M. Holts, W. M. Duncan, S. Zollner, and R. Liu, J. Appl. Phys. 88, 2523

(2000).15. K. Karrai and R. D. Grober, Appl. Phys. Lett. 66, 1842 (1995).16. D. N. Davydov, K. B. Shelimov, T. L. Haslett, and M. Moskovits, Appl.

Phys. Lett. 75, 1796 (1999).17. J. Bandet, B. Despax, and M. Caumont, J. Appl. Phys. 85, 7899 (1999).18. U. Beck, D. T. Smith, G. Reiners, and S. J. Dapkunas, Thin Solid Films

332, 164 (1998).19. F. Cerdeira, C. J. Buchenauer, F. H. Pollak, and M. Cardona, Phys. Rev. B

5, 580 (1972).20. E. Anastassakis, Proceedings of the 4th International School ISPPM, J.

Kassavod, Ed. (Varan, Bulgaria, 1985), p. 128.21. M. Yoshikawa, M. Maegawa, G. Katagiri, and H. Ishida, J. Appl. Phys. 78,

941 (1995).22. M. Yoshikawa and N. Nagai, ‘‘Vibrational Spectroscopy of Carbon and

Silicon Materials’’, in Handbook of Vibrational Spectroscopy, J. M.Chalmers and P. R. Griffiths, Eds. (John Wiley and Sons, Chichester,2002), vol. 1, p. 2593.

FIG. 6. The images of (a) the peak intensity, (b) peak frequency, and (c) peak-frequency shift in the VLSI standards, measured without the pyramidical cantilever.Each of the Raman spectra was measured within an exposure time of 1 second.

482 Volume 60, Number 5, 2006