ARTICLE IN PRESS

Ultramicroscopy 109 (2009) 1183–1188

Contents lists available at ScienceDirect

Ultramicroscopy

0304-39

doi:10.1

� Corr

E-m

journal homepage: www.elsevier.com/locate/ultramic

Bandgap measurement of thin dielectric films usingmonochromated STEM-EELS

Jucheol Park�, Sung Heo, Jae-Gwan Chung, Heekoo Kim, HyungIk Lee, Kihong Kim, Gyeong-Su Park

Samsung Advanced Institute of Technology, AE Center, Nongseo-Dong Mt. 14-1, Giheung-gu, Yongin-si, Gyeonggi-do 446-712, Republic of Korea

a r t i c l e i n f o

Article history:

Received 17 June 2008

Received in revised form

6 April 2009

Accepted 28 April 2009

PACS:

82.80.Dx

79.20.Uv

68.37.Ma

77.55.+f

Keywords:

Bandgap measurement

Monochromtor

STEM

EELS

AES

REELS

a-SiNx

SiO2

Charge trap memory

91/$ - see front matter & 2009 Elsevier B.V. A

016/j.ultramic.2009.04.005

esponding author. Tel.: +82 31280 9141; fax:

ail address: jucheol.park@samsung.com (J. Pa

a b s t r a c t

High-resolution electron energy-loss spectroscopy (HR-EELS), achieved by attaching electron mono-

chromators to transmission electron microscopes (TEM), has proved to be a powerful tool for measuring

bandgaps. However, the method itself is still uncertain, due to Cerenkov loss and surface effects that can

potentially influence the quality of EELS data. In the present study, we achieved an energy resolution of

about 0.13 eV at 0.1 s, with a spatial resolution of a few nanometers, using a monochromated STEM-EELS

technique. We also assessed various methods of bandgap measurement for a-SiNx and SiO2 thin

dielectric films. It was found that the linear fit method was more reliable than the onset reading method

in avoiding the effects of Cerenkov loss and specimen thickness. The bandgap of the SiO2 was estimated

to be 8.95 eV, and those of a-SiNx with N/Si ratios of 1.46, 1.20 and 0.92 were measured as 5.3, 4.1 and

2.9 eV, respectively. These bandgap-measurement results using monochromated STEM-EELS were

compared with those using Auger electron spectroscopy (AES)-reflective EELS (REELS).

& 2009 Elsevier B.V. All rights reserved.

1. Introduction

The bandgap measurement of thin dielectric films is essentialto improving the reliability of charge trap flash memory, which isaccomplished by bandgap engineering of the silicon nitride andoxide layers used as traps and tunneling layers, respectively [1].Traditionally, the bandgaps of dielectrics have been measured byoptical methods offering high-energy resolution (� several tens ofmeV) but very poor spatial resolution (�0.2mm). This spatialresolution clearly is insufficient for measuring bandgaps inmodern devices with horizontal and vertical structures in thenanometer range. Hence, there has been an increasing demand forboth high-energy and high-spatial-resolution techniques forbandgap measurement.

Recently, high-resolution electron energy-loss spectroscopy(HR-EELS), achieved by attaching electron monochromators totransmission electron microscopes (TEM), has proved to be apowerful tool for bandgap measurements on the nano-scale [2–8].Despite this technological breakthrough, however, there remain

ll rights reserved.

+82 31 280 9157.

rk).

restrictions for spatial resolution and precision of measurementon monochromated TEM. First, the ultimate spatial resolutionobtainable by TEM is limited by the beam size and thedelocalization phenomenon. Second, estimation of precise band-gaps in the STEM mode, which offers superior spatial resolution, isstill uncertain, due to Cerenkov loss [9–12] and various effectsthat impose artifacts on bandgap measurements for a wide rangeof thin dielectric films [13,26].

In this paper, several methods of TEM bandgap measurementfor thin dielectric films were assessed by means of comparisonswith the Auger electron spectroscopy-reflective EELS (AES-REELS)method. We report that we measured reliable bandgap values ofthin dielectric films using the linear fit method in STEM modewith an energy resolution of about 0.13 eV at 0.1 s and a spatialresolution of several nanometers using our monochromatedSchottky-FEG TEM.

2. Experiments

SiNx films from various SiH4:NH3:N2 gas mixtures at 400 1Cwere deposited at 670 W RF power by the plasma-enhanced

ARTICLE IN PRESS

101

100

10-1

10-2

10-3

1.4

1.2

1.0

0.8

0.6

0.4

0.2

0.0

-0.2-2 -1 0 1 2

Energy Loss (eV)

Inte

nsity

(Arb

itrar

y U

nit)

Nor

mal

ized

Inte

nsity

Normal: 0.7eVMonochromated: 0.13 eV

Normal: 0.7eVMonochromated: 0.13 eV

0.13 eV

J. Park et al. / Ultramicroscopy 109 (2009) 1183–11881184

chemical vapor deposition (PECVD) method on (0 0 1) p-Si. TheSiO2 films were grown by conventional thermal oxidation. TEMspecimens were prepared according to the mechanical-thinningand ion-milling method. Plasma cleaning was carried out beforethe TEM work in order to acquire the bandgap spectra.

The STEM-EELS spectra were obtained using Field-emissionTEM (FEI Titan 80-300) equipped with a Wien-type monochro-mator and a high-resolution Gatan imaging filter (GIF Tridiem865 ER300) installed at the Samsung Advanced Institute ofTechnology. The acceleration voltage of the TEM was 300 kV.The energy spread of a monochromated zero-loss peak (ZLP) was0.13–0.27 eV in full-width at half-maximum (FWHM), dependingmainly on energy dispersion, sample thickness and monochro-mator alignment. The energy dispersion was 0.01 eV/ch forthe energy resolution and 0.02 eV/ch for the bandgap. Thecollection angle was 5–10 mrad, which varied with camera length(o100 mm) and the entrance aperture size of the HR-GIF (1.0 or2.5 mm in diameter). The EELS spectra for various thicknesses(50–200 nm) were acquired in the STEM imaging mode.

In order to compare the bandgaps estimated by STEM-EELS,AES-REELS spectra were measured on a VG Microlab 350F withprimary electron energies of 1000 eV in the constant analyzerenergy mode of 10 eV pass energy. The FWHM of the elastic peakwas 0.8 eV. The energy-loss range was 0–100 eV.

For accurate composition analysis of Si and N, an HRRutherford backscattering system equipped with a magneticsector analyzer capable of obtaining a high-resolution energyspectrum of the MEIS level (KOBELCO HRBS-V500) was used. TheRBS analysis was carried out with a 400 keV He+ probe beam,which was incident at the angle of 501 with a scattering angle of70.51. In order to avoid surface damage, the measurement pointsfor the RBS were continually changed.

10-4

-4 -3 -2 -1 0 1 2 3 4Energy Loss (eV)

Fig. 1. Zero loss peaks acquired from the conventional and the monochromated

Schottky-FEG. Normalized intensities are shown on (a) a linear scale and (b) a

logarithmic scale. The full-widths at half-maximum of the conventional and the

monochromated Schottky-FEG are 0.7 and 0.13 eV, respectively. Two arrows at 1.1

and 3.9 eV show the energy-loss values at a 1/1000th maximum of the

conventional and the monochromated Schottky-FEG, respectively.

3. Results and discussion

3.1. Energy resolution of monochromated EELS

The energy resolution of an entire TEM-EELS system is one ofthe most important parameters in EELS analysis. There are fourmain factors associated with practical energy resolution in EELS:(a) the energy spread of the electron source, (b) the non-isochromaticity of the spectrometer, (c) the point spread (blur-ring) of the detector, and (d) instabilities, such as stray magneticfields, in the TEM-high-voltage and room environments. Amonochromator with a high-resolution spectrometer and ahigh-voltage tank directly reduces the energy spread of anelectron beam, naturally resulting in an attainable high-energyresolution.

In general, the FWHM of ZLPs is regarded as an energy-resolution indicator. Fig. 1(a) shows the ZLPs acquired froma normal Schottky-FEG TEM and those derived with amonochromated Schottky-FEG TEM. TEM equipped with athermally assisted Schottky field emission source typicallyproduces, under normal operating conditions, an energyresolution of about 0.7 eV and an asymmetrical ZLP, whichcomplicates the removal of the ZLP in measuring the bandgap. Bycontrast, monochromated Schottky-FEG TEM equipped with aWien-type monochromator, high-resolution electron spectrometerand a high-resolution high-tension tank offers an energy resolutionof about 0.13 eV and a relatively symmetrical ZLP (Fig. 1(a)).

Fig. 1(b) shows, on a logarithmic scale, the normalizedintensities of the ZLPs acquired from the normal and mono-chromated Schottky-FEG. When the normalized intensities areshown on a logarithmic scale, the tail shape and range of theintensities are much enhanced, allowing us easily to recognize the

noise and background levels. The FWHM of the monochromatedand conventional Schottky-FEGs were 0.13 eV and 0.7, respec-tively. The background intensities in the region of interest(1–10 eV) for a monochromated Schottky-FEG are considerablylower than those for a normal Schottky-FEG. Thus, if the thresholdfor bandgap detection is set at 1/1000th of the maximumintensity, as suggested in the literature [14,15], a horizontal linecan be drawn at this level, so that its intersection point can beassigned with the tails of the zero-loss distribution as theminimum energy loss that can be detected above the background.The lowest detection limit for a dispersion of 0.01 eV was 1.1 eV forthe monochromated ZLP and �4.0 eV for the normal schottky-FEG. It should be noted that the difference in the minimum energybetween a conventional and a monochromated schottky-FEG issubstantial and critical to the accurate determination of bandgapenergy.

3.2. Bandgap measurement of SiO2

The bandgap refers to the energy difference between the top ofthe valence band and the bottom of the conduction band. Fast

ARTICLE IN PRESS

Fig. 3. Four spectra of single scattering distributions (SSD) obtained by the

subtraction of zero-loss peaks at various thicknesses. After the ZLP was

deconvoluted from the original spectrum, some broad CL peaks were shown at

4–8 and 8–10 eV.

J. Park et al. / Ultramicroscopy 109 (2009) 1183–1188 1185

electrons passing through a solid can excite an electron–holeinterband transition, which is interpreted as an electronictransition from a bound state (either valence or core) to anunoccupied state located in the conduction band. Electron–holeexcitation initiates the onset of the loss spectrum, which can beused to determine the bandgap of oxide films. In the literature[14–18], various techniques have been applied to retrieve theonset energy from the TEM-VEELS spectrum. We applied threemajor techniques to our experimental spectra and estimated theonset energies for each technique. The first method entails a directreading of the onset energy on a logarithmic scale without ZLPsubtraction. The second method obtains the single scatteringdistribution (SSD) by ZLP subtraction and plural scattering, andthen fits and estimates the onset energy. Finally, the third method,which is used frequently in AES-REELS, applies a linear fit anddirectly reads the cross point between the extrapolated line andthe bottom line.

A thermally grown sample was prepared as a TEM specimenthrough the conventional polishing and the ion-milling method,imparting a wedge shape of uneven thickness. Fig. 2 shows thelow-loss spectra of SiO2 at the different thicknesses of thespecimen, obtained by monochromated STEM-EELS. The totalacquisition time was 1.0 s (0.1 s�10 times). The normalizedintensity at the onset was on the order of 10�3 to 10�4,according to the thickness variation. As the thickness decreased,the ‘‘apparent’’ bandgap widened and approached the knownbandgap of 8.9 eV at room temperature. Even at a thickness of45 nm, however, the onset of the spectrum was 8.5 eV, and did notreach the bandgap of SiO2. We found that this method of directlyreading the onset energy could underestimate the bandgap valueeven on this logarithmic scale, because the onset of the spectrumcould start early, before the real bandgap, due to Cerenkov lossnear 10 eV [7,13,26]. Therefore, this direct reading method wasdeemed to be relatively unreliable for measuring the bandgaps ofdielectric materials, including SiO2, especially those showingCerenkov effects.

Next, in order to test the ZLP subtraction method, we processedthe low-loss spectra to obtain SSDs. ZLP subtraction from the EELSspectra and Fourier-log deconvolution were carried out usingelectronic structure tools (EST) including the SS_COR programs[19]. For all of the EELS spectra, the center of the ZLP was fitted forthe zero energy calibration. Subsequently, the spectra werecorrected for multiple scattering events by Fourier-log deconvolu-

Fig. 2. The low-loss spectra of silicon dioxide acquired using the monochromated

STEM-EELS. As the thickness decreases, bandgap values approached the SiO2

bandgap of 8.9 eV. Even at the thickness of 45 nm, the onset of 8.5 eV does not

reach the bangap of silicon dioxide.

tion. By this procedure, the wings of the ZLP were fitted separatelywith an asymmetric Pearson VII line shape and subtracted. TheSSDs of the SiO2 finally obtained are shown in Fig. 3. It was foundthat as the thickness increased, the intensities of the humps at5–7 and 8–10 eV inside the bandgap region increased, after whichthe onset energy of each spectrum also decreased. Couillard et al.reported that the 4–10 eV hump near the bandgap (�10 eV) is dueto Cerenkov effects in SiO2 [27]. Therefore, it can be understoodthat after ZLP subtraction and plural scattering, bandgapdetermination remains difficult, because the onset points can bechanged by the hump near the bandgap (that is, by the Cerenkoveffects) as the sample thickness changes.

Cerenkov radiation is emitted when the electron velocity v

exceeds the speed of light or, more precisely, its phase velocity inthe medium through which it is moving. The electric fieldsurrounding a moving charged particle displaces and polarizesthe electrons in the atoms of the medium traveled through. Whenthe polarized medium restores itself after the electron has passedthrough, virtual photons are emitted and interfere constructively,leading to the emission of light at the cost of energy loss from thebeam electron. According to the literature [9], in order to reduceCerenkov loss, both the dielectric constant and the voltage shouldbe lower. In our case, the dielectric constants of most of thedielectrics were relatively higher, and decreasing the TEMaccelerating voltage was practically difficult due to the tediousalignments necessitated by the voltage changes. Therefore, aneasier and simpler method of bandgap measurement remains tobe found.

Finally, we applied the linear fit method, which is frequentlyused in AES-REELS [20]. By this method, the bandgap energy canbe estimated from the intersection of a straight line originatingfrom the background level with a linear fit to the onset of the losssignal spectrum. The same EELS spectra shown in Fig. 3 areredrawn on the linear scale in Fig. 4. As shown in Fig. 4, as thethickness increased, the entire intensity and slope of the low-lossspectrum increased, and furthermore, the hump intensity due tothe Cerenkov and surface effects near 10 eV also increased.However, the crossing point for each spectrum was keptconstant at 8.9 eV for the different thicknesses. From the results,it was found that the linear fit method is considerably reliable inestimating the bandgaps of SiO2 materials, despite persistentCerenkov loss and surface effects. Furthermore, it was found that

ARTICLE IN PRESS

Fig. 4. The bandgap measurements of silicon oxide using monochromator at

various thicknesses. At different thickness, crossing points are consistently at

�8.9 eV.

Fig. 5. AES-REELS spectra taken with primary electron beam of 1 keV for SiO2 thin

films. Note that the bandgap is 9.0 eV which is similar to the bandgap estimated

from STEM-EELS.

Fig. 6. Random RBS spectra of a-SiNx with different compositions. Black lines are

simulated lines for quantitative composition analysis.

Table 1Compositions of a-SiNx determined by RBS.

N Si N/Si

SiNx1 59.1 40.9 1.4670.07

SiNx2 54.5 45.5 1.2070.06

SiNx3 47.9 52.1 0.9270.05

J. Park et al. / Ultramicroscopy 109 (2009) 1183–11881186

the linear fit method is potentially applicable also to otherdielectric materials including noisy loss in VEELS spectra.

To compare the bandgap-measurement results by STEM-EELS,the AES-REELS spectra of a-SiNx were acquired with a 1 keVelectron beam. Fig. 5 shows the AES-REELS spectrum of SiO2. Thebandgap was 9.0 eV, estimated with the linear fit method. Theonset slope and shape of the low-loss spectrum were similar tothose for the STEM-EELS.

3.3. Bandgap measurement of a-SiNx

The a-SiNx films from various SiH4:NH3:N2 gas mixtures weredeposited by the PECVD method. The chemical composition of theamorphous silicon nitride was determined by Rutherford back-scattering and AES. The RBS measurements showed that anincrease of the SiH4 flow rate caused an increase in the siliconconcentration in the thin films (Fig. 6). Fig. 6 shows the RBSexperimental spectra as well as the simulated spectra of thea-SiNx deposited at the different SiH4 gas flow rates: 200, 300 and

400 sccm. The compositions and N/Si ratios obtained from the RBSmeasurements are summarized in Table 1. The N/Si ratios of threea-SiNx samples were 1.46, 1.20 and 0.92 for the SiH4 gas flow ratesof 200, 300 and 400 sccm, respectively.

Fig. 7(a) shows the bandgap spectra of a-SiNx with the variousN/Si ratios obtained by the monochromated STEM-EELS. Theirplasmon loss energies were 19.4, 21.2 and 22.7 eV, in sequence.The bandgaps of the three a-SiNx were estimated by the linear fitmethod (Fig. 7(b)). The bandgaps of the a-SiNx with the N/Si ratiosof 1.46, 1.20 and 0.92 were determined to be 5.3, 4.1 and 2.9 eV,respectively. The bandgap values estimated by the linear fitmethod were almost the same as the bandgap values of a-SiNxestimated by the direct onset method on the logarithmic scale(not shown here). As the N/Si ratio increased, the bandgapwidened. As shown in Fig. 7(a), as the nitrogen content increased,the bandgap and plasmon energy also widened and increased,respectively. The plasma energy increased with the wideningbandgap and the increasing density of the valence electrons [22].The energy gap widened with increasing nitrogen concentration,due to, first, the substitution of Si–Si bonds, located on the top ofthe valence band, Si–N bonds, positioned deeper in the valenceband, and second, the displacement of the bottom of theconduction band, due to the substitution of the Si–Siantibonding states for the Si–N antibonding ones [21]. Thedensity of the valence electrons also increased with increasingnitrogen concentration. Therefore, both the plasmon energyand the bandgap should widen with increasing nitrogenconcentration.

Fig. 8 shows the AES-EELS spectra of a-SiNx with various N/Siratios. The a-SiNx samples were bombarded with a focused low-energy electron beam, and the energy distribution of the reflectedelectrons was measured as 0–100 eV. The bandgap was estimatedfrom the intersection of a straight line originating from thebackground level with a linear fit to the onset of a loss signal

ARTICLE IN PRESS

Fig. 7. STEM-EELS spectra of a-SiNx showing (a) the plasmon loss peaks and (b)

the bandgap measurements at various compositions. Note that as the nitrogen

content increases, the bandgap and plasmon energy also increase.

Fig. 8. AES-REELS spectra taken with primary electron beam of 1 keV for a-SiNx

thin films. The bandgaps of a-SiNx with N/Si ratios of 1.46, 1.20 and 0.92 were

measured as 4.7, 4.1 and 3.1 eV, respectively.

Fig. 9. Variation of electronic energy gap (eV) with the nitrogen content (x) in a-

SiNx films. The bandgaps of various a-SiNx obtained from STEM-EELS and AES-

REELS in this work were compared with the experimental optical gaps of other

research groups [22–25]. Note that the electronic band gap increases with nitrogen

content.

J. Park et al. / Ultramicroscopy 109 (2009) 1183–1188 1187

spectrum. This crossing point yields the bandgap values. Thebandgaps of a-SiNx with N/Si ratios of 1.46, 1.20 and 0.92 weremeasured as 4.7, 4.1 and 3.2 eV, respectively. Their plasmon loss

energies were 20.8, 21.1 and 21.8 eV, in sequence. These resultswere similar in the trend to, but slightly different in the values ofthe plasmon and bandgap energies from the TEM-EELS results.Basically, it was thought that the difference in the bandgap andplasmon energies between TEM-EELS and AES-EELS was in theerror range. The difference presumably reflected the facts that AEShas a larger electron beam, even if it is focused, as well as a muchlarger scan area and higher energy resolution than TEM, andincludes some charging effects. In the case of TEM-EELS, thedifference could result from the difficulty in accurately readingthe onset, due to the lack of the linear region in TEM EELS spectra.

The bandgaps of various a-SiNx materials obtained using thelinear fit method for STEM-EELS and AES-REELS were comparedwith the experimental optical gaps of other research groups[21–25], as shown in Fig. 9. The bandgap values measured in thepresent experiment were in agreement with the bandgap valuesobtained by the other groups, and also showed a similar trend: thewidening of the electronic gap with nitrogen content [21–23].Therefore, it was speculated that the linear-fit methods areapplicable to SiNx materials even if, as revealed by comparisonwith AES-measurement results, some errors occur.

4. Conclusions

STEM-EELS bandgap measurements of SiO2 and a-SiNx di-electric films were conducted using monochromated electrons.The energy spread of a monochromated zero-loss peak was0.13 eV in full-width at half maximum, with an energy dispersionof 0.01 eV/ch. In the case of SiO2, with the direct reading methodand the ZLP subtraction method, the determination of thebandgap was slightly difficult, because the onset points weremoved by the humps, at 5–7 and 8–10 eV, caused by Cerenkoveffects. The linear fit method was more reliable than theaforementioned methods, because it could overcome the effectsof specimen thickness as well as the Cerenkov effect. In the case ofSiNx, we compared the monochromated STEM-EELS bandgapresults with those by AES-REELS for the a-SiNx with N/Si ratios of1.46, 1.20 and 0.92. The comparison showed a slight difference

ARTICLE IN PRESS

J. Park et al. / Ultramicroscopy 109 (2009) 1183–11881188

between the two methods but also a similarity to the value rangesreported by other groups.

Thus, the linear fit method with STEM-EELS was considered to bereliable in estimating the bandgaps of SiO2 dielectric materials, butwas determined to be of only limited applicability to SiNx materials,for which an appropriate linear spectral range could not be found.

References

[1] T.S. Chen, K.H. Wu, H. Chung, C.H. Kao, IEEE Electr. Dev. Lett. 25 (2004) 205.[2] M. Terauchi, M. Tanaka, K. Tsuno, M. Ishida, J. Microsc. 194 (1999) 203.[3] R. Erni, N.D. Browning, Ultramicroscopy 104 (2005) 176.[4] R. Erni, N.D. Browning, Ultramicroscopy 107 (2007) 267.[5] R. Erni, N.D. Browning, Ultramicroscopy 108 (2008) 84.[6] R. Pantel, M.C. Cheynet, F.D. Tichelaar, Micron 37 (2006) 657.[7] Lin Gu, Vesna Srot, Wilfried Sigle, Christoph Koch, Peter van Aken, Ferdinand

Scholz, Sarad B. Thapa, Christoph Kirchner, Michael Jetter, Manfred Ruhle,Phys. Rev. B 75 (2007) 195214.

[8] S. Lazar, G.A. Botton, M.Y. Wu, F.D. Tichelaar, H.W. Zandbergen, Ultramicro-scopy 96 (2003) 535.

[9] P.A. Cerenkov, Dokl. Akad. Nauk. SSSR 2 (1934) 451.[10] C.H. Chen, J. Silcox, R. Vincent, Phys. Rev. B 12 (1975) 64.

[11] J.M. Cowley, Phys. Rev. B 25 (1982) 1401.[12] P.A. Midgley, Ultramicroscopy 76 (1999) 91.[13] M. Stoeger-Pollach, H. Franco, P. Schattschneider, S. Lazar, B. Schaffer,

W. Grogger, H.W. Zandbergen, Micron 37 (2006) 396.[14] K. Kimoto, G. Kothleitner, W. Grogger, Y. Matsui, F. Hofer, Micron 36

(2005) 185.[15] S. Lazar, G.A. Botton, H.W. Zandbergen, Ultramicroscopy 106 (2006) 1091.[16] R. Rafferty, L.M. Brown, Phys. Rev. B 58 (1998) 10326.[17] B. Rafferty, S.J. Pennycook, L.M. Brown, J. Electron Microsc. 49 (2000) 517.[18] S. Schamm, G. Zanchi, Ultramicroscopy 96 (2003) 559.[19] A.D. Dorneich, R.H. French, H. Mullejans, S. Loughin, M. Ruhle, J. Microsc. 191

(3) (1998) 286.[20] Hua Jin, Suhk Kun Oh, Hee Jae Kang, Surf. Interface Anal. 38 (2006) 1564.[21] J.F. Justo, F. de Brito Mota, A. Fazzio, Phys. Rev. B 65 (2002) 73202.[22] M.M. Guraya, H. Ascolani, G. Zampieri, J.I. Cisneros, J.H. Dias da Silva,

M.P. Cantao, Phys. Rev. B 42 (1990) 5677.[23] N. Banerji, J. Serra, S. Chiussi, B. Leon, M. Perez-Amor, Appl. Surf. Sci. 168

(2000) 52.[24] S. Hasegawa, M. Matuura, Y. Kurata, Appl. Phys. Lett. 49 (1986) 1272.[25] Z. Yin, F.W. Smith, J. Non-Cryst. Solids 137 (1991) 879.[26] U. Bangert, A. Harvey, J. Davison, R. Keyse, C. Dieker, J. Appl. Phys. 83 (1998)

7726.[27] M. Couillard, M. Kociak, O. Stephan, G.A. Botton, C. Colliex, Phys. Rev. B 76

(2007) 165131.