Site-Specific Atomic and Electronic Structure Analysisof Epitaxial Silicon Oxynitride Thin Film on SiC(0001)

by Photoelectron and Auger Electron Diffractions

Naoyuki Maejima1, Fumihiko Matsui1+, Hirosuke Matsui1, Kentaro Goto1,Tomohiro Matsushita2, Satoru Tanaka3, and Hiroshi Daimon1

1Graduate School of Materials Science, Nara Institute of Science and Technology (NAIST), Ikoma, Nara 630-0192, Japan2Japan Synchrotron Radiation Research Institute (JASRI), SPring-8, Sayo, Hyogo 679-5198, Japan

3Department of Applied Quantum Physics and Nuclear Engineering, Kyushu University, Fukuoka 819-0395, Japan

(Received December 19, 2013; accepted February 5, 2014; published online March 28, 2014)

The film and interface structures of epitaxial silicon oxynitride (SiON) thin film grown on a SiC(0001) surface wereinvestigated by photoelectron diffraction. Forward focusing peaks (FFPs) corresponding to the directions from thephotoelectron emitter atom to the surrounding atoms appeared in the photoelectron intensity angular distribution (PIAD).By comparing N 1s PIAD with those of Si 2p and C 1s, we confirmed that the nitrogen atoms at SiON/SiC interfacereplace carbon atoms at stacking fault sites. Two kinds of oxygen atom sites exist in the previously proposed model[T. Shirasawa et al.: Phys. Rev. Lett. 98, 136105 (2007)]. FFP corresponding to Si–O–Si perpendicular bonds wasobserved in the O 1s PIAD, while diffraction rings were observed in the KLL Auger electron intensity angulardistribution (AIAD), which were attributed to the diffraction patterns from outermost oxygen sites. Furthermore, O K-edge X-ray absorption spectra combined with AIAD were analyzed. An electronic structure specific to each oxygenatom site was successfully separated.

1. Introduction

The development of power devices is an urgent contem-porary issue. Silicon carbide (SiC) is a key material forfinding such a solution owing to its wide band gap, highbreak down field, and high electron mobility.1) Furthermore,thermal silicon oxide layers can be grown on SiC, which iscommon to Si surfaces. Accordingly, processing techniquesin the Si industry is applied to SiC device fabrications. Inreality, the conventional oxide films have a few nm-thicktransition layer at the interface, which seriously affects thetransport property.2) An epitaxial silicon oxide layer withan atomically sharp interface was reported.3) However, theepitaxial silicon oxide layer on SiC(0001) has a danglingbond per unit cell at the interface.3,4) This problem isexpected to be solved by nitrogen incorporation leading tothe formation of an epitaxial oxynitride (SiON) layer thatcompletely lifts the interfacial dangling bond.5) Nitrogenatom, which have one lone pair electron more than theoxygen atom, removes the interface defect state.

Shirasawa et al. have reported an investigation of a newstable epitaxial SiON thin film on 6H-SiC(0001)6) byscanning tunneling microscopy and low energy electrondiffraction (LEED) I–V measurements. They suggested anatomic structure model of the epitaxial SiON thin film onSiC(0001) that consist of Si2O5 thin film, a Si2N3 layer at theinterface, and a SiC substrate. In addition, in the Si2O5 thinfilm, tridymite- and cristobalite-like 180° Si–O–Si bondscoexist with quartz-like 144° Si–O–Si bonds. The dangling-bond free interface was achieved in their model. Recently,vertical atomic layer distances have been confirmed by X-raydiffraction.7)

It is essential to characterize local electronic structures tounderstand the origin of the epitaxial SiON thin-film stability.The atomically abrupt band offset at the interface of SiON/SiC(0001) was confirmed by X-ray absorption spectroscopy(XAS) and X-ray emission spectroscopy (XES) together with

theoretical calculation.8) XAS measurements were performedin the total electron yield (TEY) and Auger electron yield(AEY) detection modes. AEY detection was expected to bemore surface sensitive than TEY detection due to the finitemean escape depth of Auger electron. Two kinds of oxygenatoms were placed at sites of different depth in the oxidelayer. However, the surface sensitivity difference in XAS bythe AEY and TEY detection modes was not sufficient toresolve the electronic structure for two different oxygen sitesthen.

Photoelectron diffraction is a powerful element-selectiveatomic structure analysis method, which has direct access toburied subsurfaces and interfaces without destroying sur-faces. Site-specific X-ray photoelectron spectroscopy (XPS)and XAS can be obtained by analyzing forward focusingpeaks (FFPs) and diffraction patterns as site specific probes.For example, the electronic and magnetic structures of eachatomic layer of a Ni thin film was investigated using atomiclayer specific FFPs for resolving site-specific XAS spec-tra.9,10) In an previous work,11) we measured Si 2p and C 1sPIADs from the vicinal 6H-SiC(0001) substrate with anepitaxial SiON thin film. There are two different Si and Csites in SiC crystals owing to stacking faults. Owing to theanisotropic step bunching along the [11�20] direction resultingin a preferential appearance of terraces with one type of localatomic site,12) threefold symmetric photoelectron intensityangular distributions (PIADs) predominantly originatingfrom top layers were obtained. We developed a numericalmethod of deducing PIAD from each site by solving aninverse matrix. In this work, we investigated the atomic andelectronic structures of the epitaxial SiON thin film byphotoelectron/Auger electron diffraction and spectroscopy.N 1s PIAD from epitaxial SiON thin film also showedthreefold symmetry. The structure of nitrogen atomsoccupying carbon stacking fault sites at the SiON/SiCinterface was confirmed. We succeeded in characterizing theatomic and electronic structures of individual oxygen sites by

Journal of the Physical Society of Japan 83, 044604 (2014)

http://dx.doi.org/10.7566/JPSJ.83.044604

044604-1 ©2014 The Physical Society of Japan

combining Auger electron intensity angular distribution(AIAD) measurements and XAS.

2. Experiments

The experiments were performed at the circularly polarizedsoft-X-ray beamline BL25SU of SPring-8, Japan.13) Theepitaxial SiON thin film was grown on the 6H-SiC(0001)surface with an off angle of 4° towards the [1�100]direction.11) The sample was introduced into an ultra-highvacuum chamber and no further sample surface treatmentwas applied. All experiments were performed at roomtemperature.

PIADs from the sample were measured using a two-dimensional display-type spherical mirror analyzer (DI-ANA).14–16) The acceptance angle of the analyzer was �60°.Circularly polarized light was incident along the direction 45°off the surface normal in the O 1s PIAD measurement, whilenormal incident geometry was used for all the othermeasurements. In the case of normal incidence, the emissionangle (ª) dependence from 45� 60° relative to the surfacenormal was measured simultaneously. By scanning thesample azimuth over 360°, 2�-steradian PIAD data werecollected. A set of 2� steradian PIADs excited by �þ and ��helicity lights was measured by switching the path of storagering electrons in twin helical undulators at 0.1Hz.17) Angle-resolved constant-final-state (CFS) mode photoelectronspectra and X-ray absorption spectra were obtained byvarying photon energy at a fixed kinetic energy.

The density of states and molecular orbital calculationswere performed using the first-principles program SCAT,a discrete varitional X� molecular orbital method,18) tointerpret the experimental results. The molecular orbital resultwas explained by the Mulliken method. The final values ofelectrical densities were calculated until the result convergesto the initial value assumed before the numerical basisfunction was obtained.

3. Results and Discussion

Figure 1 shows the angle-resolved CFS-mode photoelec-tron spectra at a kinetic energy of 600 eV. Substratecomponents (Si 2p and C 1s) and epitaxial SiON thin-film-specific components (N 1s and O 1s) were observed. Thepeak intensity ratio of N 1s to O 1s was smaller under thesurface-sensitive grazing angle condition. This result indi-cates that the oxide layer was grown on the nitride layer.

We have measured the 2�-steradian Si 2p and C 1s PIADswith a photoelectron kinetic energy at 600 eV using photonenergies of 708 and 889 eV, respectively. In the bulk crystal,pairs of mirrored local atomic sites with respect to the f1�100gplane exist and the chemical environments surrounding eachsite are equivalent. However, all the measured patternsshowed a threefold symmetry owing to the anisotropic stepbunching along the [11�20] direction resulting in a preferentialappearance of terraces with one type of local atomic site.12)

Taking the finite inelastic mean free path of photoelectronsinto account, photoelectron patterns for one kind each of theSi and C atom sites were successfully derived.11) Figure 2(a)shows the C 1s PIAD from one kind of atomic site near theinterface. The surface normal direction at the center of thePIAD matches the sample rotation axis. The raw data werenormalized by their polar angle intensity profile obtained by

averaging their azimuthal intensity variations. The inhomo-geneity of the detector can be removed during this normal-ization. The pattern in the normal direction disappears afterthis procedure.

N 1s PIAD [Fig. 2(b)] with a photoelectron kinetic energyof 600 eV was measured using an excitation photon energy of1003 eV. However, the signal-to-background ratio at N 1swas very small because of the energy-loss electron back-ground as shown in Fig. 1. We measured the backgroundPIAD, which has the same kinetic energy and excitationphoton energy that is 20 eV lower than that of N 1s. Aprimary N 1s PIAD pattern was obtained by subtracting thisbackground pattern in order to remove undesired energy-losseffects.19)

The N 1s PIAD pattern was compared with the Si 2p[Fig. 3(b) in Ref. 11] and C 1s [Fig. 2(a)] PIAD patterns.Note that the C 1s PIAD pattern for the topmost SiC unit sitehas a similar feature to the N 1s PIAD pattern, but is mirrorsymmetric with respect to the f1�100g plane. Three FFPs

Fig. 1. Angle-resolved constant final state mode photoelectron spectra ofthe epitaxial SiON thin film on 6H-SiC(0001). The photoelectron kineticenergy was fixed at 600 eV and photon energy was varied. The intensity wasnormalized by C 1s peaks. The signal-to-background ratio at N 1s was 0.07.

Fig. 2. (Color online) (a) Topmost unit site C 1s PIAD,11) (b) N 1s PIADpatterns, and (c) N 1s simulated pattern. In the simulation, the structuresuggested by Shirasawa et al. was used. The FFPs corresponding to the first-and second-nearest-neighbor atoms were indicated by the circles and dots,respectively. The local atomic structure models expected from C 1s and N 1sare shown in (d) and (e), respectively, and the model used for the simulationis shown in (f ).

J. Phys. Soc. Jpn. 83, 044604 (2014) N. Maejima et al.

044604-2 ©2014 The Physical Society of Japan

corresponding to the first-nearest-neighbor atoms indicatedby circles as well as three FFPs corresponding to the second-nearest-neighbor atoms indicated by dots were observed inthe h11�1i and h011i directions in the C 1s PIAD, respec-tively.11) In contrast, FFPs corresponding to the first- andsecond-nearest-neighbor atoms in N 1s PIAD were observedin the directions mirror-symmetric with respect to the f1�100gplane. These results indicate the substitution of carbon atomswith nitrogen atoms at a stacking fault site, as shown inFigs. 2(d) and 2(e). Although a nitrogen atom form three Si–N bonds with two Si atoms above and one Si atom directlybeneath, there are three nitrogen atoms in one unit cell at theinterface resulting in a three-fold symmetric PIAD pattern.This is also shown in the simulated pattern in Fig. 2(c), andthe atomic configuration observed around the nitrogen atomis in agreement with that observed in the N atom site in theepitaxial SiON thin film, as shown in Figs. 2(e) and 2(f ).These observations are in good agreement with the suggestedatomic structure model by Shirasawa et al.6)

The FFP directions from the emitter atom to the scattereratoms in the PIADs obtained using different helicity lightsrotate around the incident light axis. The rotation angle �� iswell described by Daimon’s formula:20)

�� ¼ tan�1 m

kR sin2 �out; ð1Þ

where m and k are the magnetic quantum number andthe wave number of photoelectron, respectively. R is theinteratomic distance between the emitter and scatterer atoms,and �out is the angle between the incident photon directionand the outgoing direction of the emitted photoelectrons.In general, the effective magnetic quantum number m� isused,20) considering the contribution of the transitionprobability from different mcore initial states at a particularangle. In the case of O 1s excitation, m� is 1.

Figure 3(a) shows the circular dichroism in O 1s photo-electron angular distributions (CDADs) from the Si2O5 layer.The photoelectron kinetic energy was 600 eV and theexcitation photon energy was 1138 eV. Red cross indicatesthe direction of incident circularly polarized light. The whiteand black pattern at the center is the result of an FFProtational shift. Here we show that this feature is wellreproduced by photoelectron diffraction simulation. We useda multiple scattering simulation code, TMSP, developed byone of our authors (T. Matsushita).21) O 1s PIADs excited by�þ and �� helicity circularly polarized lights from a Si–O–Silinear cluster with a Si–O bond length of 0.163 nm weresimulated, and a CDAD pattern was obtained. Note thatthe observed CDAD pattern is exactly reproduced in thesimulated pattern shown in Fig. 3(b). An oxygen photo-electron emitter atom corresponds to the atom indicated asO2 in the third layer, as shown in Fig. 3(c). The FFP circulardichroism pattern in the [0001] direction correspond to the Siatom indicated as Si1 in the second layer. Dots in Fig. 2(b)correspond to O2 atoms seen from the N atom at theinterface. Therefore, we succeeded in the direct observationof the 180° Si–O–Si bond.

Figure 4(a) shows an O KLL 2�-steradian AIAD with aphotoelectron kinetic energy of 504 eV and an excitationphoton energy of 540 eV. The measured O KLL AIAD isin good agreement with the simulated pattern shown in

Fig. 4(b) for the epitaxial SiON structure model shown inFig. 4(c). The six FFP patterns indicated by black dots at apolar angle of 36° correspond to O1 atoms in the first layerseen from the O2 atom in the third layer. The other diffractionpatterns can be explained by the overlaps of diffraction ringsindicated as R1, R2, and R3. These diffraction ringscorrespond to the in-plane O–O atom scatterings of r1, r2,and r3 indicated in Fig. 4(c).

Finally, we measured the O K-edge XAS at normalincidence and emission polar angles from ¹15 to 90°.Figure 5(a) shows the emission angle dependent XAS every10° from 5 to 85°. In the previous work, the XAS dataobtained in the TEY and AEY detection modes did not showmuch difference.8) On the other hand, we succeeded indetecting a gradual change as function of emission polarangle, as shown in Fig. 5(a). The intensity variation was 2%at a photon energy of 538 eV and the energy position of theabsorption maximum at approximately 540 eV was shifted by0.2 eV. Since the mean escape depth of Auger electrons variesby emission angle, the spectral changes suggest the differencein the local electronic structure between the first- and third-layer oxygen atom sites.

Therefore, we analyzed the emission-angle-resolved XASdata to obtain information specific for each site. Themeasured intensity I�x for the emission angle �x is the sum

Fig. 3. (Color online) (a) O 1s circular dichroism angular distributionpattern and (b) the corresponding multiple scattering simulation pattern. Thered cross shows the direction of incident light. (c) Side view of atomicstructure model of the epitaxial SiON thin film on 6H-SiC(0001) byShirasawa et al.8)

J. Phys. Soc. Jpn. 83, 044604 (2014) N. Maejima et al.

044604-3 ©2014 The Physical Society of Japan

of Auger electron intensities from the first- and third-layer Oatoms and can be expressed as

I�0 ¼3I1st þ 2AI3rd

3þ 2A; ð2Þ

I�1 ¼3I1st þ 2BI3rd

3þ 2B; ð3Þ

A ¼ exp � d

� cos �0

� �; ð4Þ

B ¼ exp � d

� cos �1

� �; ð5Þ

where d is the depth of the third-layer oxygen atom site(0.215 nm) and ­ is the Auger electron inelastic mean freepath (1.79 nm). In order to separate the spectra for the first-and third-layer O atom sites, I1st and I3rd , respectively, wesolved the following equation:

ð3þ 2AÞI�0ð3þ 2BÞI�1

!¼ 1 A

1 B

!3I1st

2I3rd

!: ð6Þ

Each layer spectrum obtained from this equation isnormalized per atomic site. The separated XAS data for thefirst- and third-layer O atom sites are shown in Fig. 5(b) asthe black open circles and red solid squares, respectively. Theabsorption edge is at 536.2 eV (peak ¡) in both spectra. Abroad peak appeared at 539.3 eV (peak £) in the first layerspectrum, while two absorption peaks were observed at538.1 eV (peak ¢) and 540.5 eV (peak ¤) in the third-layerspectrum. Note that the edge energies of both spectra werethe same. This result coincides with the calculation resultssuggesting that the conduction band minima are the same forthe two different SiO layers.8,22–24)

Furthermore, we calculated the electronic structure corre-sponding to each peak shown in Fig. 5(b) using the SCATcode. SiH3OSiH3 clusters with bond angles of 144 and 180°were used to represent the first- and third-layer oxygen atomsites, respectively. The present normal incident geometry issensitive to the transition from the O 1s orbital to the in-planeO 2p orbital. Here, molecular orbitals with considerablecontribution of the in-plane O 2p orbital are selected.

Fig. 4. (Color online) (a) 2�-steradian O KLL Auger electron intensityangular distribution (AIAD) pattern and (b) simulated result. There are five Oatom sites, thus the AIAD from these sites were summed.

(a)

(b)

δγβ

α

1

2

3

4 5

6 model 144

model 180

Fig. 5. (Color online) (a) Result of angle-resolved Auger-electron-yieldX-ray absorption spectrum. (b) Layer-resolved X-ray absorption near edgestructure spectrum with molecular orbitals calculated by DV-X¡. The normalincident circularly polarized light can excite the molecular orbital thedirection of which is toward the surface parallel. The red and black bars arethe O 2p density of states of the Si–O–Si 144 and 180° atomic models. Eachmolecular orbital is shown over the bar.

J. Phys. Soc. Jpn. 83, 044604 (2014) N. Maejima et al.

044604-4 ©2014 The Physical Society of Japan

Molecular orbital 3 expands in the h11�20i direction andcorresponds to the £-peak. Molecular orbitals 2 and (4+5)expand in the surface-parallel direction around the third-layeroxygen atom. Thus, they are attributed to the ¢- and ¤-peaks.

Tanaka et al. calculated the density of states for ¡-quartzhaving 144° Si–O–Si bonds and for ¢-cristobalite having180° Si–O–Si bonds.25) Qualitatively, the present clustercalculation results are in agreement with theirs. Since theepitaxial SiON thin film is a complex structure of ¡-quartzand ¢-cristobalite, a crystal long-range order does not exist.This seems to be the reason why the result of the simulationusing small clusters qualitatively explains the site-resolvedXAS data.

Tridymite and cristobalite are stable phases in the temper-ature range of 1000–1700 °C. The epitaxial SiON thin filmwas grown on a SiC substrate at a temperature of 1350 °C.Thus, it is natural to have 180° Si–O–Si bonds on such asurface. However, the distance between neighboring N atomsis about 0.3 nm, which is too small for the further tridymitecrystal growth, but sufficient for the 144° Si–O–Si bonds toterminate the surface. The resulting the epitaxial SiON thinfilm has no dangling bonds and have a low strain, which arethe origin of the epitaxial SiON thin-film stability on thissurface.

4. Conclusions

In summary, we measured photoelectron diffractionpatterns and X-ray absorption spectra of the SiON thin filmepitaxially grown on a SiC(0001) 4° off substrate preferen-tially terminated by one kind of terrace. The observed N 1sPIAD was similar to the C 1s PIAD, but mirror-symmetric.This observation is a strong indication of N atoms replacingC atoms in the outermost stacking fault layer and forming anabrupt SiON interface. Furthermore, we detected 180 and144° Si–O–Si bonds in the O 1s CDAD and O KLL AIADpatterns. Finally, we succeeded in resolving O K-edge XASfor different oxygen atomic sites. The present photoelectrondiffraction approach was shown to be useful for character-izing buried interface defect atomic and electronic structures.

Acknowledgments

We appreciate the grateful support by Dr. TetsuyaNakamura and Dr. Takayuki Muro. This research wasperformed at the Japan Synchrotron Radiation ResearchInstitute (Proposal Nos. 2009A1753 and 2009B1769).

+matui@ms.naist.jp1) W. H. G. Pensl, Silicon Carbide Recent Major Advances (Springer,

Berlin, 2004) Part I.2) V. V. Afanasev, M. Bassler, G. Pensl, and M. Schulz, Phys. Status

Solidi A 162, 321 (1997).3) J. Bernhardt, J. Schardt, U. Starke, and K. Heinz, Appl. Phys. Lett. 74,

1084 (1999).4) M. Schürmann, S. Dreiner, U. Berges, and C. Westphal, Phys. Rev. B

74, 035309 (2006).5) V. V. Afanas’ev, A. Stesmans, F. Ciobanu, G. Pensl, K. Y. Cheong, and

S. Dimitrijev, Appl. Phys. Lett. 82, 568 (2003).6) T. Shirasawa, K. Hayashi, S. Mizuno, S. Tanaka, K. Nakatsuji, F.

Komori, and H. Tochihara, Phys. Rev. Lett. 98, 136105 (2007).7) T. Shirasawa, K. Sakamoto, T. Takahashi, and H. Tochihara, Surf. Sci.

605, 328 (2011).8) T. Shirasawa, K. Hayashi, H. Yoshida, S. Mizuno, S. Tanaka, T. Muro,

Y. Tamenori, Y. Harada, T. Tokushima, Y. Horikawa, E. Kobayashi, T.Kinoshita, S. Shin, T. Takahashi, Y. Ando, K. Akagi, S. Tsuneyuki, andH. Tochihara, Phys. Rev. B 79, 241301 (2009).

9) F. Matsui, T. Matsushita, Y. Kato, M. Hashimoto, K. Inaji, F. Z. Guo,and H. Daimon, Phys. Rev. Lett. 100, 207201 (2008).

10) F. Matsui, T. Matsushita, and H. Daimon, J. Electron Spectrosc. Relat.Phenom. 178–179, 221 (2010).

11) F. Matsui, N. Nishikayama, N. Maejima, H. Matsui, K. Goto, M.Hashimoto, T. Hatayama, T. Matsushita, Y. Kato, S. Tanaka, and H.Daimon, J. Phys. Soc. Jpn. 80, 013601 (2011).

12) S. Nakamura, T. Kimoto, H. Matsunami, S. Tanaka, N. Teraguchi, andA. Suzuki, Appl. Phys. Lett. 76, 3412 (2000).

13) Y. Saitoh, H. Kimura, Y. Suzuki, T. Nakatani, T. Matsushita, T. Muro,T. Miyahara, M. Fujisawa, K. Soda, S. Ueda, H. Harada, M. Kotsugi,A. Sekiyama, and S. Suga, Rev. Sci. Instrum. 71, 3254 (2000).

14) H. Daimon, Rev. Sci. Instrum. 59, 545 (1988).15) H. Daimon and S. Ino, Rev. Sci. Instrum. 61, 57 (1990).16) M. Kotsugi, Y. Miyatake, K. Enomoto, K. Fukumoto, A. Kobayashi, T.

Nakatani, Y. Saitoh, T. Matsushita, S. Imada, T. Furuhata, S. Suga, K.Soda, M. Jinno, T. Hirano, K. Hattori, and H. Daimon, Nucl. Instrum.Methods Phys. Res., Sect. A 467–468, 1493 (2001).

17) T. Muro, T. Nakamura, T. Matsushita, H. Kimura, T. Nakatani, T.Hirono, T. Kudo, K. Kobayashi, Y. Saitoh, M. Takeuchi, T. Hara, K.Shirasawa, and H. Kitamura, J. Electron Spectrosc. Relat. Phenom.144–147, 1101 (2005).

18) H. Adachi, M. Tsukada, and C. Satoko, J. Phys. Soc. Jpn. 45, 875(1978).

19) F. Matsui, T. Matsushita, M. Hashimoto, K. Goto, N. Maejima, H.Matsui, Y. Kato, and H. Daimon, J. Phys. Soc. Jpn. 81, 013601 (2012).

20) H. Daimon, Phys. Rev. Lett. 86, 2034 (2001).21) T. Matsushita, F. Matsui, H. Daimon, and K. Hayashi, J. Electron

Spectrosc. Relat. Phenom. 178–179, 195 (2010).22) P. Krüger, B. Baumeier, and J. Pollmann, Phys. Rev. B 77, 085329

(2008).23) F. Devynck, Ž. Šljivančanin, and A. Pasquarello, Appl. Phys. Lett. 91,

061930 (2007).24) F. Devynck and A. Pasquarello, Surf. Sci. 602, 2989 (2008).25) I. Tanaka, J. Kawai, and H. Adachi, Phys. Rev. B 52, 11733 (1995).

J. Phys. Soc. Jpn. 83, 044604 (2014) N. Maejima et al.

044604-5 ©2014 The Physical Society of Japan