Induced ferromagnetism in Mn3N2 phase embedded in Mn/Si3N4 multilayers

J. Appl. Phys. 106, 043912 (2009); https://doi.org/10.1063/1.3203997 106, 043912 © 2009 American Institute of Physics. Induced ferromagnetism in phase embedded in multilayers Cite as: J. Appl. Phys. 106, 043912 (2009); https://doi.org/10.1063/1.3203997 Submitted: 20 April 2009 . Accepted: 12 July 2009 . Published Online: 26 August 2009 E. Céspedes, E. Román, Y. Huttel, J. Chaboy, J. García-López, A. de Andrés, and C. Prieto ARTICLES YOU MAY BE INTERESTED IN X-ray absorption and magnetic circular dichroism characterization of a novel ferromagnetic phase in multilayers Applied Physics Letters 93, 252506 (2008); https://doi.org/10.1063/1.3050534 Crystalline phase and orientation control of manganese nitride grown on MgO(001) by molecular beam epitaxy Journal of Applied Physics 91, 1053 (2002); https://doi.org/10.1063/1.1425435 Structural and magnetic properties of η-phase manganese nitride films grown by molecularbeam epitaxy Applied Physics Letters 78, 3860 (2001); https://doi.org/10.1063/1.1378800 Induced ferromagnetism in Mn3N2phase embedded in Mn/Si3N4 multilayers E. Céspedes,1E. Román,1Y. Huttel,1J. Chaboy,2J. García-López,3A. de Andrés,1and C. Prieto1,a兲 1Instituto de Ciencia de Materiales de Madrid, Consejo Superior de Investigaciones Científicas, Cantoblanco, 28049 Madrid, Spain 2Dep. de Física de la Materia Condensada and Instituto de Ciencia de Materiales de Aragón, CSIC-Universidad de Zaragoza, 50009 Zaragoza, Spain 3Centro Nacional de Aceleradores, Parque Tecnológico Cartuja’93, 41092 Sevilla, Spain 共Received 20 April 2009; accepted 12 July 2009; published online 26 August 2009兲 Room temperature ferromagnetism has been obtained for different sets of Mn/Si3N4multilayers prepared by sputtering. In order to find the most suitable conditions to stabilize the ferromagnetic ordering in this system, the evolution of the magnetic properties has been studied for films in which the Si3N4layer thickness was maintained constant while that of the Mn layer was varied, 关Mn共tm兲/Si3N4共3.4 nm兲兴n, and conversely, in 关Mn共0.7 nm兲/Si3N4共tsn兲兴43 samples, in which the Mn layer thickness was kept constant while varying the Si3N4layer thickness. Structural, compositional, electronic and magnetic characterizations have been performed by means of x-ray reflectometry, Rutherford backscattering spectrometry, x-ray photoemission spectroscopy, x-ray absorption, and superconducting quantum interference device for further knowledge of the magnetic-structural relationship in this system. Our results show that the peculiar magnetic behavior of these films is mainly related to the stabilization of a slightly distorted Mn3N2phase that is induced by the Si3N4 at the interfaces. For samples with larger Mn layer thickness, metallic Mn and Mn3N2phases coexist, which leads to a reduction of the total magnetization per Mn atom due to the presence of metallic Mn. For small Mn layer thickness 共tm⬍0.86 nm兲, where noncontinuous Mn3N2layers are formed, the magnetization decreases noticeably due to the superparamagnetic size limit. It has been found that the best conditions for the stabilization of the ferromagnetism in this system occur when both, the manganese-rich and the silicon nitride layers, are continuous and with similar thickness, close to 3.5 nm. © 2009 American Institute of Physics.关DOI: 10.1063/1.3203997兴 I. INTRODUCTION Manganese nitrides have gained attention among the 3d transition metal nitrides for their variety of electronic and magnetic properties which are suitable for potential applications.1–3Recently, these materials have also generated interest in the field of diluted magnetic semiconductors 共DMS兲as potential magnetic secondary phases within films of Mn-doped III-V semiconductors such as Mn:GaN, Mn- :GaAs, or MnInN.4–6 Manganese nitrides form a variety of stoichiometric phases 共␧, ␨ , ␩ , ␪ 兲exhibiting different magnetic properties.7 The most N rich phase, mononitride ␪ -MnN, has a tetragonally distorted rocksalt 共NaCl兲structure with an antiferromagnetic 共AF兲ground state.8It usually contains a considerable amount of N vacancies and is often labeled as Mn6N5+x. The closely related ␩ -Mn3N2phase also exhibits the NaCl structure but with an ordered array of N vacancies.8,9The Mn5N2and Mn2N ␨ -phases have hexagonal closed packed arrangements and present, respectively, ferromagnetic 共FM兲 and AF order.10 Finally, the ␧-Mn4N phase consists basically in a Mn face-centered cubic structure with an interstitial N and exhibits ferrimagnetism with a curie temperature 共TC兲as high as 745 K.11 These MnNxphases are currently related to inhomogeneities in ternary nitride-based alloys as the possible origin of the observed ferromagnetism. Recent studies have pointed toward Mn4N as a segregation phase showing the FM behaviors in Mn:GaN films.5Special attention has also been focused on the ␪ -MnN phase, since possible inclusions or MnN-rich regions in the matrix can have the same crystallographic structure as the semiconductor host. Although MnN has an AF ground state, it has been theoretically shown that FM order can be established depending on the MnN polymorph and the lattice parameters.6Band-structure calculations have been performed considering MnN in the hypothetical zinc-blende and wurtzite structures in order to get a better comprehension of these ternary nitride-based DMS systems.6,12 Ferromagnetism with TCof 354 K due to Mnrich precipitates in the zinc-blende phase has been recently predicted in Mn:GaN by using the mean-field approximation.4Spin-density-functional theory has been used to study GaN/MnN heterostructures composed by a thin layer of biaxially strained wurtzite 共w兲MnN intercalated with GaN layers as a new way to obtain FM wide-gap nitride systems for spintronic applications.13 FM order in this w-MnN on GaN has been predicted on the bases of structural relaxation at the GaN–MnN interface. Despite all these exciting properties, up to date, there have not been reported experimental efforts to obtain FM a兲Electronic mail: [email protected]. JOURNAL OF APPLIED PHYSICS 106, 043912 共2009兲 0021-8979/2009/106共4兲/043912/10/$25.00 © 2009 American Institute of Physics106, 043912-1 semiconductor/manganese nitride heterostructures. Even the interesting magnetic properties of MnNxfilms have been hardly experimentally studied, probably due to difficulties in sample preparation. Just a few attempts of fabrication of MnN by reactive sputtering10 and ion-assisted deposition,2 Mn3N2using reactive sputtering3and molecular beam epitaxy1and Mn4N, also by reactive sputtering,3,14 have been reported. Very recently, the magnetic properties of the Mn/Si3N4 system were experimentally studied prompted by the similarities with Mn/GaN, since silicon nitride also has a wideband-gap semiconducting behavior and can be a very stable host material with high dopant concentrations.15,16 We have reported, for the first time, ferromagnetism above room temperature 共RT兲in Mn/Si3N4multilayered films.17 The structural and electronic properties of few samples were studied by x-ray absorption spectroscopy 共XAS兲and x-ray magnetic circular dichroism, indicating the presence of a distorted noncentrosymmetric Mn3N2phase with slightly shorter Mn–N distances, which was proposed to be the origin of the ferromagnetism in this system. Here we present an extensive experimental study aimed at getting a better comprehension of the magnetic-structural relationship in the Mn/Si3N4system. We have grown by sputtering different sets of Mn/Si3N4films under similar experimental conditions as the previously reported ones. Different samples were prepared by varying the Mn:Si3N4rela- tive ratio. Evolution of the magnetic properties was studied for multilayered films with formula 关Mn共tm兲/ Si3N4共3.4 nm兲兴n, being tmthe Mn layer thickness and nthe number of bilayers, as well as for several films labeled as 关Mn共0.7 nm兲/Si3N4共tsn兲兴43, being tsn the Si3N4layer thickness, in order to find the most suitable conditions to stabilize the FM ordering in this system. II. EXPERIMENTAL AND COMPUTATIONAL METHODS Mn/Si3N4multilayered samples were prepared by sequential sputtering of Si3N4and Mn on Si共100兲substrates at RT. Si3N4layers were deposited by reactive rf-sputtering from a pure Si target 共99.999%兲with N2as reactive sputtering gas.18 The deposition rate was 2 nm/min for a rf power of 100 W. The residual pressure in the vacuum system was within the 1⫻10−7 mbar range. Mn was grown by dcsputtering from a Mn target 共99.95%兲with Ar as sputtering gas and a deposition rate of ⬃4nm/min for a dc power of ⬃10 W. Samples with nominal formula 关Mn共tm兲/Si3N4共3nm兲兴n and 关Mn共0.6 nm兲/Si3N4共tsn兲兴43 were prepared. For the first set of samples, the Si3N4layer thickness was maintained constant, while the thickness of the Mn layer 共tm兲was modified between 6 and 0.1 nm. In order to maintain the total amount of Mn constant, the number of bilayers 共n兲was progressively increased, from 5 to 75, as the Mn layer thickness was reduced. In the case of the 关Mn共0.6 nm兲/Si3N4共tsn兲兴43 samples, the Mn layer thickness was kept constant, while the Si3N4layer thickness 共tsn兲was varied from 6 to 0.7 nm. In this last case, the number of bilayers was always the same 共n=43兲. For the two sets of multilayers, the growth process started with Si3N4layer and terminated by depositing an additional 3 nm Si3N4layer on top. X-ray reflectivity 共XRR兲measurements were performed by using a Bruker D-8 diffractometer working with the Cu K ␣ wavelength. Rutherford backscattering spectroscopy 共RBS兲and nuclear reaction analysis 共NRA兲were carried out at the Centro Nacional de Aceleradores 共Sevilla, Spain兲. For the RBS experiments, a 2 MeV He2+ beam was employed, with beam diameter ⬃1 mm and scattering angle of 165°. The NRA measurements were done using the 14N共d, ␣ 1兲12C nuclear reaction with a 1.4 MeV D+beam and a detection angle of 150°. A 13 ␮ m thick Mylar foil was placed in front of the detector in order to stop the backscattered deuterons to avoid saturation of the electronic setup. For absolute N quantification of our samples, we have used as standard a 100 nm thick Si3N4film grown on Si substrate. Magnetic characterization was performed by using a commercial superconducting quantum interference device magnetometer 共Quantum Design MPMS-5S兲in the temperature range from 5 to 400 K. X-ray photoemission spectroscopy 共XPS兲measurements were performed in an ultrahigh-vacuum chamber with a base pressure of 1⫻10−10 mbar. The angle between the hemispherical analyzer 共Specs-PHOIBOS100兲and the plane of the surface was kept at 60° and the x-ray radiation was the Mg K ␣ line 共1253.6 eV兲. XPS spectra were recorded at the core levels of the elements present in the samples 共Mn, Si, N兲; in the case of Mn, Mn 3pspectra were recorded with a photon energy step of 0.1 eV and a pass energy of 15 eV in order to optimize the energy resolution for further fitting analysis and Mn species identification. Prior to the data analysis, the contributions of the Mg K ␣ satellite lines were subtracted and the spectra were subjected to a Shirley background subtraction formalism. The binding energy scale was calibrated with respect to the Si peak at 101.6 eV. Fittings were carried out using Gaussian–Lorentzian doublets. Within the experimental conditions, the sensitivity to the surface is rather limited and it can be assumed that the XPS spectra give information of the electronic structure that corresponds to the bulk. X-ray absorption experiments were carried out at the BM26 and BM8 beamlines of the European Synchrotron Radiation Facility 共ESRF兲in Grenoble. Samples were measured in fluorescence and total electron yield 共TEY兲modes, respectively. No appreciable differences were found between the two detection modes and spectra are included here without distinction. The storage ring was operated with an average current of 200 mA and an electron beam energy of 6 GeV. The fixed-exit double-crystal monochromator was equipped with a pair of Si-共111兲crystals being the energy resolution ⌬E/E⬃7⫻10−5 at the Mn K-edge 共6539 eV兲. The harmonics rejection was achieved by detuning the second crystal from the parallel alignment to 70% intensity. In all cases the onset energy of the absorption process was chosen to be the maximum of the first derivative in the edge region of the absorption spectrum. The experimental x-ray absorption near edge structure 共XANES兲spectra were normalized after background subtraction. 043912-2 Céspedes et al. J. Appl. Phys. 106, 043912 共2009兲 Extended x-ray absorption fine structure 共EXAFS兲 analysis was performed by using the VIPER software.19 Oscillations were obtained after removing the background by a cubic spline polynomial fitting and EXAFS signal ␹ 共k兲was obtained by normalizing the magnitude of the oscillations to the edge jump. The corresponding pseudoradial distribution function around Mn atoms was obtained by weighting the EXAFS signals by the square wave number ␹ 共k兲·k2multi- plying by a Hanning window and Fourier transform 共FT兲. For comparison, the FT of EXAFS signal was performed within the same k-range. Theoretical EXAFS signals were generated by using the amplitude and phase backscattering functions obtained using the FEFF6 code.20 The calculation of the EXAFS spectra was performed in a single scattering approach. The scattering potentials were obtained within the standard muffin-tin 共MT兲approximation and by using the nonoverlapping MT sphere approximation. The potential also includes the Hedin–Lundqvist 共HL兲self-energy with the Quinn correction to account for low energy electron-hole losses. Standard fits of EXAFS parameters were performed by minimizing the difference between filtered data and simulated signal where no constrains were imposed. The computation of the Mn K-edge XANES spectra was carried out using the multiple-scattering code CONTINUUM 共Ref. 21兲based on the one-electron full-multiple-scattering theory.22,23 For a complete discussion of the XANES simulation procedure we refer the reader to Refs. 24 and 25. The potential for the different atomic clusters was approximated by a set of spherically averaged MT potentials built by following the standard Mattheis’ prescription.26 The MT radii were determined following the Norman’s criterion.27 The Coulomb part of each atomic potential was generated using charge densities obtained from a nonlocal selfconsistent Dirac–Fock code.28 The atomic orbitals were chosen to be neutral for the ground state potential, whereas different choices were used to build the final state potential and we have found during the present calculations that the screened and relaxed Z+1 option29 lead to the best results for the simulation of the experimental absorption spectra at the Mn K-edge. The calculated theoretical spectra have been further convoluted with a Lorentzian shape function to account for the core-hole lifetime30 共⌫=1.2 eV兲and the experimental resolution 共⌫=1 eV兲. III. RESULTS AND DISCUSSION A. XRR and RBS XRR spectra of the studied multilayers are shown in Fig. 1for both samples series prepared with different Mn-layer thickness 共tm兲关Fig. 1共a兲兴and with variable Si3N4layer thickness 共tsn兲关Fig. 1共b兲兴. Well-defined peaks of heterostructure periodicity in the out-of-plane direction can be observed in the spectra from samples with larger tmand tsn thicknesses, while these periodicity features become smoother as tmand tsn are reduced. In fact, noncontinuous Mn layers are expected for very small tmvalues, where Mn layer thickness and interface roughness become comparable. It should be noted in the figure that samples with a large bilayer number, as for instance 75 bilayers, have a total thickness too large to exhibit the short oscillations corresponding to finite size film interferences. For several samples, multilayer periodicity has been additionally confirmed by transmission electron microscopy studies.17 Sample description and microstructural features such as average bilayer thickness are summarized in Tables Iand II for samples with variable tmand tsn, respectively. Measured layer thickness is approximately 15% higher than nominal. From now on, samples are named by their experimental thickness obtained by XRR. In order to obtain the manganese atomic density in Mnrich layers, the total amount of Mn has been determined by RBS. Figures 2共a兲and 2共b兲display spectra obtained using incident ions of 2 MeV and their corresponding simulations using the SIMNRA software package. The calculated Mn atomic density and film atomic composition obtained jointly by RBS and NRA for some representative samples are summarized in Tables Iand II. Finally, a comparison between the experimental manganese layer thickness values obtained from XRR spectra and the ones calculated from RBS data considering pure metal manganese density is shown in Fig. 2共c兲. The obtained difference means that, even if pure Mn was sputtered, layers do not consist on metallic Mn, since to obtain the actual thickness 共given by the XRR data兲a smaller Mn density should be considered in calculations from RBS. On the other hand, no appreciable oxygen amount was de- 123456 Intens i ty ( arb. un i ts ) tm=3.8 nm (x10) tm=1.9 nm (x20) tm=0.86 nm (x43) tm=0.5 nm (x75) tm=0.28 nm (x75) tm=7.6 nm (x5) tm=5.4 nm (x8) [Mn(tm)/Si3N4(3.4nm)]n(a) 2θ ( º ) tsn=1.6 nm tsn=1.0 nm tsn=6.7 nm tsn=3.4 nm [Mn(0.7nm)/Si3N4(tsn)]43 (b) FIG. 1. 共a兲XRR spectra of some of the multilayered samples with different Mn-layer thickness 共tm兲. Experimental tmvalues are shown in the figure. 共b兲 XRR spectra of the prepared samples with different Si3N4-layer thickness 共tsn兲. Experimental tmvalues are shown in the figure. All spectra have been vertically shifted for clearness. 043912-3 Céspedes et al. J. Appl. Phys. 106, 043912 共2009兲 tected by RBS and NRA experiments 共less than 5 at. %兲. Both, the reduced Mn density in RBS simulations and the absence of oxygen indicate that manganese nitridation occurs at the Mn-rich layers. Previous works have related the oxygen content to the deposition method. These works are based on the tendency of Si3N4to develop the corresponding oxynitride which can lead to strong modification of the physical properties. However, in contrast to deposition from Si3N4ceramic targets, reactive sputtering from Si, as it has been done here, leads to lower oxygen content.18,31 B. Magnetic characterization The magnetic characterization results of the Mn/Si3N4 samples are displayed in Fig. 3. Magnetization per Mn atom has been calculated by using the amount of Mn determined from the RBS measurements. Figure 3共a兲shows the hysteresis loops recorded at RT and 5 K for selected samples with tmranging from 3.5 to 0.14 nm. In all cases, the diamagnetic contribution of the silicon substrate has been subtracted from the recorded data. For the tm=0.14 nm sample, a high paramagneticlike component is observed at first glance. While this sample exhibits a very weak FM ordering, the RT magnetization of the tm=3.5 nm sample indicates the existence of an average magnetic moment of ⬃0.4 ␮ Bper Mn atom. This value corresponds to about 110 emu/cm3at RT, being the highest saturation magnetization among the samples studied here. This sample shows a coercive field HC ⬃60 Oe at RT. Similar HCvalues, ranging between 50 and 80 Oe, are found for the whole set of samples. The origin of ferromagnetism in the Mn/Si3N4system has been attributed to the presence of a distorted Mn3N2FM phase with a TChigher than 400 K, which seems to be favored by the Si3N4tetrahedral structure.17 In this work, ferromagnetism has been obtained for most of the prepared samples. The temperature dependence of magnetization M共T兲measured under an applied magnetic field of 100 Oe, can be seen in Fig. 3共b兲. An ordering temperature well above 400 K can be inferred for samples with tmⱖ0.5 nm. For samples with smaller tm, although an enhancement of TC could be evoked from the M-Tcurves, their smaller FM response together with the complexity of Mversus Tmeasurements over 400 K for thin film samples, make difficult to assert an increase of TC. The evolution of the saturation magnetization 共MS兲at RT versus the measured Mn layer thickness 共tm兲for all the obtained FM samples is represented in Fig. 3共c兲. For small Mn layer thickness, MSincreases with tm, exhibiting a maximum for a Mn layer thickness of the order of the Si3N4one 共⬃3.5 nm兲, while MSdecreases for further tmincrease. The importance of the Si3N4layer thickness into the magnetic properties is also evidenced in the inset of Fig. 3共a兲, that shows the M共T兲curves of several samples prepared with constant Mn layer thickness 共tm =0.7 nm兲, same number of repetitions 共n=43兲and different Si3N4layer thickness, tsn 共ranging from 6.7 to about 1 nm兲. Although it should be noted that smaller magnetization values are found for the samples prepared varying tsn in comparison with the tmvariable series, it can be noticed that the highest FM signal within this variable tsn series is obtained for the sample labeled as 关Mn共0.7 nm兲/Si3N4共1nm兲兴43. The fact that the 关Mn共3.5 nm兲/Si3N4共3.4 nm兲兴10 sample exhibits TABLE I. Description of selected samples, nominally named 关Mn共tm兲/Si3N4共3nm兲兴n: nominal Mn thickness 共tm兲, number of bilayers 共n兲, nominal bilayer thickness 共⌳兲, XRR-obtained bilayer thickness 共XRR-⌳兲, XRR- obtained Mn thickness 共XRR-tm兲and atomic composition, obtained from simulation of RBS and NRA. Nominal tm 共nm兲n Nominal ⌳ 共nm兲 XRR-⌳ 共nm兲 Mn-layer atomic density 共at./cm2兲 XRR-tm 共nm兲 Composition 共at. %兲 关Mn兴关Si兴关N兴 6 5 9 11 4.6⫻1016 7.6 0.42 0.14 0.44 3 10 6 6.9 2.2⫻1016 3.8 0.31 0.16 0.53 1.5 20 4.5 5.3 1.1⫻1015 1.9 0.25 0.18 0.57 0.7 43 3.7 4.2 5.9⫻1016 0.86 0.17 0.25 0.58 0.4 75 3.4 3.6 4.1⫻1015 0.5 0.13 0.28 0.59 0.2 75 3.2 3.4 2.9⫻1015 0.28 0.10 0.30 0.60 0.1 75 3.1 3.1 1.9⫻1015 0.17 0.07 0.35 0.59 TABLE II. Description of selected samples, nominally named 关Mn共0.6 nm兲/Si3N4共tsn兲兴43: nominal Si3N4thick- ness 共tsn兲, number of bilayers 共n兲, nominal bilayer thickness 共⌳兲, XRR-obtained bilayer thickness 共XRR-⌳兲, XRR-obtained Si3N4thickness 共XRR-tsn兲and atomic composition, obtained from simulation of RBS and NRA. Nominal tsn 共nm兲n Nominal ⌳ 共nm兲 XRR-⌳ 共nm兲 Mn-layer atomic density 共at./cm2兲 XRR-tsn 共nm兲 Composition 共at. %兲 关Mn兴关Si兴关N兴 6 43 6.6 7.4 4.7⫻1015 6.7 0.08 0.34 0.58 3 43 3.6 4.1 4.8⫻1015 3.4 0.13 0.27 0.60 1.5 43 2.1 2.3 4.8⫻1015 1.6 0.20 0.20 0.60 0.7 43 1.7 1.7 4.7⫻1015 1.0 0.25 0.14 0.61 043912-4 Céspedes et al. J. Appl. Phys. 106, 043912 共2009兲 the highest MSvalue points out that the FM alignment is favored when both, the manganese-rich and the silicon nitride layers have comparable thicknesses. C. XPS and XAS The above results suggest a relationship between the way in which the distorted Mn3N2FM phase develops at the interface and the magnetic properties. For this reason we have undertaken a detailed study on the formation of such phase in this multilayer system. In order to study the electronic properties of Mn-rich layers, we have performed an XPS study of two samples, the tm=7.6 nm sample, representative of the thick Mn layer samples, and the tm=0.5 nm one, where noncontinuous Mn is expected. Depth profiles were performed by removing thin layers of material with a calibrated sputtering gun operated with Ar+ ions of 1.4 eV kinetic energy. The photoemission spectra were collected after each sputtering sequence in order to follow the evolution of the electronic structure as a function of depth into the samples. The depth-profile Si, N, and Mn compositions of the tm=7.6 nm sample are shown in Fig. 4. It can be observed that the depth-profile concentrations of the different elements are in rather good agreement with the layer period for the tm=7.6 nm sample. As stated above, XPS spectra of the different elements present in the FIG. 2. 共Color online兲共a兲RBS spectra taken at energy of incident ions of 2 MeV 共dots兲and their simulations 共lines兲of several Mn/Si3N4samples with different Mn-layer thickness 共tm兲.共b兲RBS spectra of several samples with different Si3N4-layer thickness 共tsn兲. All spectra have been vertically shifted for clearness. 共c兲Comparison between experimental Mn layer thickness values obtained from XRR and calculation from RBS. FIG. 3. 共Color online兲共a兲Hysteresis loops of several samples measured at 5 K and RT. The inset shows the effect of modifying the Si3N4layer thickness on the magnetization curves 共H=1 kOe兲for a Mn layer thickness constant 共tm=0.7 nm兲.共b兲Magnetization vs temperature 共H=100 Oe兲for several samples with different Mn layer thickness 共tm兲.共c兲Variation of the saturation magnetization at RT with the Mn layer thickness 共tm兲for all the FM samples obtained. A dotted-line is included just as a guide to the eyes. 043912-5 Céspedes et al. J. Appl. Phys. 106, 043912 共2009兲 samples have been measured after each sequence of sputtering. In Fig. 5we display the Mn 3pspectra of these two samples that were measured at the experimental point located at the depth of 5.4 nm 共cf. Figure 4兲. For the case of tm=0.5 nm sample, since Mn layers are very thin and embedded into Si3N4layer, it is not possible to distinguish different Mn layers from the XPS signal and therefore, this depth is not very significant. Anyway, it was chosen to be the same in both cases. The Fig. 5also displays the corresponding peak decomposition into the different Mn valence states. Through the fitting analysis it clearly appeared that the samples exhibit different Mn valence states. For the tm =7.6 nm sample, majority of metal and Mn2+ contributions has been obtained 共about 40% Mn metal, 31% Mn2+, 22% Mn3+, and 7% Mn4+ are given from the fits, where an uncertainty between 5%–10% can be expected兲. Despite the small additional contribution from higher oxidized species, if we consider the majority species, the metallic Mn/Mn2+ ratio is about 4/3. For the tm=0.5 nm sample, mostly Mn2+ contribution was deduced from the XPS analysis 共about 6% Mn metal, 65% Mn2+, 20% Mn3+, and 9% Mn4+兲. The extra Mn3+ and Mn4+ high valence contributions are approximately the same in both samples. However, the Mn metallic phase is drastically reduced in this former sample, being the Mn2+ valence state the dominant one. The obtained Mn average oxidation states are about +1.9 for the tm=7.6 nm sample and +2.3 for the tm=0.5 nm one. Aimed at determining the Mn electronic state, the local structure of the Mn sites and the presence of Mn–N alloys, we have performed a combined XANES and EXAFS study of the 关Mn共tm兲/ Si3N4共3.4 nm兲兴nand 关Mn共0.7 nm兲/Si3N4共tsn兲兴43 samples. The normalized Mn K-edge XANES spectra of representative samples are shown in Fig. 6. The threshold energy position evidences an average Mn oxidation state close to +2, in agreement with XPS results. Details of the prepeak region can be seen in the inset, showing an evolution of the prepeak intensity from the manganese foil reference to the samples with thinner Mn layer. As shown in Fig. 6, the shape and the main spectral features of the spectra change as tmis varied. Therefore, we have studied the possible formation of a Mn–N alloy in the samples by comparing the experimental Mn K-edge XANES spectra and the results of theoretical simulations. It should be noted that despite the current 50 40 30 20 10 0 at (%) 76543210 De p th ( nm ) Mn N Si t m = 7.6 nm FIG. 4. XPS depth profile of the tm=7.6 nm sample. Intensity (arb.u.) t m = 7.6 nm (a) Mn 3p Mn 0 Mn 2+ Mn 3+ Mn 4+ Intensity (arb.u.) 54 52 50 48 46 4 4 Bindin g ener gy ( eV ) (b) t m = 0.5 nm Mn 3p Mn 0 Mn 2+ Mn 3+ Mn 4+ FIG. 5. Mn 3pXPS spectra of the tm=7.6 nm and tm=0.5 nm with the corresponding decomposition analysis into the different Mn valence states. FIG. 6. 共Color online兲Mn K-edge XANES spectra of Mn/Si3N4samples with different Mn layer thickness 共tm兲values and a metallic Mn foil. Spectra have been vertically shifted for clearness. The inset shows a zoom of the prepeak region. 043912-6 Céspedes et al. J. Appl. Phys. 106, 043912 共2009兲 progress in XANES computation, several problems, as the treatment of the inelastic losses of the photoelectron, still remain as open problems.32 Therefore we have started the computation on a reference material, metallic ␣ -Mn, in order to fix the main parameters of the calculations. The first step of the calculations was deserved to determine the size of the cluster needed to reproduce all the spectral features present in the experimental XANES spectra. To this end we have performed the computation of the Mn K-edge XANES in the case of ␣ -Mn 共I-43m兲by increasing progressively the number of atoms in the cluster. It should be noted that because Mn occupies four different crystallographic positions 2a 共Mn1兲,8c共Mn2兲,24f共Mn3兲, and 24g共Mn4兲, XANES computations have to be performed for four different clusters in which the absorbing Mn atom lies at the different Mnisites, being the theoretical spectrum made up from the weighted sum of these contributions according the crystallographic statistical ratio. In this way, we have determined that the computations performed for clusters including contributions from neighboring atoms located within the first 7 Å around photoabsorbing Mn account for all the observed spectral features, and that further coordination shells do not contribute significantly to the XANES spectrum. We have also studied the effect of both the maximum angular momentum quantum number lmax and the overlapping factor imposed to the MT spheres to build up the potential27 into reproducing the Mn K-edge XANES spectrum. We have found that the best reproduction of the experimental spectrum is obtained by using lmax=4 and a 1% overlapping factor to build upa7Åcluster. The results of these computations are shown in Fig. 7共a兲. The agreement between the computations and the experimental signal is noticeable as the calculations reproduce the steplike profile at the threshold 共A兲, the main peak 共B兲and the two structures located at 28 共C兲and 47 eV 共D兲above the edge. During the calculations we have tested the performance of the energy dependent HL and Dirac–Hara 共DH兲exchange and correlation 共ECP兲potentials. Moreover, we have also made computations by using only the real part of the HL ECP 共hereafter, real HL兲and, in addition, we have built a “complex” DH one by adding to it the imaginary part of the HL one. As shown in Fig. 7共a兲, the performances of the different EPC potentials are similar although in the case of the HL the calculated absorption maxima falling short of the observed ones. By contrast, computations performed by using the DH ECP show a better reproduction of the relative energy position among the peaks. This is evidenced by the reproduction of the main resonance D and the negative dip E located, respectively, at 47 and 62 eV above the edge. Moreover, the addition of the imaginary part of the HL potential improves the calculations yielding the use of the complex DH ECP the best reproduction of the intensity ratio among the different spectral features. It should be noted that the assessment of the quality of the theoretical computations is based on the correct reproduction of the shape and energy position of the different spectral features and of their relative energy separation and intensity ratio as well. Possessing these results we have further performed the computation of the Mn K-edge spectrum in the case of both ␪ -MnN and ␩ -Mn3N2. In the case of MnN, computations have been performed for a cluster containing 179 atoms that includes all the contributions from neighboring atoms located within the first 7.4 Å around the absorbing Mn atom. In the case of Mn3N2, Mn occupies two different crystallographic sites 2a共Mn1兲and 4e共Mn2兲. Hence, XANES computations have been performed for two clusters in which the absorbing Mn atom lies at the different Mn1 and Mn2 sites. These clusters include the scattering contributions within the first 7.5 Å around the absorber and contain 153 and 150 atoms for Mn1 and Mn2, respectively. The Mn K-edge absorption of Mn3N2has been calculated by adding the theoretical computations obtained for both clusters according to the crystallographic statistical ratio. FIG. 7. 共Color online兲共a兲Comparison between the Mn K-edge of a Mn metallic foil 共쎲兲and the theoretical spectra calculated ona7Åcluster by using lmax=4 and different ECP potentials: real HL 共blue, dots兲, real DH 共green, dashed line兲, complex HL 共black, solid line兲and complex DH 共purple, dot-dashed兲.共b兲Comparison between the experimental Mn K-edge XANES of the tm=0.28 nm Mn sample 共쎲兲and the theoretical computation performed for both MnN 共blue, dots兲and Mn3N2共red, solid line兲by using the complex DH ECP potential. 共c兲Comparison between the experimental Mn K-edge XANES of the tm=0.28 nm 共black, 쎲兲,tm=3.8 nm 共green, 䊐兲 and tm=7.6 nm 共red, 䊊兲samples and the theoretical signals calculated as the weighted sum of the computed spectra for both Mn3N2and ␣ -Mnasa function of the Mn3N2fraction n:100%共cyan, solid line兲, 75% 共black, dots兲, 50% 共red, dashed兲, 25% 共blue, dot-dashed兲and 0% 共green, 䊊兲. 043912-7 Céspedes et al. J. Appl. Phys. 106, 043912 共2009兲 The results of these computations are compared to the experimental spectrum of the tm=0.28 nm sample in Fig. 7共b兲. The computed spectra show similar oscillations in the high energy region. However, their spectral profiles at the threshold region differ, especially concerning the pre-edge peak 共A兲. This peak is experimentally observed and typically assigned to a noncentrosymmetric environment of the absorbing site, such as in tetrahedral coordination 共Td兲symmetry. The theoretical simulation reproduces this peak for ␩ -Mn3N2, while it is absent for MnN. These results suggesting the presence of ␩ -Mn3N2in our samples have been verified by studying the modification of the XANES spectra as a function of the thickness. Indeed, as the Mn thickness increases the ␩ -Mn3N2contribution should be hindered as the XANES would be made by the superposition of the XANES spectra of both ␩ -Mn3N2and bulk-Mn-like. This is shown in Fig. 7共c兲, where the XANES spectra of the samples with different Mn thickness 共tm兲are compared to the theoretical spectra calculated as the weighted addition of the computed spectra for ␩ -Mn3N2and ␣ -Mn in the form: n⫻Mn3N2+m ⫻ ␣ -Mn. As shown in the figure, the modification of the Mn thickness strongly changes the spectral profile in the high energy region. For example, in the case of the sample with the small Mn thickness 共tm=0.28 nm兲the spectrum shows a negative dip 共D兲at ⬃33 eV above the edge. As the Mn thickness increases, samples tm=3.8 nm and tm=7.6 nm, this spectral feature evolves up to yield a positive resonance. This is exactly the situation obtained in the theoretical simulation of Fig. 7共c兲: whereas the Mn3N2dominates the spectrum, feature D remains negative but, however, as the contribution of bulk Mn increases, the sign of this contribution is reversed. Similar analysis can be followed for the energy position of the main resonance E at ⬃57 eV above the edge. Experimentally, this feature is shifted toward higher energies as the Mn thickness increases in the samples, in identical manner to the behavior of the theoretical signal constructed by varying the weight of the ␩ -Mn3N2and ␣ -Mn contribution to the XANES spectra. The obtained evolution agrees with the Mn3N2:Mn ratio obtained by XPS. Therefore, the ab initio calculations have evidenced the presence of both metallic and Mn3N2phase 共with Mn Td symmetry兲in sample tm=7.6 nm, while no sign of Mn metallic phase appears in the tm=0.28 nm spectrum, which presents just nitride contribution. The fact that increasing the Mn-layer thickness leads to a reduction of the relative contribution of ␩ -Mn3N2compared to the bulklike Mn indicates that the Mn nitride formation in the sputtered Mn layer is related to the Si3N4interfaces, leading, for thin enough Mnrich layers, to all the deposited Mn turned into manganese nitride. Further confirmation of these results has been obtained by studying the Mn local environment from the analysis of the EXAFS part of the spectra. Figure 8shows the evolution of the FT magnitude of the EXAFS signal at the Mn K-edge along the series with tmvariable and a Mn metal foil, given for comparison. At first glance, clear differences between the FT magnitude of the tm=7.6 nm sample and the others, as well as the extinction of the FT second peak for small tm values, are observed. The comparison between experimental data and simulations is shown in Fig. 9, and the obtained results are summarized in Table III. Analysis shows a complex situation that can be explained as follows. A contribution around 2.05 Å corresponding to Mn–N distance is present for all the sample series and besides, samples with tmhigher than 0.5 nm present a second Mn–Mn near 3 Å. Those distances agree with Mn3N2compounds with slightly compressed distances.9 Additionally, a small Mn–Si contribution at about 3.1 Å has been found for tm=0.86 nm and tm=3.8 nm samples, what can be assigned to the link between the Mn3N2and Si3N4 phases. Finally, the Mn–Mn distance found at 2.67 Å in sample tm=7.6 nm is in agreement with distances in metallic FIG. 8. Experimental EXAFS FT of some representative 关Mn共tm兲/Si3N4共3.4 nm兲兴nsamples and a Mn metal foil. FIG. 9. 共a兲k-space comparison between the filtered EXAFS signals 共dots兲 and simulations 共lines兲at the Mn K-edge of samples with different Mn layer thickness 共tm兲.共b兲FT of the 共a兲plotted signals; modulus and imaginary parts have been presented for dual comparison with calculated contributions 共lines兲. 043912-8 Céspedes et al. J. Appl. Phys. 106, 043912 共2009兲