Trace element content and magnetic properties of commercial HOPG samples studied by ion beam microscopy and SQUID magnetometry
TTrace element content and magnetic properties of commercial HOPG samplesstudied by ion beam microscopy and SQUID magnetometry
D. Spemann a , P. Esquinazi a, ∗ , A. Setzer a , W. B¨ohlmann a a Institute for Experimental Physics II, Universit¨at Leipzig, Linn´estr. 5, 04103 Leipzig, Germany
Abstract
In this study, the impurity concentration and magnetic response of nine highly oriented pyrolytic graphite (HOPG)samples with di ff erent grades and from di ff erent providers were determined using ion beam microscopy and SQUIDmagnetometry. Apart from sideface contaminations in the as-received state, bulk contamination of the samples inmost cases consists of disk-shaped micron-sized particles made of Ti and V with an additional Fe contaminationaround the grain perimeter. The saturation magnetization typically increases with Fe concentration, however, there isno simple correlation between Fe content and magnetic moment. The saturation magnetization of one, respectivelysix, out of nine samples clearly exceeds the maximum contribution from pure Fe or Fe C. For most samples thetemperature dependence of the remanence decreases linearly with T – a dependence found previously for defect-induced magnetism (DIM) in HOPG. We conclude that apart from magnetic impurities, additional contribution tothe ferromagnetic magnetization exists in pristine HOPG in agreement with previous studies. A comparative studybetween the results of ion beam microscopy and the commonly used EDX analysis shows clearly that EDX is not areliable method for quantitative trace elemental analysis in graphite, clarifying weaknesses and discrepancies in theelement concentrations given in the recent literature. Keywords:
HOPG, trace element content, magnetic properties
1. Introduction
In the last years the possibility to have magnetic or-der in di ff erent kinds of solids above room temperaturewithout nominally magnetic ions, like the usual transi-tion or rare earth elements, has attracted the attentionof the solid state community. Although some theoreti-cal and experimental works in the past provided somehints for the existence of this apparently unusual phe-nomenon, it has been only recently that we becameaware that di ff erent kinds of defects, like vacancies, hy-drogen or a combination of those with nominally non-magnetic elements can trigger magnetic order in solids[1–5].This phenomenon, named defect-induced magnetism(DIM), has been mostly studied, theoretically and ex-perimentally, in the single element graphite / graphene.More than ten years ago, systematic studies of the mag-netic properties of di ff erent graphite samples with dif- ∗ Corresponding author. Tel / Fax: +
49 341 9732751 /
69. E-mailaddress: [email protected] (P. Esquinazi)
Email address: [email protected] (D. Spemann) ferent magnetic impurity concentrations suggested thatan extra magnetic contribution, other than from impu-rities, should exist [6, 7]. In general, the small ferro-magnetic moment observed in commercial, as-receivedhighly oriented pyrolytic graphite (HOPG) samplesmakes the detailed knowledge of the contribution frommagnetic impurities imperative to understand its origin.In this work, we have studied HOPG samples with dif-ferent grades from three di ff erent commercial sources,i.e. a total of nine HOPG samples. This study, there-fore, gives a fairly reproducible spectrum on the di ff er-ent magnetic contributions in the HOPG samples avail-able nowadays.An accurate measurement of impurity concentrationsin the ppm range and in micrometer small grains is notsimple and only possible with an experimental methodwhich provides elemental imaging with excellent detec-tion limits in the ppm and sub-ppm range together withreliable quantification, preferably in a non-destructiveway. The method used in this work, Particle InducedX-ray Emission (PIXE), has the necessary requirementsfor this kind of studies. In this work, we show that a Preprint submitted to Carbon October 29, 2018 a r X i v : . [ c ond - m a t . m t r l - s c i ] J un arge contamination with magnetic elements is found atthe sidefaces of as-received HOPG samples, very prob-ably originating from the cutting of the sample priorshipping. The observed amount of the sideface impu-rities can overwhelm by far the usually observed impu-rity concentration in the bulk and, if no thorough samplecleaning is done, clearly prevents the measurement ofthe magnetic contribution from DIM in pristine HOPGsamples.On the other hand, the sole measurement of the mag-netic moment of a sample with a known amount of im-purities does not provide always with a clear statement,whether the ferromagnetism is or is not due to impuri-ties [8–11]. In general and for small impurity concentra-tion and grains [8], only an upper estimate of the ferro-magnetic signal from those grains can be simply done.This is due to the fact that not only the grain size, butalso the, in general di ffi cult to quantify, stoichiometryand structure of the existing ferromagnetic phases de-termine the magnetic signal. Therefore, demonstratedfor the case of graphite, we will show in this work thattogether with a careful impurity measurement, the tem-perature dependence of the remanent magnetic momenthelps to discern, which magnetic contributions are ac-tive in the sample.The manuscript is divided in three more sections anda conclusion. In the next section 2 we provide details onsample preparation, the used elemental analysis meth-ods and the commercial Superconducting Quantum In-terferometer Device (SQUID) used for magnetometry.Taking into account the, in general, limited knowl-edge on the possibilities of PIXE in contrast to usualmethods for elemental analysis like Scanning ElectronMicroscopy combined with Energy Dispersive X-rayanalysis (SEM / EDX) [9–11], the manuscript includesa complete trace element analysis obtained with PIXEin section 3 and a comparison with the results obtainedwith EDX analysis in section 3.4. Section 3.1 discussesthe impurity distribution at the sidefaces of the commer-cial HOPG, a fact that we believe is of importance forany future discussion on the “intrinsic” bulk impurityconcentration presented in section 3.2. An elementalanalysis of single metallic grains found in HOPG sam-ples is given in section 3.3. In section 4 we presentand discuss the magnetic characterization done with theSQUID in all HOPG samples and compare it with theinformation obtained from the elemental analysis. Theconclusion is given at the end of the manuscript.
2. Experimental
The samples studied are commercially availableHOPG from Advanced Ceramics (now Momentive Per-formance Materials), NT-MDT and SPI Supplies [12].From each of these companies, samples were purchasedin the three available structural grades designated asZYA, ZYB and ZYH. In case of SPI Supplies the corre-sponding designation is SPI-1, -2 and -3. In the follow-ing, the samples are named by the company (AdvancedCeramics is abbreviated as AC) and the structural grade,e.g. AC ZYA denotes the ZYA-grade sample from Ad-vanced Ceramics.The HOPG samples were wire-cut into pieces of(5 ×
5) mm size each and thoroughly cleaned in anultrasonic bath with ethanol for SQUID magnetome-try. On one piece of ZYA-grade HOPG from each com-pany trace elemental analysis was performed in the as-received state, i.e. without sample cleaning to checkfor possible contaminations, especially at the sidefaces.For this purpose, the samples were only cleaved witha CuBe-knive and glued on Si substrates using varnishwith the cleaved surface on top. In order to characterizethe “intrinsic” bulk trace element content using PIXEand SEM / EDX analysis, the samples were additionallycleaned three times with ethanol in an ultrasonic bath,glued on Si substrates using varnish and the top surfaceremoved by tape stripping to prepare a fresh one ontowhich the ion / electron beam was directed. Trace elemental analysis of the samples was per-formed at the LIPSION facility of the University ofLeipzig [13] with PIXE [14] and Rutherford Backscat-tering Spectrometry (RBS) [15] using a 2.28 MeV pro-ton microbeam focused to 1 − µ m diameter. The pro-ton microbeam was raster-scanned across the samplesurface and the characteristic X-rays and backscatteredprotons simultaneously recorded for each scan pixel. Inthis way, non-destructive quantitative imaging of the el-emental content is possible with micron lateral resolu-tion. Detailed information about the LIPSION facilitycan be found in [16].In contrast to the commonly used EDX analysis withits comparably poor minimum detection limits (MDL),PIXE allows true trace elemental analysis in carbonwith MDLs (cid:46) . µ g / g for 3d-elements like Fe. A com-parison with neutron activation analysis showed thatFe concentrations as low as 0.17 µ g / g can be accu-rately determined using the PIXE method [8]. In addi-tion, MeV protons penetrate much deeper into the mate-2ial than electrons of several tens of keV. According toSRIM-2013 simulations [17] 2.28 MeV protons pene-trate 47 µ m deep into graphite where they generate 90%of the total X-ray yield from Fe atoms within the first27 µ m [18]. At this depth the beam diameter has in-creased by only ≈ . µ m due to scattering processes,i.e. the proton microbeam is still well-focused allow-ing PIXE to be used as a bulk-sensitive analysis tech-nique with good imaging capabilities even for trace el-ements buried several microns below the graphite sur-face. Since the X-ray emission process used in PIXErelies on the ionization of inner-atomic shells (in caseof Fe the innermost K-shell is used) which are practi-cally una ff ected by the chemistry of the sample, PIXEcan be considered as being “chemically blind” as areEDX or XRF. As a consequence, the actual distributionof impurity atoms does not a ff ect the X-ray productionfrom a certain amount of these impurities, no matterwhether they are homogeneously distributed or enrichedin small grains. Since ion channeling can be excludedin our measurements, the lattice site location of impu-rity atoms does not influence their detection e ffi ciencyby the PIXE method as well. This ensures that all impu-rity atoms can be detected, no matter in which chemicalstate they are or how they are distributed within the vol-ume probed by the ion beam – an important prerequisitefor reliable quantitative elemental analysis.The PIXE spectra were recorded using a high-purityGUL0110 Germanium detector from Canberra withan active area of 95 mm subtending a solid angleof 150 msr and an energy resolution of 144 eV at5.9 keV. The spectra were analyzed using GeoPIXE II[18]. Whereas for the calculation of bulk concentra-tions, graphite with a thickness greater than the protonrange was used as matrix in the data analysis, a thinlayer of Fe with a mass thickness of 0.1 mg / cm wasassumed for the analysis of the sideface contamination.This accounts for the negligible energy loss of the pro-tons and x-ray absorption in the thin surface layer ofcontamination at the sideface of the samples. The thick-ness itself is arbitrarily chosen and cancels out in thecalculation of the mass / area value of the contamination.Figure 1 shows a typical PIXE spectrum from ZYA-grade HOPG from Advanced Ceramics together withthe extracted elemental concentrations.Whereas PIXE provides excellent sensitivity, but nodirect depth profiling capabilities, RBS inherently al-lows depth profiling of element concentrations if theyare su ffi ciently high. As will be shown later, RBS canbe used to determine the thickness of metallic particlesand their location below the surface in a non-destructiveway, i.e. without the need to prepare cross-sections from Ti V Cr Fe Ni
Q=4.98 µC Advanced Ceramics ZYA
Figure 1: PIXE spectrum from ZYA-grade HOPG from AdvancedCeramics recorded with a collected proton charge of Q = . µ C.The green curve are the measured data, the violet and red curve are thebackground simulation and fit to the data, respectively, from GeoPIXEII. The extracted concentrations as well as minimum detection limitsare given and the corresponding X-ray lines for the detected elementsindicated. the sample for EDX or TEM analysis. Furthermore,RBS was used in this study to accurately determine theapplied proton charge from the RBS yield of the carbonbulk.The RBS spectra were recorded using an annularPIPS detector from Canberra with an area of 275 mm ,an e ff ective backscattering angle of 172 ◦ , a solid an-gle of 86 msr and an energy resolution of 10.6 keV for2.28 MeV protons. Afterwards, the spectra were ana-lyzed using XRUMP [19]. For comparison with previously published studies[9, 11] and ion beam microscopy selected samples werefurther analyzed with SEM and EDX using the DualBeam Microscope Nova NanoLab 200 from FEI Com-pany. Prior elemental analysis of single metallic parti-cles, backscattered electron (BE) imaging was used tolocate them due to the Z -contrast between the carbonbulk and the heavier 3d-elements of the grains. ThenEDX spectra and element maps were recorded using20 keV electrons and analyzed with the EDAX soft-ware. Magnetization measurements were performed with aSQUID magnetometer MPMS-7 from Quantum Designwith Reciprocal Sample Option (RSO) and the mag-netic field applied parallel to the graphene planes ofthe samples (within an experimental resolution of ± ◦ ).3everal years of experience in measuring graphite sam-ples with SQUID [8, 20, 21] and the excellent repro-ducibility of the used apparatus allows us a sensitivityof (cid:46) × − emu.
3. Trace element content in HOPG
In order to check for contaminations in the as-received state the HOPG samples of ZYA grade wereanalyzed with the proton beam incident on both top sur-face and sideface with a tilt angle of 15 ◦ between beamand sample normal. The yield from the secondary elec-tron background (see Fig. 1 for comparison) was used todi ff erentiate between both areas designated as “surface”and “sideface” in Fig. 2 where the X-ray yields, i.e. ele-mental concentrations, are displayed in false color scalefor the SPI-1 sample. As can be seen in Fig. 2(a), thesideface is strongly contaminated with Cr, Fe, and Ni,all three showing an identical distribution indicating thatthey originate from the same source. The quantitativeanalysis of this contamination reveals a composition of16.0% Cr, 77.2% Fe and 6.8% Ni by weight which fitswith the frequently used non-magnetic, austenitic SAEgrade 301 stainless steel [22]. It is therefore reasonableto assume that this contamination originates from thecutting of the samples prior shipping using a stainlesssteel tool.Contaminations of similar distribution and composi-tion can be found on the sidefaces of AC ZYA and NT-MDT ZYA samples as well, however, to a less severedegree. As Tab. 1 shows, 702 ng Fe per cm sidefacearea was found for SPI-1, whereas for the AC and NT-MDT samples Fe concentrations amount to 15.8 ng / cm and 155 ng / cm , respectively. In addition, Cl, K andCa contaminations were detected on the sidefaces ofAC ZYA and SPI-1 with distributions di ff erent fromCr, Fe and Ni. Even though they were not measuredit is reasonable to assume that the sidefaces of the ZYBand ZYH samples are contaminated as well in the as-received state.After thorough ultrasonic cleaning, however, thesidefaces are free of contaminants. In order to checkthis, the large scan from which the bulk trace elementcontent was determined was placed such that the side-face area was included as well, see Fig. 2(b). Clearly,the sideface does not show enhanced concentrations ofCr, Fe and Ni anymore compared to the top surface. Weconclude that ultrasonic cleaning is mandatory prior useof these HOPG samples in contamination-critical appli-cations. Due to the excellent sensitivity of PIXE, trace ele-ment analysis can be performed by simply scanning theproton beam over a large area of the sample and col-lecting the X-ray photons from the trace element atoms.For this purpose, a 2 .
24 mm × .
24 mm sized scan wasmade covering a substantial portion of the whole samplearea. The PIXE spectra (see as example Fig. 1 for theAC ZYA sample) were analyzed and the concentrationsextracted are given in Tab. 1.The findings can be summarized as follows: (i)the ZYA-grade samples have Fe concentrations below1 µ g / g; (ii) the ZYB and ZYH samples of NT-MDT andSPI all have similar Fe concentrations of the order of10 µ g / g, whereas the AC ZYB has the lowest Fe con-tamination of < . µ g / g ((0 . ± . µ g / g was deter-mined for a similar sample a few years ago [8]) and theAC ZYH the highest Fe concentration of 22.6 µ g / g; (iii)most of the samples also contain Ti and V with varyingconcentrations between the di ff erent samples.From the concentrations it is immediately clear thatthe contamination is not due to stainless steel particlesas was the case for the sidefaces. Nevertheless, it isinteresting to compare the Fe contamination of the bulkand sideface. In case of the 5 × SPI-1 sample thesideface contains twice as much Fe than the bulk, whichagain illustrates the importance of ultrasonic cleaning.We would like to point out that, with the exception ofAC ZYA and AC ZYB samples, we have not analyzedany of the other samples before and can therefore notmake any statement about the variation of the concen-tration values between di ff erent batches of these sam-ples. In [11] the Fe concentration in a SPI-2 sampledetermined by Instrumental Neutron Activation Analy-sis (INAA) was almost three times higher than in oursample. In that publication, however, the authors didnot provide any information on sample cleaning.AC ZYA samples have been used in our studies onDIM and therefore frequently analyzed in the last tenyears. We found that their trace element content showslittle variation between di ff erent batches. In case of ACZYB one sample was analyzed a few years ago [8],again showing similar trace element concentrations asgiven in Tab. 1.As already stated in [8] and confirmed recently in[9, 11] the trace elements are not distributed homoge-neously within the sample, but located in micrometerlarge grains that are homogeneously dispersed withinthe bulk. This is illustrated in Fig. 3 where the Ti, V andFe maps of a 400 µ m × µ m scan are shown for theAC ZYA and SPI-3 sample in (a) and (b), respectively.4 i e a c ed f s i de f a c e e c e s i d f a C on c en t r a t i on
200 µm × 200 µm2.24 mm × 2.24 mm e a c e s i d f s u f a r c e Back r s u f a c e Cr s u r f a c e Fe s u r f a c e Ni (a)(b) C on c en t r a t i on e cs i d f a e f s i de a c e u r f a c e s Back r f e s ua c Cr s u f a r c e Fe s u r f a c e Ni Figure 2: X-ray yield in false color scale for the background and the elements Cr, Fe and Ni: (a) SPI-1 sample in the as-received state (200 µ m × µ m scan size, Q = . µ C applied proton charge) showing a severe sideface contamination with Cr, Fe and Ni, probably originating from astainless steel tool used for sample cutting. Some loose particles of contamination can even be found on the top surface close to the sideface; (b)SPI-1 after ultrasonic cleaning (2 .
24 mm × .
24 mm scan size, Q = . µ C applied proton charge) with the same maximum concentration valuesof the color scale as in (a). As can be seen between the two dashed lines, the sideface is not contaminated anymore. (a)(b)
Concentration
Cr Fe NiTi V FeTi V Fe
Figure 3: Ti, V and Fe distribution in a 400 µ m × µ m sized scanarea for (a) AC ZYA and (b) SPI-3. Ti and V are strongly correlatedand show the same distribution. Most of the (Ti,V) grains contain Fe,but some do not. In addition, some Fe grains do not contain Ti or V. A detailed inspection of the maps reveals that Ti and Vare strongly correlated showing identical distributionswithin the scan area. In most of the (Ti,V) grains Fe ispresent too. However, there are a few (Ti,V) grains thatdo not contain Fe and a few Fe grains that do not containTi or V. The maps indicate that the grain density is sig-nificantly higher in SPI-3 and that the grains itself areslightly smaller compared to AC ZYA. Assuming thatgrains can be detected up to a depth of 27 µ m (see sec-tion 2.2) their density in AC ZYA can be estimated to about 6 × cm − .It should be noted that using BE and EDX imagingin [9] no grains were detected in SPI-3 at all, probablydue to their small size and the inferior sensitivity of themethods used. Consequently, the sample was assumedto be free of Fe contamination [9], in clear contrast toour findings, see Tab. 1. As can be seen in Fig. 3, finding single grains of con-tamination is easy with ion beam microscopy. A care-ful analysis of these grains should allow to draw someconclusions on their magnetic properties. Therefore,single grains were selected from larger scan areas andeach of the grains analyzed in detail using a smallerscan. Whereas PIXE allows to map elemental distri-butions with high sensitivity, RBS can be used to de-termine the depth below surface and thickness of thegrain. The simultaneous recording of PIXE and RBSdata, therefore, allows a reliable quantitative analysisin a non-destructive way, i.e. without the need of spe-cial sample preparation. As an example, Figs. 4(a) and(b) show the PIXE and RBS analysis, respectively, of asingle grain in the AC ZYA sample. As stated above,Ti and V have identical distributions, di ff erent from Feand Ni that are located at the outside of the grain, bothwith a similar and irregular distribution along the grain’sperimeter. This is also illustrated in the composite mapof Fig. 4(a) showing the distribution of Ti, Fe, Ni in red,green and blue color, respectively, with brighter colors5ample Sideface Concentrations in HOPG bulk ( µ g / g)Fe (ng / cm ) Ti V Cr Fe Co NiAC ZYA 15 . ± . . ± . . ± . < .
30 0 . ± . < .
19 0 . ± . . ± . . ± . < . < . ∗ < . < . . ± . . ± . < .
30 22 . ± . < .
25 4 . ± . ± < . < . < .
28 0 . ± . < . < . . ± . . ± . < .
29 10 . ± . < . < . . ± . . ± .
14 0 . ± .
04 10 . ± . < . < . ± < . < . < .
27 0 . ± . < . < . . ± . < . < .
31 9 . ± . < . < . . ± . . ± . < .
33 8 . ± . < . < . n.d.: not determined; ∗ Fe concentration in a similar AC ZYB sample: (0 . ± . µ g / g [8] Table 1: Sideface contamination with Fe and bulk concentrations of trace elements in HOPG. With the exception of AC ZYB the ZYA sampleswith the highest structural quality have the lowest Fe concentrations < µ g / g. In addition to the elements listed the concentrations of Mn, Cu andZn were determined for all samples as well and found to be < . µ g / g for Mn, < . µ g / g for Cu and < . µ g / g for Zn. zya_ac_c_reg1 newsim reset retlt 1read acac1r.dat -ascimev 2.28beam h+geom cornellphi 9theta 0conv 2.542 7 fwhm 16omega 86curr 0.7cha 0.048simresread kohlenstoff_h.adtla 1 th 49500 /cm2 c c 1 /la 2 th 1500 /cm2 c ti 1 v 6 /la 3 th 100000 /cm2 c c 1 /fuzz 1 5000 9 /retreg 400 900pl 1 ov 0d:cd spemann\messung\hopg050912\ Ti , V ,Fe, NiFe correspondsto thicknesscorrespondsto depth of grain C (b) TiFeNi
Ti VFe Ni (a)
Concentration
Figure 4: Ion beam analysis of a single grain in AC ZYA (17 . µ m × . µ m scan area): (a) Element maps and composite map of Ti, Fe and Nishowing a homogeneous distribution of Ti and V inside the grain and the location of Fe and Ni at the perimeter; (b) RBS spectrum extracted fromthe grain (the green curve in the Fe map represents the extracted scan area). From the XRUMP analysis (red curve) the metallic content, thicknessand depth of the grain can be determined. referring to higher concentrations of the respective ele-ment.Fig. 4(b) shows the RBS spectrum extracted fromthe grain alone. The broad peak around channel 720are protons backscattered from the metals where peakwidth and area reflect grain thickness and the total num-ber of metal atoms, respectively. The grain is alsovisible as “missing” carbon in the dip around chan-nel 530. From this dip position the depth of the graincan be determined. As quantitative analysis shows, this7 . µ m × . µ m sized grain has a mass thickness of(0 . ± .
01) mg / cm , consists on average of 13.7% Ti,82.0% V, 2.4% Fe and 1.9% Ni by weight and is lo- cated 4.35 µ m below the graphite surface. It contains(1 . ± .
1) pg Fe and (0 . ± .
10) pg Ni. Assumingfor simplicity that the grain is made of pure Vanadiumwith a mass density of ρ = . / cm , the geometricalthickness can be calculated to d ≈
210 nm. This and theanalysis of other grains show that they are not sphericalas assumed in [9], but flat disks that are oriented par-allel to the graphene planes. This finding including thelocation of Fe at the perimeter of the grains agrees wellwith the EDX and TEM analysis of an AC ZYA samplereported in [11]. Taking into account the high temper-atures T > ◦ C and pressures used in the produc-tion of HOPG from pyrocarbons [23], this flat shape is6o be expected. Metallic particles in the pyrocarbons,whatever their origin might be, melt at these tempera-tures and spread out perpendicular to the direction ofpressure, i.e. perpendicular to the c-axis of the formedHOPG. Furthermore, the formation of iron carbides isto be expected under these conditions. Indeed, electrondi ff raction analysis showed that Fe in the grain is notmetallic, but present as cementite Fe C [11] in agree-ment with the findings in [24–26].Figure 5 shows composite maps of grains for all in-vestigated samples with the exception of SPI-1 and NT-NDT ZYA where no grains could be detected within thedetection limits of PIXE imaging. Indeed, these twosamples are the only ones that do not contain Ti or V(see Tab. 1). They do, however, contain Fe of com-
TiFeNi
TiVFe
TiVFe
KTiFeTiVFe
TiVFe
TiVFe
TiVFe
TiVFe
ZYA ZYB ZYH N T - M D T SP I A d v an c ed C e r a m i cs Figure 5: Composite maps of single grains for all investigated samplesexcept SPI-1 and NT-MDT ZYA where no grains could be found. Thedisplayed elements and assigned colors are given in each map. Asdemonstrated, size and composition of the grains di ff er substantiallybetween di ff erent samples, but also within the same sample (see ACZYH). parable amount as AC ZYA, presumably more or lesshomogeneously distributed within the bulk and not con-centrated in grains as for the latter one. As can be seen from the maps, the grains di ff er substantially in sizefrom 2 µ m (NT-MDT ZYB) to ≈ µ m (AC ZYH) andcomposition between di ff erent samples, but also withinthe same sample as is demonstrated for AC ZYH. Here,three grains of very di ff erent shape and size are dis-played. In view of these variations, it is obvious, thatthe estimation of bulk concentrations from the analy-sis of a few single grains can lead to substantial errors.Indeed, taking the 1.2 pg Fe from the grain in Fig. 4and the grain density of 6 × cm − estimated fromFig. 3 one gets about 3 µ g / g Fe as bulk concentrationfor AC ZYA where the true value is (0 . ± . µ g / g.Obviously, the metal content of this grain is above theaverage compared to the other grains in AC ZYA. EDX is a wide-spread technique for elemental analy-sis and imaging and has been used recently in the char-acterization of contaminations in HOPG in [9, 11]. Inorder to compare its capabilities and limitations withthose of ion beam microscopy some of the HOPG sam-ples were studied using EDX. Figure 6 shows the X-rayspectra recorded from a NT-MDT ZYB sample usingEDX and PIXE, respectively, using a large scan area.
NT-MDT ZYB
EDXPIXE
Fe (10.4 µg/g)CO Ti V
Figure 6: EDX spectrum (black line) and PIXE spectrum (green line,together with GeoPIXE II fit, see Fig. 1 for explanation) recorded onNT-MDT ZYB. In both cases, a large scan was made to obtain reliablebulk concentration values, however, no peaks for the trace elementscan be detected in the EDX spectrum, demonstrating its insu ffi cientsensitivity for trace element analysis in HOPG. Whereas the PIXE spectrum shows peaks for Ti, V andFe, no peaks can be discerned for these trace elements inthe EDX spectrum, despite Ti and Fe having concentra-tions (cid:38) µ g / g. As a detailed analysis shows, a typicalMDL for Fe amounts to ∼ µ g / g in EDX analysis,about a factor 1000 larger than for PIXE and far above7he Fe bulk concentrations in all the HOPG samples. Itis clear, that EDX cannot be used to measure the bulkconcentrations of trace elements in HOPG directly.In case of HOPG samples, where the trace elementsare strongly concentrated in grains, EDX can at leastbe performed on single grains. Figure 7 shows such anEDX analysis on a grain in AC ZYA. First, BE imaging X-ray energy (keV)
AC ZYA Particle 1 C oun t s Ti V NiFeVC (b)
C TiV NiFeBE (a)
Figure 7: EDX analysis of a single grain in AC ZYA: (a) BE andelemental maps of a 2 . µ m × . µ m sized grain. The Fe and Ni mapindicate that both elements are enriched at the outer edge as alreadyseen in Fig. 4. (b) EDX spectrum showing peaks for the 3d-metalsinside the grain. The peaks of Fe and Ni su ff er from a comparablypoor statistics. is used to detect single grains directly below the graphitesurface due to the Z -contrast in electron yield. Then, asmall scan is made and the emitted X-rays are recorded(see Figs. 7(a) for the BE and elemental maps and (b) forthe EDX spectrum). From the spectrum, however, nodirect information on depth and thickness of the graincan be obtained in contrast to RBS in ion beam mi-croscopy, making a quantitative analysis, e.g. the de-termination of the metallic content, di ffi cult. Qualita-tive analysis though shows that the Ti / V concentrationratio matches quite well the results from PIXE / RBS onthe grain in Fig. 4 from the same sample, whereas theFe and Ni concentrations are both twice as large as for PIXE. This might be due to di ff erences in the compo-sition of individual grains as pointed out earlier and / ordue to the rather poor statistics in the EDX spectrum(see Fig. 7(b)) and the insu ffi cient knowledge of grainthickness and depth.Assuming that the metallic content of single grainscan be accurately determined with EDX, bulk concen-trations can in principle be estimated from these data aswas done in [9, 11]. This approach, however, has severalweaknesses: (i) it requires that all the grains in a sam-ple are comparable in metallic content and compositionwhich is not necessarily the case as Fig. 5 shows; (ii) thenumber density of grains must be determined, e.g. us-ing BE imaging. Since only a very low number of grainsis present in HOPG in the near-surface area of (cid:46) µ mdepth even for large scan areas (in [11] only three grainsare visible for AC ZYA in a 1 . × . ffi culties lead us to the conclu-sion that EDX is useful for identifying and imagingmetallic grains in HOPG, but cannot be considered asa reliable method for quantitative trace element analy-sis in graphite, in contrast to truly bulk-sensitive tech-niques as e.g. PIXE / RBS or INAA. Indeed, the Fe con-centration in AC ZYA was estimated from SEM analy-sis (and magnetization data) to 6 µ g / g in [11], whereasPIXE analysis always gave < µ g / g Fe for numerousof AC ZYA samples in the last ten years. This discrep-ancy might be due to the grain density which is statedto be about 0.25 per 100 µ m × µ m × . µ m vol-ume, i.e. 5 × cm − for AC ZYA [11], about a fac-tor eight higher than our estimations and what can bejudged from the BE image in [11] itself.
4. Magnetic properties
As explained in the introduction, one of the aims ofthis work is to correlate the magnetization behavior ofthe HOPG samples with the one we can estimate tak-ing into account the impurity concentration obtained byPIXE. In general, information on the impurity concen-trations is mandatory in order to understand the originof any unusual magnetic behavior of nominally non-magnetic samples. However, as we will point out below,the concentration values alone are not su ffi cient to pre-dict the behavior of the magnetization, just because themagnetic impurities in a graphite matrix can show dif-ferent magnetic response upon several details, like theirmagnetic state (composition) and grain size. As an ex-ample, we note that Fe in graphite not always shows a8erromagnetic behavior [7]: a sample with an inhomo-geneous Fe concentration of up to 0.38% (in weight)shows no magnetic order. If we implant single Fe atomsrandomly distributed in a disordered graphite lattice onedoes not expect magnetic order, as confirmed after im-planting Fe up to concentrations of 4000 µ g / g [27].For the characterization of the magnetic behavior ofthe HOPG samples we measured the field and temper-ature hysteresis of the magnetic moment. In order tocompare di ff erent samples with di ff erent masses, themagnetization values given in this study are always nor-malized to the whole sample mass. Figure 8 MDTMDTMDT
Figure 8: (a) Magnetization at 300 K for the three AC samples. Alinear diamagnetic background M D = − . × − µ H emu / gT hasbeen subtracted in all the three samples. (b) The same as in (a) but forthe NT-MDT samples. The subtracted linear diamagnetic backgroundwas M D = − . × − µ H emu / gT for the NT-MDT ZYA and NT-MDT ZYH samples and M D = − . × − µ H emu / gT for the NT-ZYB sample. shows the field hysteresis for six samples from twodi ff erent companies at 300 K. The diamagnetic linear-in-field background has been already subtracted fromthe raw data (see caption for their values). Due tothe parallel field direction, this background is relativelysmall and the ferromagnetic hysteresis is clearly recog- nized even without subtraction in most of the samples.From these hysteresis curves we obtain the magnetiza-tion values at saturation M sat ( µ H = ff erenttemperatures and the remanence, i.e. M ( µ H = | µ H | = µ g / g Fe in graphite means about 2.1 ppm of Fe)with exception of AC samples where the lowest Fe con-centration is found for the ZYB-grade sample and thesample with the highest Fe concentration ( (cid:39) . all Fe present in the sample would behave aspure ferromagnetic Fe, Fe O or Fe C, the latter onebeing the most likely case as discussed above. We rec-ognize that a few samples show saturation magnetiza-tion values above the one estimated assuming a specificmagnetic behavior for the Fe impurity. Let us com-pare the sample AC ZYH with sample AC ZYA. As-suming that 1 µ g / g of ferromagnetic Fe (alternativelyas compounds Fe O or Fe C) in graphite would pro-duce a magnetization at saturation of 2 . × − emu / g(1 . × − emu / g for both Fe O and Fe C [26, 28]),if all the measured Fe would be ferromagnetic, for sam-ple AC ZYH we would have the magnetization values atsaturation of 5 . × − emu / g (3 . × − emu / g) and forsample AC ZYA 1 . × − emu / g (1 . × − emu / g),i.e. a ratio of (cid:39)
32 between the two samples. The ra-tio between saturation values at 5 K between those twosamples is (cid:39)
52, i.e. about 62% larger than the aboveestimated ratio. Nevertheless, and making the unrealis-tic assumption of the impurity in the samples being pureFe, we would conclude that the measured Fe concentra-tion roughly explains the absolute values as well as thedi ff erence in magnetization at saturation if all Fe wouldbe ferromagnetic.9
YA ZYB ZYH ZYA ZYB ZYH -MDT-MDT-MDT-MDT
ZYA ZYB ZYH -MDT
Figure 9: (a) Magnetization at saturation at 300 K and 5 K for the ninesamples vs. their grade. (b) Remanence vs. grade. The saturation andremanent magnetization have been obtained from the field hysteresisloops after subtraction of the diamagnetic background, see Fig. 8 asexample. (c) Fe concentration determined by PIXE vs. grade for allthe samples. The lines are a guide to the eye only to allow a bettercomparison between the subfigures. -MDT
Figure 10: Magnetization at saturation at 5 K vs. the maximummagnetization estimated assuming that all the Fe impurities wouldbe ferromagnetic Fe (full symbols), ferrimagnetic magnetite Fe O or cementite Fe C (open symbols), in double logarithmic scale. Thedashed line indicates a one-to-one correspondence between the mea-sured absolute values and the estimated ones.
This apparent correlation is less clear for other sam-ples of di ff erent origin. For example, for the samplesNT-MDT ZYH and NT-MDT ZYA the ratio betweenmagnetization at saturation at 5 K is 2 . × − / . × − = .
2, see Fig. 10, a ratio that does not agree withthe expected ratio of 19 if all
Fe would contribute tothe ferromagnetic signal. Furthermore, if we comparesamples with similar Fe concentration but of di ff erentorigins, i.e. the three samples with expected Fe mag-netization around 1 . × − emu / g or the three around2 × − emu / g (see full symbols in Fig. 10), we rec-ognize also that a large variation of the measured mag-netization for similar Fe concentrations exists. A factthat should be not surprising since in general Fe is nothomogeneously distributed in the micron-sized impu-rity grains as revealed by PIXE elemental imaging (seesection 3.3) and shown in [11] for an AC ZYA sample.It is rather unlikely that it would provide the expectedmaximum magnetic moment at saturation for Fe, Fe O or Fe C. In fact, the saturation magnetization for theNT-MDT ZYA sample, where no local Fe enrichmentin the form of grains was found, is 2.3 or even 3.6 timeslarger than the highest expected saturation magnetiza-tion from Fe or Fe C, respectively. Even more, takinginto account that Fe is most likely present in the formof Fe C, only the saturation magnetizations of the ACZYA, SPI-1 and SPI-3 samples are still compatible withthe observed impurity content, whereas the majority ofsamples shows magnetization values clearly above thepossible contribution from all found magnetic impuri-ties (note that apart from Fe, the maximum possible con-10ribution of all other magnetic elements remains negli-gible, in comparison).One might speculate that ferromagnetic nanoparti-cles, which can provide enlarged magnetic moments perimpurity atom (e.g. Fe nanoparticles as small as 14 nmprovide magnetic moments of 3 µ B per Fe atom [29]),are responsible for the excess in magnetic moment com-pared to the impurity concentration. However, from ourmeasurements we can rule out any significant contribu-tion from such nanoparticles, e.g. features like blockingtemperature and superparamagnetism that would showup in the measurements if su ffi ciently small nanoparti-cles would be present in the samples were not observed.Furthermore, magnetism from Fe, magnetite or cemen-tite nanoparticles is always characterized by a strong de-crease of the saturation and remanent magnetization aswell as coercive force with increasing temperature (see,e.g., [30]). However, the opposite is observed in oursamples. As was shown in [31], even for large nanopar-ticles of up to more than 100 nm diameter and their ag-glomerates the coercive force at 300 K drops down toonly 34% of the value at 5 K, whereas for our samplesthe coercive force at 300 K still is 80% of the 5 K value.We note that in [9] it was stated that no ferromag-netic signal and no impurity grains were found for SPI-2and -3 in contrast to our findings. Furthermore, a largevariation in saturation magnetization was reported be-tween di ff erent NT-MDT ZYB and ZYH samples, againin contrast to our findings were the M sat values as wellas the Fe contents are nearly identical for both samples(see Fig. 9). Interestingly, the NT-MDT ZYA sampleshowed the highest saturation magnetization among allsamples in [9], despite having a Fe concentration abouta factor 20 lower than the other samples – according toour trace element analysis. One may tend to explainthis clear deviation with strong variations in the impu-rity content between di ff erent batches of NT-MDT (andSPI samples as well). This, however, and from our ex-perience with the HOPG samples from Advanced Ce-ramics, seems unlikely. It might, therefore, also indicatethat no simple correlation exists between impurity con-tent and magnetic properties – as shown in this study.In [11] the saturation magnetization of the SPI-2 sam-ple is about twice as large and the Fe content measuredwith INAA almost a factor three higher than for oursample. Since INAA is a suitable method for trace el-ement analysis, the deviations between both findingsmight give an indication on the extent of variation be-tween di ff erent batches of SPI-2 samples. However,since no information was given in [11] on sample clean-ing and due to the lack of imaging capabilities in INAAanalysis, it cannot be excluded that contaminations like the sideface contamination shown in Fig. 2(a) have beenoverseen in [11]. Nevertheless, assuming that Fe ispresent as Fe C in line with their own findings, thesaturation magnetization is about 35% larger than ex-pected from the Fe concentration. More surprisingly,the saturation magnetization of the AC ZYA samplewas found to be about a factor eight larger than in ourstudy. From this value Venkatesan et al. estimated theFe concentration in AC ZYA to be 6 µ g / g [11], a valuethat clearly contradicts our measurements as discussedabove. Whatever the contributions to this large magneti-zation are, it is far above the value that can be attributedto the impurity content inside the HOPG bulk itself.From our findings presented above we conclude thatin some of the HOPG samples, specially in those withlow enough Fe concentration, an extra mechanism con-tributes to the observed magnetic order. This conclusionis similar to that obtained in [7]. As noted in the intro-duction, further support to this conclusion is obtainedby the temperature dependence of the remanent magne-tization discussed in the next section. Di ff erences in the temperature dependence of themagnetization at saturation or at remanence can alsoprovide a way to discern whether ferromagnetic con-tributions from magnetic impurities are at work in thesamples. For all measurements done under a mag-netic field, the temperature dependence of the total mag-netic moment of a HOPG sample (in the case dis-cussed here with field parallel to the graphene layers)is given by the sum of: (i) The intrinsic Landau dia-magnetism of the HOPG sample, which is given by asmall misalignment between the field and the parallelto the graphene planes direction; (ii) The possible tem-perature dependence of the magnetic contribution fromthe substrate / sample holder, which in general should benegligible, if an appropriate holder is used; (iii) Thetemperature dependence of the ferromagnetic contribu-tion itself. In the case of sample NT-MDT ZYB, forexample, we measured M (5 K) = − . × − emu / gand M (300 K) = − . × − emu / g at a field of 1 T.We note that the overall change of the diamagnetic sig-nal is 8 × − emu / g, whereas (2 ± × − emu / g isthe apparent change of the ferromagnetic contribution atsaturation field. The uncertainty in the temperature de-pendence of the corresponding diamagnetic backgroundis a non-negligible source of error when a quantitativecomparison of the M sat ( T ) with appropriate models isrequired. We note that the di ff erence in the temperaturedependence between the magnetic moment at 2 kOe ap-plied field and that at remanence of a HOPG sample,11ee Fig. 10 in Ref. [7], is not intrinsic but it is due to theconstant diamagnetic background subtraction assumedfor simplicity in that work. If one takes into accountthe weak temperature dependence of the diamagneticbackground at 2 kOe, an extra temperature dependentcorrection of (cid:46) × − emu has to be taken into ac-count, enough to remove the di ff erence between the twomagnetic moment’s temperature dependence. M agne t i c M o m en t ( B / M n ) Temperature T(K)
FCC ( H = 0.1T) REM ( H = 0T) m
FCC /m REM
Ru-LSMO
Figure 11: Red squares show the magnetic moment (in units of Bohrmagneton per Mn atom) of the 40 nm thick Ru-doped LaSrMnO ox-ide film measured by cooling in a field of 0.1 T (FCC). The openblue circles are the remanent magnetization measured at zero field bywarming. The remanence is completely irreversible, as expected. Theblack stars show the ratio between the magnetic moments; in this casethe same y − scale applies but unitless. Therefore, one choice is to compare the remanentmagnetization measured at zero applied field. In caseof the ferromagnetism found in the HOPG samplesthe remanent (zero field) magnetization M rem ( T ) canshow a similar temperature dependence as the satu-ration one M sat ( T ) (or at applied fields below satura-tion as observed experimentally in [20]), at least ina temperature region clearly below the correspondingCurie temperature. This is expected when the appliedfield does not change the energy landscape of the do-main walls and when there is no magnetic anisotropythat strongly changes with temperature. In general thisis achieved applying the magnetic field parallel to aneasy axis of the ferromagnetic material. The similar-ity between the temperature dependence of M sat ( T ) and M rem ( T ) can be observed in hard as well as soft ferro-magnets. As an example of some ferromagnets where M rem ( T ) / M sat ( T ) (cid:46) . ± .
5% up to300 K, below the Curie temperature T C (cid:39)
350 K.In the case of our HOPG samples let us take the mea-sured values for the sample NT-MDT ZYB: At 5 K wehave a M rem = . × − emu / g and M sat = . × − emu / g; at 300 K these values are 4 . × − emu / gand 2 . × − emu / g, respectively. Note that the fieldwas always applied parallel to the graphene planes ofthe samples, i.e. parallel to an easy axis. The ratios at5 K and 300 K between the two magnetizations are 0.23and 0.21 with an error (cid:38) .
01. In other words, theseresults clearly indicate that the changes of both magne-tization values with temperature are similar.In Ref.[20] the temperature dependence of the smallferromagnetic moment of graphite produced by pro-ton irradiation has been determined with certain ac-curacy at di ff erent finite fields, subtracting the “afterirradiation” signal minus the “before irradiation” sig-nal, avoiding in this case arbitrary diamagnetic subtrac-tions. Those results showed that the temperature de-pendence of the magnetization remains the same, inde-pendently of the applied field. Furthermore, we notethat the changes in magnetization with temperature arerelatively small, i.e. 10 . . .
20% within two orders ofmagnitude change in temperature. This indicates thatthe measured behavior occurs still far away from theCurie temperature and the number of domains does notchange drastically with temperature. Second, the mag-netically ordered regions are very probably relativelysmall in size, with a relatively weak, if at all, magneticanisotropy and / or pinning of domain walls. In this case,the change in M rem ( T ) would be a thermally driven ro-tation of the magnetization vector, a process that is ba-sically spin wave excitations. Obviously and due to itsirreversible behavior, only the warming up curve of the M rem ( T ) curve can be compared with appropriate spinwave excitations models. As a function of tempera-ture and by warming, the average magnetization vectorgoes through new potential minima that cannot be over-whelmed when the temperature is lowered, showing atemperature hysteresis typical for magnetically orderedmaterials, see Fig. 12.We expect that di ff erent mechanisms will contributeto the remanence and influence its temperature depen-dence, depending whether large or small magnetic re-gions contribute. (i) For large enough ferromagnetic12e particles or regions where due to the density of de-fects and their distribution within the layered structurethe magnetization vector can be considered to be ina 3D potential, we expect to observe a T -dependenceof the remanent magnetization compatible to excita-tion of spin waves following the, e.g. 3D Bloch T / model. In this case we may have a law of the type M ( T ) = M (0)(1 − CT / ), where C is a constant re-lated to the spin wave sti ff ness and a fitting parameter.We note that this simple law applies only at T ≤ . T C ( T C is the Curie temperature of the material) [33].(ii) For samples with a weaker 3D ferromagneticcontribution, however, a quasi-linear temperature de-pendence for the remanent magnetization has been ob-served [5, 8]. This dependence can be understoodwithin the 2D Heisenberg anisotropic spin wave model.We note that this mechanism, first observed in pro-ton irradiated HOPG samples [20] (see also [34]) sug-gests that defects, whatever their origin, within thegraphene planes are responsible for triggering the ob-served magnetic order. The quasi-linear temperaturedependence is an indication of excitation of 2D spinwaves that reduce the magnetization linearly with T [35–37]. The 2D spin-waves magnetization follows as M ( T ) (cid:39) M sw ( T ) M I ( T ) obtained using perturbation the-ory techniques up to third order in spin waves [36, 38]; M sw ( T ) is the magnetization due to spin waves and M I ( T ) is due to an Ising model with the exchange renor-malized by the spin waves, for more details see [20] andreferences therein.(iii) Superparamagnetism is a possible third mecha-nism that can a ff ect the temperature dependence of theremanent magnetization. According to [39] it follows asimple 1 / T dependence that can be added to the temper-ature dependence due to the other contributions. Thiscontribution that has its origin in small enough ferro-magnetic clusters, is expected to contribute mainly atlow enough temperatures.As an example of the di ff erent contributions we showin Fig. 12 the results for the remanent magnetization ofthe NT-MDT samples measured after cooling the sam-ple in a field of 1 T, where all three mechanisms canbe observed. The observed hysteresis between warm-ing and cooling, see Fig. 12, is a clear evidence for theexistence of a ferromagnetic state with Curie temper-ature above 300 K. The sample NT-MDT ZYA showsa behavior compatible with the sum of the contribu-tions due to the mechanisms described in (ii) and (iii),i.e. a quasi-linear contribution plus a superparamag-netic state responsible for the low-temperature behav-ior (the value of the used parameters in the fits are in-cluded in the figure caption), see Fig. 12(a). Sample MDTMDTMDT
Figure 12: Remanent magnetization (at zero field) vs. temperature forthe three NT-MDT samples (after cooling them from 300 K to 5 K un-der a field of 1 T) on warming (5 K →
300 K) and cooling (300 K → T c =
600 K,spin-wave critical temperature due to low-energy spin-wave excita-tions T SWc =
870 K and anisotropy ∆ = .
001 (see [20]) plus4 . × − / T [emu K / g] due to a superparamagnetic contribution. (b)As in (a) but for the NT-MDT ZYB sample. The continuous blue linefollows the case (i) described in the text with n = /
2. For com-parison the dashed red line follows the 2D-Heisenberg anisotropicspin waves model with parameters T c =
500 K, T SWc = ∆ = . T c =
550 K, T SWc = ∆ = . n = /
2. Theinset blows out the low temperature region. ∝ (1 − CT / ) very probably in this case dueto the large Fe impurities or disordered clusters. Theobserved behavior in this sample cannot be fitted withthe 2D anisotropic Heisenberg spin wave model, inde-pendently of the chosen parameters. Figure 12(b) andits inset show the calculated curves; one can clearly re-alize, especially at low temperatures that the data do notfollow the 2D anisotropic Heisenberg spin wave model.On the other hand, the behavior of the remanent mag-netization of sample NT-MDT ZYH is compatible withthe 2D anisotropic Heisenberg spin wave model, seeFig. 12(c) and its inset, even though the amount of Fecontamination and the size of the grains are practicallythe same as in NT-MDT ZYB (see Tab. 1 and Fig. 5 forcomparison).In spite of the above discussed quantitative and qual-itative di ff erences between the measured magnetic re-sponse in some of the HOPG samples, the apparent cor-relation between Fe impurity concentration and mea-sured magnetization at saturation shown in Fig. 9 mayleave a skeptical reader with the feeling that all the mag-netic signal is due to impurities. We note, however, thatin the way the HOPG samples are produced [40], thereshould be a correlation between defect concentration(as well as hydrogen or other non-magnetic impurities)and the magnetic impurity concentration. The HOPGsamples with less Fe impurity, grade ZYA, are obtainedat larger annealing temperatures than those with gradeZYB or ZYH. Therefore, it should be not surprisingthat both, the Fe concentration and the defect density,are correlated. In fact, a direct correlation between thenumber of Fe atoms and defects, i.e. an increase in thenumber of cation vacancies proportional to the Fe con-centration, was already reported in TiO [41, 42].Independently done studies of the change of the mag-netic signals after annealing at 2100 ◦ C, a temperaturewell below the temperatures at which HOPG is pro-duced, show a reduction of the saturation magnetizationby a factor of two. This can only be explained by assum-ing that part of the magnetism observed is due to defectswhich can be removed by annealing and therefore sup-ports the existence of a DIM contribution in HOPG sam-ples of high grade and low impurity concentration [43].Finally, we note that XMCD measurements at the nearsurface region of untreated HOPG samples revealed theexistence of magnetic order at the carbon K-edge, whichis not related to any magnetic impurity[44]. Those re-sults support our main conclusion.
Conclusions
In this work we have done a complete trace elementanalysis using PIXE and RBS in di ff erent HOPG sam-ples of di ff erent sources. The main impurity that cancontribute to the magnetic response of the samples isFe, showing a maximum concentration of (cid:39) µ g / gfor ZYH grade samples. The contamination at thesidefaces of as-received HOPG samples is notable andcan exceed the one in the sample bulk. A thoroughcleaning is mandatory prior use of HOPG samples incontamination-critical applications. The analysis of sin-gle metallic grains indicates that they are not spheri-cal but quasi flat disks oriented parallel to the grapheneplanes, in agreement with previous reports. The size andcomposition of the grains di ff er substantially betweendi ff erent HOPG samples but also within the same sam-ple.We have studied the elemental distribution of someof the samples with EDX in order to compare its capa-bilities and limitations with those of ion beam analysis.Our results clearly show that EDX cannot be used tomeasure bulk concentrations of trace elements in HOPGwith such a small impurity concentration directly butonly and, to a certain extent, in single grains. More-over, our comparative studies indicate that EDX cannotbe considered a reliable method for quantitative traceelement analysis in graphite, clarifying several weak-nesses and discrepancies in the element concentrationestimates done in the literature.From the field and temperature hysteresis of the mag-netic moment of the HOPG samples we conclude thatin some of the samples an extra contribution, other thanthose from magnetic impurities, to the observed ferro-magnetic magnetization response exists. In agreementwith previous reports this extra contribution is compat-ible with quasi two-dimensional defect-induced mag-netism. The rough correspondence between the mag-netization at saturation and the total Fe concentrationindicates also that defect density and the impurity con-centration can be correlated. No general answer, how-ever, can be given even knowing the nominally mag-netic impurity concentration, to the question whethermagnetic impurities are or are not the reason for the ob-served magnetic response in a given HOPG sample. Acknowledgement
We gratefully acknowledge Francis Bern for pro-viding us with his SQUID measurements of Ru-dopedLSMO film (Fig.11). This work was supported by the14eutsche Forschungsgemeinschaft under contract DFGES 86 / References [1] Stoneham, M.. The strange magnetism of oxides and carbons.J Phys: Condens Matter 2010;22:074211.[2] Andriotis, A.N., Sheetz, R.M., Menon, M.. Defect-induceddefect-mediated magnetism in ZnO and carbon-based materials.J Phys: Condens Matter 2010;22:334210.[3] Volnianska, O., Boguslawski, P.. Magnetism of solids resultingfrom spin polarization of p orbitals. J Phys: Condens Matter2010;22:073202.[4] Yazyev, O.V.. Emergence of magnetism in graphene materialsand nanostructures. Rep Prog Phys 2010;73:056501.[5] Esquinazi, P., Hergert, W., Spemann, D., Setzer, A., Ernst,A.. Defect-induced magnetism in solids. IEEE Transactions onMagnetics 2013;49(8):4668–4674.[6] Kopelevich, Y., Esquinazi, P., Torres, J., Moehlecke, S..Ferromagnetic- and superconducting-like behavior of graphite.J Low Temp Phys 2000;119:691–702.[7] Esquinazi, P., Setzer, A., H¨ohne, R., Semmelhack, C., Kopele-vich, Y., Spemann, D., et al. Ferromagnetism in orientedgraphite samples. Phys Rev B 2002;66:024429–1–10.[8] Esquinazi, P., Barzola-Quiquia, J., Spemann, D., Rother-mel, M., Ohldag, H., Garc´ıa, N., et al. Magnetic order inGraphite: Experimental evidence, intrinsic and extrinsic di ffi / test-structures-HOPG.html; 04.09.2013.[24] Tsuzuki, A., Sago, S., Hirano, S.I., Naka, S.. High-temperature and pressure preparation and properties of iron car-bides Fe C and Fe C. J Mat Sci 1984;19:2513–2518.[25] Vo˘cadlo, A., Brodholt, J., Dobson, D., Knight, K., Marschall,W., Price, D., et al. The e ff ect of ferromagnetism on the equa-tion of state of Fe C studied by first-principles calculations.Earth and Planetary Science Lett 2002;203:567–575.[26] Talyzin, A., Dzwilewski, A., Dubrovinsky, L., Setzer, A., Es-quinazi, P.. Structural and magnetic properties of polymerizedC with Fe. Eur Phys J B 2007;55:57–62.[27] H¨ohne, R., Esquinazi, P., Heera, V., Weishart, H., Setzer,A., Spemann, D.. The influence of iron, fluorine and boronimplantation on the magnetic properties of graphite. J MagnMagn Mater 2008;320:966–977.[28] St¨ablein, F., Schroetter, K.. Die Bestimmung der mag-netischen S¨attigung von Eisencarbid. Z Anorg U Allg Chem1928;174:193–215.[29] Barabara, B., Marchand, A., Mollard, P., Devaux, X., Rous-set, A.. On a series of nanoparticles of iron epitaxed on Al O :a new field, temperature and concentration (of Fe) scaling plotof the magnetization curves. In: Dormann, J.L., Fiorani, D., ed-itors. Magnetic Properties of Fine Particles. Amsterdam: North-Holland; 1992, p. 171–177.[30] O’Grady, K., Chantrell, R.W.. Remanence Curves of FineParticle Systems I: Experimental Studies. In: Dormann, J.L.,Fiorani, D., editors. Magnetic Properties of Fine Particles. Am-sterdam: North-Holland; 1992, p. 93–102.[31] Lipert, K., Ka´zmierczak, J., Pełech, I., Narkiewicz, U.,´Slawska Waniewska, A., Lachowicz, H.K.. Magnetic proper-ties of cementite (Fe C) nanoparticle agglomerates in a carbonmatrix. Materials Science-Poland 2007;25:399–404.[32] Sanz, R., Cerd´an, M., M.E. McHenry, M.D.M.. Temper-ature dependent magnetization and remanent magnetization inpseudo-binary (Fe TiO ) x − (Fe O ) − x (0 . < x < .
0) ti-tanomagnetites. EEE Transactions on Magnetics 2011;47:4124–4127.[33] Kittel, C.. Introduction to Solid State Physics, 7th ed. JohnWiley & Sons, Inc.; 1996. P. 455.[34] Xia, H., Li, W., Song, Y., Yang, X., Liu, X., Zhao, M., et al.Tunable magnetism in carbon-ion-implanted highly oriented py-rolytic graphite. Adv Mater 2008;20:4679–4683.[35] Doring, W.. ¨Uber den Ferromagnetismus des Fl¨achengitters. ZNaturforsch 1961;A16:1008.[36] Levanyuk, A.P., Garc´ıa, N.. The two-dimensional Heisen-berg ferromagnet with various types of interactions: tempera-ture dependence of magnetic parameters. J Phys: Condens Matt1992;4:10277.[37] Serena, P.A., Garc´ıa, N., Levanyuk, A.P.. Monte Carlo calcu- ations on the two-dimensional anisotropic Heisenberg model.Phys Rev B 1993;47:5027.[38] Brezin, E., Zin-Justin, J.. Spontaneous breakdown of continu-ous symmetries near 2 dimensions. Phys Rev B 1976;14:3110.[39] Mart´ınez, B., Roig, A., Obradors, X., Molins, E., Rouanet,A., Monty, C.. Magnetic properties of γ -Fe O nanoparticlesobtained by vaporization condensation in a solar furnace. J ApplPhys 1996;79:2580–2586.[40] Inagaki, M.. New Carbons: Control of Structure and Functions.Elsevier; 2000. ISBN 0080437133.[41] Navarro, A.M., Errico, L.A., Torres, C.R., Cabrera, A.F., Bilo-vol, V., Renter´ıa, M., et al. Ferromagnetism in Transition MetalDoped Rutile TiO . In: Low, I.M.J., editor. Rutile – Properties,Synthesis and Applications. New-York: Nova Science Publish-ers, Inc.; 2012, p. 159–194.[42] Navarro, A.M., Torres, C.R., Bilovol, V., Cabrera, A.F.,Errico, L.A., Weissmann, M.. Study of the relation betweenoxygen vacancies and ferromagnetism in Fe-doped TiO nano-powders. J Appl Phys 2014;115:223908.[43] Miao, X., Tongay, S., Hebard, A.F.. Extinction of ferromag-netism in highly ordered pyrolytic graphite by annealing. Car-bon 2012;50:1614–1618.[44] Ohldag, H., Esquinazi, P., Arenholz, E., Spemann, D., Rother-mel, M., Setzer, A., et al. The role of hydrogen in room-temperature ferromagnetism at graphite surfaces. New Journalof Physics 2010;12:123012.nano-powders. J Appl Phys 2014;115:223908.[43] Miao, X., Tongay, S., Hebard, A.F.. Extinction of ferromag-netism in highly ordered pyrolytic graphite by annealing. Car-bon 2012;50:1614–1618.[44] Ohldag, H., Esquinazi, P., Arenholz, E., Spemann, D., Rother-mel, M., Setzer, A., et al. The role of hydrogen in room-temperature ferromagnetism at graphite surfaces. New Journalof Physics 2010;12:123012.