PHYSICAL REVIEW B 1 NOVEMBER 2000-IVOLUME 62, NUMBER 17

Magnetic, transport, and structural properties of FeÕCoÕCuÕ†CoÕIr ÕCo‡ sandwichesand FeÕCoÕCuÕ†CoÕIr ‡ multilayers prepared by ion-beam sputtering

S. Colis*IPCMS-GEMME, CNRS-UMR 7504, 23 rue du Loess, F-67037 Strasbourg Cedex, France

A. DiniaIPCMS-GEMME, CNRS-UMR 7504, ULP-ECPM, 23 rue du Loess, F-67037 Strasbourg Cedex, France

C. Meny, P. Panissod, C. Ulhaq-Bouillet, and G. SchmerberIPCMS-GEMME, CNRS-UMR 7504, 23 rue du Loess, F-67037 Strasbourg Cedex, France

~Received 3 May 2000!

A study of the structure, transport, and magnetic properties of Co/Ir sandwiches prepared by ion-beamsputtering is presented. Oscillations of giant magnetoresistance~GMR! and coupling strength versus the Irthickness are observed. In the spin-valve-type sandwich, at low Ir spacer thickness a shift is observed betweenGMR and coupling oscillations. This is due to the presence of a magnetic Fe5nm /Co0.5nm /Cu3nm buffer, whichhas an important contribution to the spin dependent scattering, and also to the nature of the indirect exchangecoupling. The maximum GMR and coupling strength are about 3% and20.47 erg/cm2, respectively. The mostinteresting result concerns the nature of the indirect exchange coupling as shown by the NMR analysis. Thiscoupling is homogeneous, and consistent with a biquadratic coupling, instead of the usually observed antifer-romagnetic coupling. This is further supported by low-temperature magnetization measurements and TEMinvestigations, which show that the deposited layers are laterally continuous and free of bridges for 0.5-nm Irspacer layer.

dj

aesu,el.

geifeaticanoro

eegs

thin

tonthl-

nd-

plinglarictheling

re-

di-nd-ex-ws.tagem-sed

ing.a-ner,

r atu-re-

riesto

I. INTRODUCTION

Since the discovery of exchange coupling between acent ferromagnetic layers through nonmagnetic spacers1 andthe related giant magnetoresistance~GMR!,2 the study ofthese systems has received a great deal of attention, becof their considerable fundamental and technological interUsually, antiferromagnetically coupled sandwiches and mtilayers give rise to a high saturation field and, sometimesa high magnetoresistance. However, large saturation fiare not suitable for read heads and sensors applications

overcome this drawback, van den Berget al.3 have devel-oped a different generation of angular sensors, showingood performance, in particular for angle and position dettion systems. In these hard-soft sensors, the strongly antromagnetically coupled sandwich is used to replace the mnetic hard layer, and is called artificial antiferromagne~AAF! sandwich. Therefore the achievement of a perfecttiferromagnetic coupling in a sandwich stack is very imptant for angular sensors. To study this indirect exchange cpling, a large number of combinations of metals has bused, giving rise to systems with oscillatory antiferromanetic coupling and different coupling strength valueAmong these systems, Co/Ru and Co/Rh were foundpresent very strong-coupling strength, in particularCo/Rh system, which showed the strongest AF couplstrength ever observed in sandwiches or multilayers.4 Fromthe electronic point of view, the Ir spacer was expectedgive large AF coupling values. This has been recently cfirmed on Co/Ir multilayers prepared using different growtechniques.5–8 However, there is no report, to our know

PRB 620163-1829/2000/62~17!/11709~10!/$15.00

a-

uset.l-todsTo

ac-r-g-

--u-n-.toeg

o-

edge, on the presence of the AF coupling in Co/Ir sawiches. Moreover, Yanagiharaet al.8 have recently evi-denced the presence of a second order, biquadratic, couon a 100-period Co/Ir multilayer prepared by molecubeam epitaxy~MBE! on a MgO substrate. Such biquadratcoupling gives rise to an orthogonal alignment betweenmagnetization vectors of the adjacent layers. This couphas been earlier observed9–12 and different explanations onits origin have been given, like loss of spins,10 couplingfluctuations,11 or pinholes in the spacer layer.12 Furthermore,the evidence of this effect was based, in most of theseports, on magnetization curves simulations.

The aim of this paper is to show the presence of an inrect exchange coupling in ion beam sputtered Co/Ir sawiches, which exhibits a biquadratic and homogeneouschange coupling feature. The paper is organized as folloThe description of the preparation method and the advanof the stack design are presented in Sec. II. In Sec. III, rootemperature GMR and magnetization curves are discusand indicate a possible existence of a biquadratic couplThis biquadratic coupling is consistent with the NMR mesurements~Sec. IV!. NMR also provides some informatioabout the concentration profile at the interfaces. Moreovan investigation by x-ray diffraction~XRD! and TEM of thestructure, grain size, and interdiffusion between Co and Ithe Co/Ir interface is also presented. Finally, in Sec. V simlations of the low-temperature magnetization curves areported and analyzed in terms of biquadratic coupling.

II. SAMPLE ARCHITECTURE AND EXPERIMENTALMETHOD

Two types of samples have been prepared. First, a seof Co3nm /IrX /Co3nm sandwiches was deposited in order

11 709 ©2000 The American Physical Society

hea

th

n

ea

ngeu

wpinibl

theos te

ndcr

ch

ndetawm

dn

anne

si

icurmin

entt

ea

gt

e

-as a

ee--of

es,ties,

ardtoingAr

s ofeenruc-

thin

ndcerth

f Ir.thcere-er-

theers

-

-is

11 710 PRB 62S. COLISet al.

analyze the variation of GMR and coupling strength with tthickness of the Ir spacer layer. Second, due to the smthickness of the sandwiches, a@Co3.5nm /Ir0.5nm#103

multilayer was prepared to investigate the structure oflayers and the morphology of the Co/Ir interfaces.

All samples were deposited at room temperature oglass substrate covered by a Fe5nm /Co0.5nm /Cu3nm bufferlayer. A Cu2nm /Cr2nm capping layer was used to cover thsamples and protect them against oxidation. Since the mnetic and transport properties of the sandwiches are strodependent on the morphology of the magnetic/nonmagninterfaces, an important effort was made to optimize the sface quality and the structure of these layers. The grodeposition rates and the thickness of the buffer and caplayers were varied in order to improve, as much as possthe magnetic and transport properties.13 The bufferFe5nm /Co0.5nm /Cu3nm presents the advantage to reducesurface roughness of the whole stack. Moreover, its magnnature gives a contribution to the GMR signal and, as a csequence, an increase in the total GMR is expected. ThuGMR signal results from two contributions: the first oncoming from the AAF Co/Ir/Co sandwich and the secoone, called spin-valve GMR, which results from the interation between the Fe/Co soft buffer layer and the AAF halayer. As the buffer layer is decoupled from the sandwithis hard soft system is ideal for magnetic sensors.3

All samples were prepared by ion-beam sputtering~IBS!technique using a two grid Kaufmann14 ion source. The basepressure was about 531029 mbars. The Ar1 ions are inci-dent on the sputter target at 400 V with an angle of 45° athe beam current was around 5 mA. The growth of theposited films was monitored by a vibrating quartz crysoscillator, which is placed close to the substrates. The grodeposition rates were typically 0.75 nm/min for Co, 0.5 nmin for Ir, 0.65 nm/min for Fe, and 1.7 nm/min for Cu.

The magnetoresistance of the samples was measureing a low-frequency ac lock-in technique, with a convetional four in line gold-plated contacts~CIP ‘‘current in-plane’’ configuration!. At room temperature, themeasurements were performed up to 17 kOe applied mnetic field, both parallel and perpendicular to the in-placurrent direction, in order to detect any anisotropic magtoresistance contribution.

The magnetization measurements were carried out uan alternating gradient force magnetometer~AGFM! ~Ref.15! and a superconducting quantum interference dev~SQUID! magnetometer, respectively, for room-temperatand low-temperature measurements. In both cases, thenetic field was applied in the plane of the film and reacha maximum of, respectively, 13 kOe~AGFM! and 50 kOe~SQUID!. The AGFM measurements have been performwith the magnetic field parallel to a reference direction aalso at several angles with respect to this direction. Sincemagnetization curves are superimposed, we deduced thahave no uniaxial anisotropy.

The x-ray measurements were performed using a Siempowder diffractometer.u/2u scans were carried out usingparallel monochromatic Co-Ka1

radiation. Small-angle x-raydiffraction was used to check the superlattice period lenand the superlattice quality.

Nuclear magnetic resonance~NMR! measurements wer

ll

e

a

g-ly

ticr-thg

e,

eticn-he

-d,

d-lth/

us--

g-e-

ng

eeag-g

ddhewe

ns

h

carried out at 1.5 K on a Co3nm /Ir0.5nm /Co3nm sandwich anda @Co3.5nm /Ir0.5nm#103 multilayer, using a broadband automated spectrometer. Measurements have been takenfunction of the applied radio frequency~RF! field H1strength, at frequencies between 50 and 250 MHz. A thrdimensional~3D! surface, intensity as a function of both frequency and H1 strength, is then obtained. The advantagethis method is to visualize the structural inhomogeneitialong the frequency axis, and the magnetic inhomogeneialong the rf field axis.

Finally, transmission electron microscopy~TEM! obser-vations were carried out using a Topcon EM002B standmicroscope, which operates up to 200 kV with a pointpoint resolution of 0.18 nm. Specimens were prepared usstandard technique of mechanical thinning combined withion-beam etching.

III. MAGNETIC AND TRANSPORT PROPERTIES

A. Sandwiches

To study the magnetic and transport properties, a seriesymmetrical sandwiches with a variable Ir thickness has bprepared. The deposited stack presents the following stture: Fe5nm /Co0.5nm /Cu3nm /Co3nm /IrXnm/Co3nm /Cu2nm /Cr2nm . Oscillations of GMR and exchange coupling strengvs Ir spacer thickness5 have been observed and reportedFig. 1. As clearly seen, the first maximum in the GMR athe exchange coupling do not occur at the same Ir spalayer thickness. The first maximum of the coupling strengand of the GMR appear at, respectively, 0.5 and 0.3 nm oFrom the theoretical point of view, the coupling strengshould present a maximum for one monolayer spathickness,17 but experimentally this is hard to achieve, bcause of the easy mixing of the Co and Ir at the Co/Ir intfaces. Below a certain Ir thickness, bridges appear acrossspacer and the direct coupling between the two Co layfavors the ferromagnetic~FM! coupling, instead of the antiferromagnetic~AF! coupling, lowering the intensity of theindirect exchange coupling.

We will try now to understand the origin of the shift between the GMR and the coupling strength oscillations. It

FIG. 1. Magnetoresistance~full squares! and coupling strength~open circles! versus Ir spacer thickness of a Co3nm /IrXnm/Co3nm

sandwich deposited on a Fe5nm /Co0.5nm /Cu3nm buffer layer. Thesolid and dotted lines are only a guide for the eye.

thhcathn

g

lyort ti

tho

ftethththrob-

exunthnn

, atheatenn

ichionthe

of

lyich.hei-er,is

n-

i-erti-

ever,ua-

sa

he

sag-

he

PRB 62 11 711MAGNETIC, TRANSPORT, AND STRUCTURAL . . .

important to note first that this effect is observed due topresence of the soft magnetic Fe/Co layers in the buffer. Tadds an interesting contribution to the spin-dependent stering mechanism. Thus we have to examine carefullymagnetization and the GMR curves for ferromagnetic aantiferromagnetic sandwiches.

For very thin Ir spacer layers, the coupling is ferromanetic as shown in Fig. 2 for the Co3nm /Ir0.3nm /Co3nm sand-wich. Therefore, the contribution to the GMR comes mainfrom the spin-dependent scattering due to the differententations of the soft magnetic buffer layers, with respecthe magnetization of the whole Co/Ir/Co sandwich. Thissimilar to a hard/soft structure, where the Fe/Co layer inbuffer and the Co/Ir/Co sandwich are, respectively, the sand the hard layers. As shown in Fig. 2 using arrows, areversing the magnetic field, the magnetization vector ofsoft buffer is switched from the parallel configuration, wirespect to the magnetization vectors of the Co layers insandwich, to the antiparallel configuration. Such a 180°tation will give rise to the maximum GMR that can be otained, which is about 3%.

Increasing the thickness of the Ir spacer layer, thechange coupling is increased to reach its maximum aro0.5-nm Ir spacer layer. If we suppose in this case thatcoupling between the Co layers through Ir is perfect aantiferromagnetic,5 the magnetization vectors of the adjace

FIG. 2. Room-temperature magnetoresistance@~a! current in-plane configuration! and magnetization ~b! curves of aCo3nm /Ir0.3nm /Co3nm sandwich. The magnetic applied field waparallel to the film plane. The dotted arrows correspond to the mnetization of the buffer, while the full arrows correspond to tmagnetization of the two Co layers of the trilayer.

eist-ed

-

i-oseftre

e-

-dedt

Co layers are antiparallel and contained in the film planezero field. After reversing the applied magnetic field, tmagnetization vector of the soft buffer layers should rotby 180°, from the parallel to the antiparallel configuratio~or vice versa! with respect to the next Co magnetizatiovector. In this situation the bottom Co layer of the sandwwill be antiparallel to both the top Co layer and the detectone. In addition, if we consider that the resistance ofsample does not change significantly with the depositionan additional Ir monolayer~0.5-nm Ir compared to 0.3 nmfor the ferromagnetically coupled sandwich!, this situationleads to a spin-valve GMR contribution, which is slightlower than the previous one, in the ferromagnetic sandwThus, adding to this signal, the GMR contribution due to tAAF Co/Ir/Co sandwich, the total GMR signal in the antferromagnetic sandwich should be close, or slightly highto the GMR in the ferromagnetic sandwich. Surprisingly, this in contradiction with what is experimentally observed. Ideed, the GMR value of the Fe5nm /Co0.5nm /Cu3nm /Co3nm /Ir0.5nm /Co3nm stack, corresponding to the first maxmum in the coupling oscillation, reported in Fig. 3 is lowby 40% than the GMR value obtained for the ferromagnecally coupled sandwich.

A reduction up to 15% in the GMR amplitude for thstrongly exchange coupled Co/Ir/Co sandwich can, howebe explained if the indirect exchange coupling has a biq

g-

FIG. 3. Room-temperature magnetoresistance@~a! current in-plane configuration# and magnetization ~b! curves of aCo3nm /Ir0.5nm /Co3nm sandwich. The magnetic applied field waparallel to the film plane. The dotted arrows correspond to the mnetization of the buffer, while the full arrows correspond to tmagnetization of the two Co layers of the trilayer.

in

icth

aootu

enithuethththFuze

nerldesetsionr tcotiowpoyesbe

ce

ca

tiuredtiove

ttticinO

senstea

ighR

eite

-m

lf,ed

za-ichif-

ion.rthemeesthetheof

eshes

pedf athe

sgeurysi-

onichave

llec-

ang

11 712 PRB 62S. COLISet al.

dratic nature. This can be calculated using the followexpression:16,25 MR; 1

2 (12cosu), whereu is the angle be-tween the moment of the bottom Co layer of the sandwand the moment of the Fe/Co layer of the buffer. Forferromagnetic~0.3 nm of Ir! and the biquadratic~0.5 nm ofIr! coupling,u is, respectively, 180° and 135°. Moreover,reduction up to 50% can also be explained considering bthe presence of biquadratic coupling and simultaneous rtion of the buffer and sandwich magnetizations. The simtaneous rotation of the buffer and the sandwich net momis more likely to occur in the case of the sandwich w0.5-nm Ir than in the sandwich with 0.3-nm Ir. Indeed, simlations of the magnetization have shown that with a givanisotropy, the coercive field is sensitively reduced bypresence of a small biquadratic coupling, compared tocase without coupling. Therefore the coercive fields ofsandwich and the buffer layers are close to each other.thermore, the absence of a plateau and of a maximum atfield in the magnetoresistance curve indicate that there iswell defined antiferromagnetic state. In addition, around zfield, the variation of the resistivity against the magnetic fiefollows a steep variation, instead of a parabolic one, aspected for a perfect antiferromagnetic coupling. On the baof the above analysis we can conclude that the presencbiquadratic coupling qualitatively explains the GMR resul

Therefore we have to look carefully to the magnetizatcurve for the same sandwich, reported in Fig. 3, in ordehave a more precise idea on the nature of the exchangepling. The figure shows a steep increase of the magnetizaup to about 0.7 times its saturation value, followed by a sluptake to the saturation. This high remanence is more imtant than the one expected from the Fe/Co detection laand cannot be attributed to errors in the layers thicknevaluation. The existence of a non-negligible differencetween the expected and the measured remanence valuehowever, be explained by the presence of a biquadraticchange coupling.

To have an idea on the exchange coupling strength,culations have been done using the formula:2J5HS•(MR2MS)•tCo/2, whereHS is the saturation field,MR and MSthe remanent and, respectively, the saturation magnetizaand tCo the thickness of one Co layer. The maximum copling strength is obtained for 0.5-nm Ir thickness and corsponds to 0.54 erg/cm2. This value is certainly overestimatesince it corresponds to a linear variation of the magnetizauntil the saturation. This is far from our magnetization curshowing a large remanence and a strong curvature belowsaturation. On one hand such a shape of the magnetizaloop could result from an inhomogeneity of the magneanisotropy within the layers, and/or a large scale lateralhomogeneity of the coupling strength between the layers.the other hand such a shape is also expected in the cabiquadratic coupling. However, the GMR observatioquoted above and the NMR results that will be presenlater favor the last hypothesis. Nevertheless, our valuessmaller than the values reported on Co/Ir multilayers5–8 andevidences the difficulty to obtain sandwich stack with a hand perfect indirect exchange coupling. However, the GMvalue of this sandwich is about 2%, which is seven timlarger than the one measured on Co/Ir multilayers depos

g

he

tha-l-ts

-neeer-rooo

x-isof.

ou-

on

r-ers-an,x-

l-

on--

n,heion

-nof

dre

sd

by the same technique~0.3%!.5 This can, however, be explained by the strong contribution to the GMR coming frothe magnetic Fe/Co buffer.

B. Multilayers

In order to study the properties of the Co/Ir stack itsemagnetization and resistivity measurements were performalso on the@Co3.5nm /Ir0.5nm#103 multilayer, where the influ-ence of the buffer layer is largely reduced. If the magnetition loop does not differ much from the one of the sandwwith the same Ir thickness, the GMR curve is, however, dferent due to the strong decrease of the buffer contributThe GMR ratio for this multilayer is of 0.5%, which is foutimes less than the sandwich value, but still larger thanvalue obtained for Co/Ir multilayers deposited by the saIBS ~Ref. 5! technique on a Fe/Ir buffer. This demonstratagain the importance of the nature of the buffer layers onproperties of our samples. The shape of the GMR loop ofmultilayer is very close to those of sandwiches, but boththem are still far from a parabolic variation. This indicatthat the magnetic and transport properties of the sandwicand multilayers are similar.

The GMR and magnetization curves analysis develoabove gives a qualitative indication of the existence obiquadratic coupling in our sandwiches. However, sincestructure and the morphology of the interfaces of thin filmand multilayers18 have a strong influence on the exchancoupling and on the GMR, a detailed structural study of osamples is therefore very important to understand the phcal origin of the magnetotransport properties.

IV. STRUCTURE AND INTERFACE INVESTIGATIONS

A. XRD measurements

Due to the small thickness of the layers, x-ray diffractidoes not allow us to see any diffraction peak for a sandwstack. For this reason, x-ray-diffraction measurements hbeen performed on a@Co3.5nm /Ir0.5nm#103 multilayer, in or-der to obtain information on the quality of the interfaces. Athe diffraction patterns were obtained with the scattering v

FIG. 4. Low angle x-ray-diffraction pattern spectra for@Co3.5nm /Ir0.5nm#103 multilayer recorded at room temperature usithe Cu radiation withlKa1

50.154 nm.

algaee

4.n

eren

io-

ceei

el

ge-l’’th

plesrs

her-a

ntaltheec-tionis-g

be-de,obe-s too-

thean-

t be

e-they

tat-isuse

t beon.s-ed.ar

cil-

rf

getiotwetheth

e

nti-.

PRB 62 11 713MAGNETIC, TRANSPORT, AND STRUCTURAL . . .

tor perpendicular to the film plane. Figure 4 shows the smangle x-ray-diffraction spectrum. The presence of the Brapeaks and the Kiessig fringes is an indication of a smrange roughness of the interfaces quality, with a well definperiodicity. From the Bragg peaks, we determine the suplattice period, which gives rise to an average value of60.1 nm. This is in good agreement with the nominal o(L54 nm!.

B. NMR measurements

In order to understand the origin of the observed intlayer exchange coupling and its nature, NMR measuremwere performed at low temperature on a@Co3.5nm /Ir0.5nm#103

multilayer and on a Co3nm /Ir0.5nm /Co3nm sandwich. Thespectra were recorded for different values of the excitatradio-frequency~rf! field, a method which allows us to measure the restoring field~anisotropy and/or exchange field!acting on the magnetic moments.19,20

1. Multilayers

The observed NMR intensity vs frequency (f ) and rf field(H1) for the multilayer is presented in Fig. 5 as a 3D surfaIn such a picture, the intensity distribution along the frquency axis reflects the structure of the Co layer andinterfaces, whereas the intensity distribution along the rf fiaxis reflects the distribution of the restoring~coupling and/oranisotropy! field (Hres). In the raw NMR data, the restorinfield distribution is convoluted with the nuclear-spin rsponse to the rf field, which leads to an intrinsic, ‘‘naturawidth. Within this resolution, a single peak is observed in

FIG. 5. NMR signal intensity as a function of frequency andfield as observed in a@Co3.5nm /Ir0.5nm#103 multilayer. The narrowdistribution of the rf field evidences the presence of an homoneous coupling. Insets show the oscillations of the magnetizavectors of the two adjacent Co layers, which correspond to thevibration modes. The ‘‘scaled rf field’’ is the actual rf field valumultiplied, for convenience, by an instrumental factor such thatrestoring field is read directly as the ‘‘scaled rf field’’ value for thmaximum signal. The curves along this axis reflect coarselyrestoring field distribution.21

l-glldr-2e

-ts

n

.-tsd

e

present samples, whereas largely inhomogeneous samwould exhibit a much flatter distribution over several ordeof magnitude of the field strength~see Fig. 10 of Ref 20!.19,23

This shows that all the electronic moments magnetize coently, over the whole stack in low field, as expected forhomogeneous ferromagnet. However, the experimecurves exhibit also an upturn for the largest values ofexcitation field, close to the maximum available on the sptrometer. This shows the existence of a second oscillamode of the electronic moments in large fields. The extence of two oscillation modes, with very different restorinfields, is indeed possible if a biquadratic coupling existstween the two adjacent Co layers. In the first vibration mofor low values of the rf field, the Co moments of the twlayers oscillate in phase, keeping constant the 90° angletween them, and the measured restoring field correspondthe magnetic anisotropy only. In the second mode, the mments of the two layers oscillate in phase opposition, andmeasured restoring field arises from both the magneticisotropy and the exchange coupling.

Although they are related, these two modes should noconfused with the ‘‘symmetric’’ and ‘‘antisymmetric’’modes observed by ferromagnetic resonance~FMR! in anti-ferromagnetic materials. Owing to the low excitation frquency the electronic spins are far from resonance andfollow adiabatically the external NMR rf field: from thenucleus viewpoint the electronic spins are essentially nuing. A direct consequence is that the ‘‘symmetric’’ modenot observed in a purely antiferromagnetic system becathe in-phase mutation of the electronic moments cannoexcited by the rf field in the absence of a net magnetizatiOnly the ‘‘antisymmetric’’ mode, corresponding to the tranverse susceptibility of the antiferromagnet, can be observThe situation is, of course, different in case of noncollinecoupling where the net electronic moment is driven in oslation by the rf field.

The NMR spectra, corrected for enhancement factor19,20,22

of the @Co3.5nm /Ir0.5nm#103 multilayer, is reported in Fig. 6.

-no

e

e

FIG. 6. NMR spectra~normalized to the sample total interfacsurface area! of a @Co3.5nm /Ir0.5nm#103 multilayer ~solid line! and ofa Co3nm /Ir1nm /Co3nm sandwich~dotted line!. The width at the halfmaximum of the main lines and the tails of both spectra are idecal, which reflects similar structures and similar Co/Ir interfaces

eee

laaicul

fo

ntte

tncthweumths%tumovethe

hepe

oixelkthtictorte

con

esf de-and-sistra

dh aalithbeme

hetheingtrater

om-ichde-essitedns.plyes,n if

orehatallare

ar-atcesvi-

ique

heo-ein

es,rro-

ere

-ere

les.theast,nnotthethe

ofi-

ms

11 714 PRB 62S. COLISet al.

The hyperfine field (HHF), the concentration profile at thinterfaces and the structure of the bulk Co layer have bdeduced. Generally, an NMR spectrum of a Co/X system,presents a main peak, which corresponds to the bulk Coers, while the low-frequency tail corresponds to the Cooms from the grain boundaries and from interfaces, whhave different chemical environment. The frequency of bCo atoms is expected to be around 226 MHz for a~0001! hcpsample with an in-plane magnetization and at 220 MHzhcp Co with the magnetization along thec axis. The fre-quency for fcc Co is expected at 217 MHz.

In our case the NMR spectrum of the multilayer presea main resonance line at 219 MHz. This cannot be attributo hcp Co with the magnetization along thec axis since themagnetization measurements have clearly evidenced thamagnetization lies in the film plane. Therefore this resonafrequency is intermediate between the NMR frequency offcc and the hcp structures, which means that there in nodefined structure. Moreover, the width at the half maximof this resonance line is large, which can be explained bypresence of stacking faults. The simulation has allowed uextract the amount of stacking faults, which is around 48This is a clear evidence of the absence of a defined strucOn the other hand, the concentration profile of the Co atoat the interfaces is presented in Fig. 7. As expected, the Cinterface is strongly interdiffused and it extends over fimonolayers. Moreover, the concentration of Co atoms inmiddle of the Ir spacer is about 30%, and consequently this no pure Ir atomic plane in the whole stack.

From the contribution of each mixed atomic layer to ttotal spectrum, it is possible to compute the average hyfine field value of each layer, which provides an estimatethe average magnetic moment per Co atom in each mlayer.18,24 The magnetization profile normalized to the bumagnetization is presented in Fig. 7. Co atoms located inmiddle of the Ir spacer lose up to 60% of their magnemoment, and if we express the profile in terms of magnedead layer, 0.14 nm of Co are nonmagnetic at each Cinterface. This value is much smaller than the value repoby Dinia5 ~0.4 nm/interface!, in the case of Co/Ir multilayersdeposited on a Fe/Ir buffer. This is a proof of the importanof the nature of buffer layer and of the growth conditionsthe interface morphology.

FIG. 7. Concentration and hyperfine field profile of Co atoacross an 0.5-nm Ir spacer layer for the Co3nm /Ir1nm /Co3nm sand-wich.

n

y-t-hk

r

sd

theeell

eto.

re.s/Ir

ere

r-fd

e

ic/Ird

e

2. Sandwiches

It is already known that the roughness of the interfaccan increase, or decrease, with increasing the number oposited layers, and consequently differences between swiches and multilayers can occur. To verify this hypotheand to give consistency to our study, another NMR spechas been recorded for a Co3nm /Ir1nm /Co3nm sandwich~thedotted line in Fig. 6!. In this figure the spectra are normalizeto the total interface surface area of the sample. With sucnormalization the NMR spectra of samples with identicinterfaces will be superimposed. From the comparison wthe multilayer spectrum, interesting observations canmade~i! first, the main resonance line appears at the safrequency~219 MHz!, and presents the same width at thalf maximum as the multilayer one, which indicates thatstructure is exactly the same with a large amount of stackfaults; ~ii ! second, the low-frequency tails of the two specare absolutely identical, which proves the similar characof interfaces for both samples. The shape of the signal cing from interfaces is the same in both structures, whmeans that Co/Ir interfaces have the same interdiffusiongree in sandwiches and in multilayers. Thus the roughndoes not change with increasing the number of deposlayers as will be also confirmed by the TEM observatioTherefore the structural results found for a multilayer apas well as for a sandwich. Moreover, at low frequencithere are no peaks coming from the Co/Cu interfaces evein the sandwich their impact on the total signal is much mimportant than in the case of the multilayer. This means tCo/Cu interfaces are interdiffused and gives only a smsignal over a large frequency range. The Co/Cu interfacesstrongly intermixed almost as Co/Ir ones, this being a pticularity of our deposition technique. Nevertheless, Pers25

obtained using classic sputtering, abrupt Co/Cu interfawith almost zero magnetic dead layer. This is another edence on the correlation between the deposition technand the physical properties of the samples.

NMR results have shown a strong interdiffusion at tCo/Ir interfaces, which extends over more than five monlayers. This interdiffusion is probably at the origin of thbiquadratic coupling as supported by NMR analysis,agreement with Slonczewski’s loose spin model.10 He at-tributes this effect to the magnetic atoms at the interfacwhich are weakly exchange coupled to the rest of the femagnetic parts and give rise to a 90° coupling.

C. TEM measurements

Cross section and plan view TEM measurements wperformed on both @Co3.5nm /Ir0.5nm#103 multilayer andCo3nm /Ir0.5nm /Co3nm sandwich. In this way the local structure, the grain size, and the quality of the interfaces wdirectly investigated for both samples.

Figure 8 shows a cross section image of the two sampThe contrast is strong between Co and Ir, mainly due tolarge difference between their atomic numbers. In contrCo, Cu, Fe, and Cr have closer atomic numbers and cabe well distinguished and therefore the capping andbuffer appear without contrast. However, the period ofmultilayer can be precisely measured, giving the valueabout 3.860.2 nm. This is in good agreement with the nom

uh

tect

hatns.no

-rnsandare

nsthethee

al-nedichandre

theal.

es

o/ss

th

,ular

PRB 62 11 715MAGNETIC, TRANSPORT, AND STRUCTURAL . . .

nal value~4 nm!. The Ir layers appear to be continuous btheir thickness is larger than the nominal value 0.5 nm. Tis due to the strong interdiffusion at the Co/Ir interfaces26 asshown by NMR interface profile~Fig. 7!. From the contrastanalysis between Co and Ir, the Ir layers and the Co/Ir infaces do not present lateral oscillations and the interfaroughness do not increase with increasing the number of

FIG. 8. High-resolution bright-field cross-section TEM imagof a @Co3.5nm /Ir0.5nm#103 multilayer ~a! and ~c!, and of aCo3nm /Ir0.5nm /Co3nm sandwich~b!. The Ir layers~dark bands! ap-pear thicker than 0.5 nm, due to the large interdiffusion at the Cinterfaces. The inset of~c! presents a diffraction image of the crosection. The most intense spots come from the~111! diffractionplanes and correspond to a direction which make 45° withgrowth direction.

FIG. 9. Diffraction image of a@Co3.5nm /Ir0.5nm#103 multilayer~a! and a Co3nm /Ir0.5nm /Co3nm sandwich~b!. The small black circlein the region of the most intense diffraction ring~a! indicates theregion, where the dark field images were taken. The Co* indicatesthe fcc phase of Co.

tis

r-eshe

periods. This is in agreement with the NMR observation tthe coupling is homogeneous and free of lateral fluctuatio

A magnification of a high-resolution cross sectio~XHREM! image of the multilayer shows that there is ncoherence between the grains structure along theZ axis. Theaverage coherence length along theZ axis is small and doesnot exceed 8 nm~maximum two periods!. A selected areadiffraction pattern~SAD pattern! was also taken in this region and presented in the inset of Fig. 8. The SAD pattepresent two types of spots, coming from the superlatticethe atomic planes. The spots along the growth directionextremely weak and correspond to the~220! diffractionplanes. Dark field~DF! images taken under these conditiogive a similar size of grains as the values extracted fromXHREM images. The most intense spots come from~111! planes with the diffraction vector lying at 45° from thgrowth direction.

The plan view diffraction patterns~Fig. 9! present a seriesof continuous rings, which are characteristic of polycrystline samples. This evidences the presence of a well deficrystallographic structure in the grains for both the sandwand the multilayer. These rings are mainly due to the hcpfcc Co phases. However, the diffraction rings of the Ir aonly observed in the case of the multilayer, mainly due tolarger thickness of Ir, giving rise to a larger diffracted sign

Ir

e

FIG. 10. Dark-field images of a@Co3.5nm /Ir0.5nm#103 multilayer~a! and Co3nm /Ir0.5nm /Co3nm sandwich~b!. The grains are smallrandomly oriented and superposed, which explains the irregshape of the borders and their diffused contrast.

11 716 PRB 62S. COLISet al.

FIG. 11. SQUID magnetization loops on@Co3.5nm /Ir0.5nm#103 multilayer recorded at 5, 70, 140, 240, and 295 K@~a!–~e!#. ~f! presents thecurve from~a! without the buffer contribution~solid line! and the simulation curve with a trilayer model~dotted line!.

nsnereg

ect,

nefpeusr

owanturti

een

e ofby

and

eldhis

en-ent

nottheex-isac-ndFe

ion,ion

The dark field images~Fig. 10! were taken in the regionof the most intense ring of the plan view diffraction patter~Fig. 9!. This ring is actually the result of the superpositioof many rings coming from fcc and hcp Co, Cr, Ir, and Fwhich means that the grains presented in Fig. 10 will corspond to all these materials. The observed grains with relar borders are small, with an average lateral~in-plane! sizeof about 10 nm, and similar for both samples, in agreemwith the cross-section observation. This small size is direrelated to our deposition technique.27 On the other handmost of the white spots of the images are coming fromsuperposition of grains with almost parallel diffractioplanes. This superposition is possible since the small sizthe grains along theZ axis is largely below the thickness othe whole stack, and explains very well the irregular shaof the white spots borders. These spots are more numerothe case of the multilayer because of the larger numbegrains along theZ direction.

Besides the NMR technique, this is another way to shby direct imaging, that the structural properties and the quity of the interfaces are identical for both the sandwich athe multilayer. As a consequence, the impact of the strucand of the interfaces on the magnetic and transport propewill be the same.

,-u-

ntly

a

of

sin

of

,l-drees

V. LOW-TEMPERATURE MAGNETIZATIONMEASUREMENTS AND SIMULATIONS

Temperature-dependent SQUID magnetometry have bperformed on the same@Co3.5nm /Ir0.5nm#103 multilayer, inorder to give another experimental support to the presencthe biquadratic coupling at low temperature as shownNMR. Measurements were carried out at 5, 70, 140, 240,293 K and presented in Figs. 11~a!–11~e!.

The first interesting observation is that the saturation fiincreases strongly with decreasing the temperature. Tmeans that the indirect exchange coupling is largelyhanced at low temperature. This result is also in agreemwith Slonczewski’s loose spin model,10 predicting a strongtemperature dependence of the 90° coupling, which canbe easily explained by the fluctuation mechanism. Onother hand, the observed large remanent magnetizationcludes the possibility that the indirect exchange couplingantiferromagnetic. The remanence can be, however,counted for by considering a pure biquadratic coupling, aby tacking into account the bulk values of the Co andmagnetic moments at 5 K~Ref. 28! for magnetization calcu-lation. This leads to values of 45 and 60 emu/m2, respec-tively, for the remanence and the saturation magnetizatwhich are in very good agreement with the magnetizat

ththon

ses

sio

oungurtheerotc

yo

put-a-wn

s is

er-ghehaveandup-and

ntsar-

e’’

PRB 62 11 717MAGNETIC, TRANSPORT, AND STRUCTURAL . . .

curve recorded at 5 K. To have a more precise idea onstrength of the exchange coupling, we tried to simulatemagnetization curve after removing the magnetic buffer ctribution, and using a trilayer model@Fig. 11~f!#. The bestresult was obtained using a ratioJbiq /Jf erro52.3 (Jbiq521.55 erg/cm2 andJf erro50.55 erg/cm2). However, sig-nificant differences are evidenced between the roundnesthe theoretical and the experimental loop. This can beplained either by the interdiffusion at the Co/Ir interfacewhich leads to a nonperfect biquadratic coupling, or bymagnetic domain structure. Since in the structural analythe NMR measurements have clearly evidenced that the cpling is homogeneous, the first hypothesis can be ruledHowever, considering small fluctuations of the coupliand/or fluctuations of the local anisotropy axis, the curvatof the magnetization loop is very well reproduced nearsaturation field. In contrast, the differences between theperimental curve and simulation are still important, near zfield. In these conditions, the second explanation seems tmore favorable. Indeed, recently Tiusan29 demonstrated thamagnetic walls pinned by structure and/or magnetic defecan be at the origin of this curvature~from zero field up tosaturation! of the magnetization loop, and this is very likelto occur in our case since all the samples present a nnegligible amount of structural defects.

ee-

ofx-,as,u-t.

eex-obe

ts

n-

VI. CONCLUSION

This study has clearly demonstrated that ion-beam stering is suitable to grow Co/Ir/Co sandwiches with biqudratic exchange coupling. TEM observations have shothat the texture is~111!, having an angle of 45° with thegrowth direction, and that the average size of the grainsmall particularly along theZ axis. Moreover, NMR andTEM have shown that the Co/Ir interfaces are strongly intdiffused. This interdiffusion, which limits the couplinstrength and the GMR ratio, is however at the origin of tmain results observed in these sandwiches. Indeed, weshown that the indirect exchange coupling is biquadratichomogeneous on the whole film surface. This result is sported by the NMR measurements, the GMR curves,low-temperature magnetization measurements.

ACKNOWLEDGMENTS

The authors thank R. Poinsot for SQUID measuremeand T. Romero for technical assistance. This work was ptially supported by the European Community ‘‘TunelSensproject ~Grant No. BRPR-CT98-0657!.

.d-

c

r

*Email address: colis@ipcms.u-strasbg.fr1P. Grunberg, R. Schreiber, Y. Pang, M.B. Brodsky, and H. Sow

ers, Phys. Rev. Lett.57, 2442~1986!.2M.N. Baibich, J. Broto, A. Fert, F. Nguyen Van Dau, F. Petrof

P. Etienne, G. Creuzet, A. Friedrich, and J. Chazelas, Phys. RLett. 61, 2472~1988!.

3H.A.M. van den Berg, W. Clemens, A. Gieres, G. Rupp, MVieth, J. Wecker, and S. Zoll, J. Magn. Magn. Mater.165, 524~1997!.

4S. Zoll, A. Dinia, M. Gester, D. Stoeffler, H.A.M. van den BergA. Herr, R. Poinsot, and H. Rakoto, J. Magn. Magn. Mater.165,442 ~1997!.

5A. Dinia, M. Stoeffel, K. Rahmouni, D. Stoeffler, and H.A.Mvan den Berg, Europhys. Lett.42, 331 ~1998!.

6Y. Luo, M. Moske, and K. Samwer, Europhys. Lett.42, 565~1998!.

7H. Yanagihara, K. Petit, M.B. Salamon, E. Kita, and S.S.P. Pkin, J. Appl. Phys.81, 5197~1997!.

8H. Yanagihara, E. Kita, and M.B. Salamon, Phys. Rev. B60,12 957~1999!.

9C. Chesman, M.A. Lucerna, M.C. de Moura, A. Azevedo, F.Mde Aguiar, S.M. Rezende, and S.S.P. Parkin, Phys. Rev. B58,101 ~1998!.

10J.C. Slonczewski, J. Appl. Phys.73, 5957~1993!.11J.C. Slonczewski, Phys. Rev. Lett.67, 3172~1991!.12K. Petit, S. Gider, S.S.P. Parkin, and M.B. Salamon, Phys. Rev

56, 7819~1997!.13S. Colis, A. Dinia, D. Deck, G. Schmerber, and V. Da Costa,

Appl. Phys.88, 1552~2000!.14H.R. Kaufmann, J.J. Cuomo, and J.M.E. Harper, J. Vac. S

Technol.21, 725 ~1982!.15P.J. Flanders, J. Appl. Phys.63, 3940~1988!.16B. Dieny, J. Magn. Magn. Mater.136, 335 ~1994!.

-

f,ev.

.

,

.

ar-

.

. B

J.

ci.

17D. Stoeffler and F. Gautier, J. Magn. Magn. Mater.140, 529~1995!.

18S. Zoll, A. Dinia, J.P. Jay, C. Me´ny, G.Z. Pan, A. Michel, L. ElChantal, V. Pierron-Bohnes, and P. Panissod, Phys. Rev. B57,4842 ~1998!.

19P. Panissod, inFrontiers in Magnetism of Reduced DimensionSystems, Vol. 49 of NATO Advanced Study Institute, Series 3:High Technology, edited by V.G. Bar’yakhtar, P.E. Wigen, andN.A. Lesnik ~Kluwer Academic, Dordrecht, 1998!.

20P. Panissod, C. Me´ny, M. Wojcik, and E. Jedryka, inMagneticUltrathin Films, Multilayers and Surfaces—1997, edited by J.Tobin, D. Chambliss, D. Rubinski, K. Barmak, P. Dederichs, Wde Jonge, T. Katayama, and A. Schuhl, MRS Symposia Proceeings No. 475~Materials Research Society, Pittsburgh, 1997!,pp. 157–168.

21The scaled RF field is the value (HHF /H1op)•H1, whereH1 isthe actual RF field,HHF is the hyperfine field, andH1op is theRF field strength for the maximum signal in a nonmagnetisample~a calibrated instrumental constant!. In a magnetic mate-rial the RF field acting on the nuclei is enhanced by the factoHHF /Hres , whereHres is the restoring field. Consequently themaximum signal is reached for a valueH1o f of the external RFfield such that (HHF /Hres)•H1o f5H1op . The restoring field isthen measured byHres5(HHF /H1op)•H1o f , i.e., the value ofthe scaled RF field for the maximum signal~see Ref. 20!.

22P. Panissod, J.P. Jay, C. Me´ny, M. Wojcik, and E. Jedryka, inMagnetic Ultrathin Films, Multilayers and Surfaces, edited byA. Fert, H. Fujimori, G. Guntherodt, B. Heinrich, W. F.Egelhoff, Jr., E.E. Marinero, and R.L. White, MRS SymposiaProceedings No. 384~Materials Research Society, Pittsburgh,1995!, pp. 61–72.

23P. Panissod, J.P. Jay, C. Me´ny, M. Wojcik, and E. Jedryka, Hy-perfine Interact.97Õ98, 75 ~1996!.

ur

t

ticevi-

arethe

n

11 718 PRB 62S. COLISet al.

24Y. Henry, C. Meny, A. Dinia, and P. Panissod, Phys. Rev. B47,15 037~1993!.

25N. Persat, Ph.D. thesis, Louis Pasteur University, Strasbo1998.

26T.B. Massalski,Binary Alloy Phase Diagram, 2nd ed.~AmericanSociety of Metals, Metals Park, OH, 1990!, Vol. 2.

27A. Dinia, N. Persat, and C. Ulhaq-Bouillet~unpublished!.28According to many authors, the values of the magnetic momen

5 K are around 1440 emu/cm3 for Co and, respectively, 1740emu/cm3 for Fe. These values were taken from C. Kittel,Phy-

g,

at

sique de l’etat solide, 5th ed., edited by Bordas~Dunod Univer-site, Paris, 1983!, p. 465. Notice that differences of the magnemoments between the bulk and the thin layer material weredenced. In our calculations we supposed also that interfacesabrupt and that there is no loss of magnetic moments atinterfaces.

29C. Tiusan, T. Dimopoulos, K. Ounadjela, M. Hehn, H.A.M. vaden Berg, V. Da Costa, and Y. Henry, Phys. Rev. B61, 580~2000!.