HIGHER physics

Exchange bias of the interface spin system at the Fe/MgO interface Y. Fan1†, K. J. Smith1, G. Lüpke1*, A. T. Hanbicki2, R. Goswami2, C. H. Li2, H. B. Zhao3 and B. T. Jonker2*

The ferromagnet/oxide interface is key to developing emerging multiferroic and spintronic technologies with new functionality. Here we probe the Fe/MgO interface magnetization, and identify a new exchange bias phenomenon manifested only in the interface spin system, and not in the bulk. The interface magnetization exhibits a pronounced exchange bias, and the hysteresis loop is shifted entirely to one side of the zero field axis. However, the bulk magnetization does not, in marked contrast to typical systems where exchange bias is manifested in the net magnetization. This reveals the existence of an antiferromagnetic exchange pinning layer at the interface, identified here as FeO patches that exist even for a nominally ‘clean’ interface. These results demonstrate that atomic moments at the interface are non-collinear with the bulk magnetization, and therefore may affect the net anisotropy or serve as spin scattering sites. We control the exchange bias magnitude by varying the interface oxygen concentration and Fe–O bonding.

T he characteristics of the ferromagnet/oxide interface deter- mine multiferroic, electronic, transport and magnetic proper- ties, providing new functionality and leading to a wide range

of emerging device technologies1–6. Magnetic tunnel junctions (MTJs) incorporating such interfaces are critical elements in magnetic memory and disk drives, and show promise for much broader application in reprogrammable architectures7–11. The ever-increasing demand for non-volatile memory, ‘instant-on’ electronics and reprogrammable logic coupled with low power con- sumption has led to the rapid development of such junctions7,8.

MTJs typically incorporate MgO as an insulating layer between two Fe or Fe-based contacts (Fig. 1a)9,10, and discrete exchange bias layers (for example, IrMn, FeMn, not shown) are used to pin the magnetization of one of these Fe layers in a given direction to control the magnetic field response. Fe/MgO/Fe-based MTJs are predicted to exhibit very large tunnel magnetoresistance (TMR) (.1,000%) because of their matching band symmetries12. Although a large change in resistance DR/R ¼ (RAP2RP)/RP ≈ 200% has been observed experimentally as the magnetization of the two Fe electrodes is changed from parallel (P) to antiparallel (AP)9,10, this value is much lower than predicted, even after much develop- ment to optimize the structure.

This discrepancy has been attributed to several factors13, including defects and traps within the MgO, structural defects and disorder, and to the intermixing observed experimentally at the Fe/MgO interface, which alters these band symmetries (oxygen randomly diffuses into the two Fe layers at the interface to form FeOx , Fig. 1b

14–18). Oh et al. note that the formation of FeO cannot be avoided, and suggest that FeO and MgO coexist at the interface in an entropically stabilized phase19. These earlier studies focused on the chemistry, electronic and atomic structure of the Fe/MgO interface, and did not address the interface spin orientation. This spin orientation sig- nificantly affects spin transport, as well as the magnetic and multifer- roic properties of ferromagnet/oxide heterostructures.

Here, we use magnetization-induced second harmonic generation (MSHG) to selectively probe the magnetization at the Fe/MgO

interface, and discover an exchange bias not previously observed that is also markedly different from its typical manifestation (Fig. 2). We observe a pronounced shift in the interface magnetization hysteresis loop from zero field, the classic signature of exchange bias20,21. Such a shift is not observed in the ‘bulk’ magnetization (that is, the net magnetization of the Fe film), which we measured using standard magnetometry and the magneto-optic Kerr effect (MOKE), in striking contrast to studies of exchange bias systems studied to date. This signals the presence of an antiferromagnetic (AF) exchange pinning layer at the interface, as illustrated by the model we propose in Fig. 1b, and shows that the magnetic moments at the interface are not parallel to the net magnetization of the Fe layer as the magnetization is switched, contrary to expectation. We control the magnitude of the exchange bias field by varying the interface oxygen concentration, confirming that this effect is induced by Fe–O bonding and compound formation, and is likely to be present even for ‘clean’ MgO surfaces. We believe that this is another factor to be considered in understanding the TMR observed in this system, and also has implications for developing electric field control of magnetic anisotropy at the Fe/MgO interface1,3,22.

Current understanding of exchange bias Exchange bias occurs at the interface between an antiferromagnet (AFM) and a ferromagnet (FM), where the hard magnetization of the AFM biases the magnetization of the softer FM20,21. Exchange bias is created by cooling the AFM/FM structure in an applied field through the Neel temperature of the AFM (the temperature at which AF order sets in). The very strong exchange coupling between the interface layers of the FM and AFM tends to pin or ‘bias’ the magnetization of the FM in a specific direction. This results in an offset of the hysteresis loop so that it is no longer centred at zero applied field, but shifted by an amount correspond- ing to the exchange bias field HE , as illustrated in Fig. 3. The exchange bias increases the magnitude of the applied magnetic field needed to reverse the magnetization of the FM from the normal coercive field Hc to Hc þ HE.

1 Department of Applied Science, College of William and Mary, 251 Jamestown Road, Williamsburg, Virginia 23187, USA, 2 Materials Science and Technology Division, Naval Research Laboratory, 4555 Overlook Avenue SW, Washington, DC 20375, USA, 3 Department of Optical Science and Engineering, Fudan University, 220 Handan Road, Shanghai 200433, China; †Present address: Seagate Technology, 1200 Disc Drive, Shakopee, Minnesota 55379, USA.

*e-mail: [email protected]; [email protected]

ARTICLES PUBLISHED ONLINE: 2 JUNE 2013 | DOI: 10.1038/NNANO.2013.94

NATURE NANOTECHNOLOGY | VOL 8 | JUNE 2013 | www.nature.com/naturenanotechnology438

© 2013 Macmillan Publishers Limited. All rights reserved.

Exchange bias is of great technological importance in tailoring the operating characteristics of most magnetic devices, including hard disk read heads, magnetic memory and magnetic sensors. However, it remains poorly understood because it is generally observed only indirectly through the response of the bulk magneti- zation, and continues to be extensively studied in a wide variety of systems23–26. Seminal work by Valev et al. used MSHG to study the length scale over which exchange bias occurred by varying the thickness of a Cu spacer layer in a CoO/Cu/Fe system27. They found that the magnetic interaction between the Fe and antiferro- magnetic CoO layer was sufficiently strong to induce order in the CoO, even at Cu spacer layer thicknesses for which there was no observable shift in the hysteresis loop.

The ferromagnet/oxide interface The Fe/MgO(001) interface can be fabricated with high structural quality because there is a small lattice mismatch between Fe and MgO (3.8%), with the in-plane Fe[100] axis rotated 458 with respect to that of the MgO (ref. 15). Submonolayer FeO formation has been observed for the growth of Fe on MgO(001) and attributed to the presence of residual or excess oxygen following growth of the MgO (ref. 15). Bulk FeO is a known AFM with a Néel temperature of 198 K, which can be enhanced to nearly 800 K if a thin FeO film is embedded into an FM matrix28. Although an exchange bias might be expected, there have been no reports of this effect occurring in the Fe/MgO(001) bilayer system, possibly because the interface FeO does not pin enough Fe interface atomic moments to generate

a b

MgO

MgO

Fe

O

Mg

[001]

[010]

[100]

Figure 1 | Model of TMR structure and of atomic moments near the Fe/MgO interface. a,b, Schematic of Fe/MgO/Fe MTJ (a) and model of the atomic

magnetic moments at the Fe/MgO interface (b) giving rise to the exchange bias layer detected with MSHG. Fe atoms with their magnetic moment shown in

blue are those coordinated and bonding with an O atom at the interface and exhibit compensated in-plane AF order leading to exchange bias. Fe moments

shown as open red arrows are exchange-biased by AF order at the interface. Fe moments shown as filled red arrows constitute the bulk magnetization

and exhibit no exchange bias. The coordinate axes refer to the Fe(001) lattice, and the in-plane magnetic easy axes of the Fe film are along [100]. The

MgO(001) in-plane axes are rotated by 458 relative to those of the Fe. The atomic structure of the interface is taken from ref. 15, although the enhanced Fe/FeO interlayer spacing and rumpling of the FeO layer are not shown.

H

a

Interface

Bulk Fe

M

MgO

MOKE

M O

K E

MSHG

M SH

G

M O

K E

M SH

G

M O

K E

M SH

G

HE = 0 Hc = 11

HE = 0 Hc = 8

HE = 0 Hc = 15

HE = −19 Hc = 11

HE = −8 Hc = 9

HE = 0 Hc = 13

H (Oe)

0 25−25

H (Oe)

0 25−25

H (Oe)

0 25−25

H (Oe)

0 25−25

H (Oe)

0 25−25

H (Oe)

High b

c

d

e

f

g

Normal Minimum

0 25−25

Figure 2 | Measurement geometry and MOKE/MSHG data. a, Schematic of the optical measurements. MOKE measures the net magnetization (bulk) of the

Fe film, and MSHG selectively probes the interface magnetization only. b, MOKE data for a 10 nm Fe/MgO(001) sample with a high density of oxygen on

the MgO surface before Fe deposition. The curve exhibits a symmetric magnetization loop. c, MSHG data from the same sample. The curve exhibits a

pronounced offset along the horizontal axis, the classic signature of exchange bias. d,e, Corresponding data for a sample prepared with no oxygen exposure

of the MgO surface, using conditions typically used to produce a clean starting surface for growth of Fe/MgO/Fe MTJs, with a normal amount of oxygen at

the interface. The MSHG curve in e exhibits a pronounced exchange bias. f,g, Corresponding data for a sample with a minimum density of oxygen on the

MgO surface, exhibiting no exchange bias. The exchange bias fields HE and coercive fields Hc are indicated in each panel. Black and red curves are taken

with the magnetic field sweeping from negative to positive, and from positive to negative, respectively. Data are acquired over a field range of +366 Oe, and no additional switching is observed beyond +50 Oe. All data are obtained at room temperature with the magnetic field applied along the in-plane easy axis, Fe[100].

NATURE NANOTECHNOLOGY DOI: 10.1038/NNANO.2013.94 ARTICLES

NATURE NANOTECHNOLOGY | VOL 8 | JUNE 2013 | www.nature.com/naturenanotechnology 439

© 2013 Macmillan Publishers Limited. All rights reserved.

a detectable exchange bias in the net magnetization of the Fe film. A direct probe of the interface magnetization is therefore needed.

The samples were formed by depositing 10-nm-thick Fe films on MgO(001) single-crystal substrates by molecular beam epitaxy at room temperature. The substrates were first rinsed with isopropyl alcohol and cleaned by heating to 773 K in ultrahigh vacuum (UHV). The initial oxygen density on the MgO surface was set to three concentrations by (i) exposure to oxygen at 300 K in an attached vacuum chamber in the presence of an ultraviolet lamp to obtain a high oxygen density, (ii) keeping the substrate in UHV but not exposed to oxygen, producing what we call here a ‘normal’ surface oxygen density (this is considered to be a clean starting surface for growth of Fe/MgO/Fe MTJs, as the MgO growth is typically done at substrate temperatures ,600 K) or (iii) further heating the substrate to 973 K in UHV, thereby desorbing the surface oxygen29 to achieve a minimum oxygen density. Heating to such high temperatures is generally not performed during growth of MTJs to minimize potential interdiffusion of other layers in the structure.

After growth, all Fe/MgO samples were exposed to air to oxidize the Fe surface, suppressing surface contributions to the MSHG signal30, and then field annealed along the in-plane Fe[100] easy axis. High-resolution transmission electron microscopy (TEM) images of each type of sample are shown in Fig. 4, and confirm the high quality of the Fe/MgO interface in each case. The interface quality and structure are comparable to those reported for thin-film Fe/MgO/Fe(001) samples, which exhibit high TMR values9,10. For each sample type, the (200) MgO and (110) Fe lattice fringes are continuous across the Fe/MgO interface. The interface exhibits a defect structure consistent with the 3.8% lattice mismatch and monolayer step fluctuations on the MgO substrate surface. We note that any FeO formation is difficult to distinguish because, in addition to the local strain and defects noted above, MgO and FeO have the same rocksalt structure and similar lattice parameters (0.421 and 0.433 nm, respectively).

Previous work has shown that although TEM has been unable to detect the structural presence of FeO at Fe/MgO epitaxial interfaces, Fe core level shifts due to chemical bonding clearly demonstrated FeO formation17. We therefore use electron energy loss spectroscopy (EELS) within the TEM to probe the chemical composition of the interface, and these data unambiguously show the presence of FeO for the sample with high oxygen interface density. Although the EELS L-edge fine structure of Fe and several of its oxides are very similar, FeO can be readily distinguished because there is a chemical shift of the L3 (2p3/2) threshold with respect to the L2 (2p1/2) edge

31. As a result, the energy separation of the L3 and L2

‘white line’ peaks is smaller for FeO (12.9 eV) than for Fe (13.2 eV) or its other oxides (also 13.2 eV, see table III of ref. 31). EELS spectra from the interface of our normal and high interface oxygen density samples are shown in Fig. 4d (spectra are normalized to the L3 peak and offset for ease of comparison). Upon visual inspection, it is apparent that the energy separation of the L3 and L2 peaks is noticeably smaller for our high oxygen interface sample than for our normal sample, indicating the presence of

Hc

M

0 H

HE

AFM

(i)

(i)(ii) (ii)

(iii)

(iii)

(iv)

(iv)

FM

Figure 3 | Classic model of exchange bias. Relative orientations of the

atomic moments in the AFM and FM are shown schematically, illustrating

the lateral offset in the magnetization curve20,21. The magnitude of the

exchange bias field HE and coercive field Hc are defined in the figure.

2 nm

2 nm

2 nm

a

b

c

N or

m al

iz ed

in te

ns ity

705 715 720 725 730710

Energy (eV)

L3

L2

Normal

High

d

Figure 4 | TEM images of sample interfaces. a–c, Images showing the

Fe/MgO interface region for high (a), normal (b) and minimum (c) oxygen

density on the MgO(001) surface before Fe deposition. The markers indicate

the position of the Fe/MgO interface, with the Fe above and the MgO

below. The interface exhibits a defect structure consistent with the 3.8%

lattice mismatch and monolayer step fluctuations on the MgO substrate

surface. MgO and FeO have the same structure and similar lattice constants,

making them difficult to distinguish. d, Fe L2,3 EELS spectra from normal and

high interface oxygen density samples, normalized to the L3 peak and offset

for clarity. The solid lines are fits to the data using the procedure of ref. 31.

The black vertical lines indicate the peak positions derived from the fits, and

show that the L32L2 peak separation is smaller for the high interface oxygen

sample, indicating the presence of FeO.

ARTICLES NATURE NANOTECHNOLOGY DOI: 10.1038/NNANO.2013.94

NATURE NANOTECHNOLOGY | VOL 8 | JUNE 2013 | www.nature.com/naturenanotechnology440

© 2013 Macmillan Publishers Limited. All rights reserved.

FeO, which is readily detectable in the former sample. To quantify this, we fit these data using the procedure of ref. 31, and determined that the separation of the peaks in the high oxygen interface sample is 12.9+0.1 eV, and in the normal sample is 13.2+0.1 eV, verifying the presence of FeO at the high oxygen interface sample. The amount of interface FeO in the normal sample is expected to be much smaller, and may be at or below our EELS detection threshold.

Selectively probing the interface and bulk spin systems The magnetic properties of the Fe/MgO interface and Fe bulk were determined using MSHG and MOKE, respectively, as illustrated in Fig. 2a. The MSHG signal is produced only where inversion sym- metry is broken, and is therefore sensitive only to the interface mag- netization30,32–34 and not the Fe bulk. In contrast, MOKE averages the magnetization of the entire film, and provides a measure of net magnetization. Further details on sample preparation and optical measurements may be found in the Methods.

Figure 2 shows the MOKE and MSHG data obtained at room temperature with the applied field along the easy in-plane axis Fe[100] for Fe/MgO(001) samples prepared with a high interface oxygen density (Fig. 2b,c), a normal density typical of a clean MgO surface (Fig. 2d,e) and minimum interface oxygen density (Fig. 2f,g). In the MSHG data of Fig. 2c, the hysteresis loop is comple- tely shifted to the left of the zero field axis, clearly indicating pro- nounced exchange bias of the interface magnetization. The exchange bias field given by the shift of the hysteresis loop along the H-axis is HE ¼ 19+4 Oe. A smaller exchange bias of the interface spin system is observed for samples with a normal interface oxygen density (Fig. 2e). The interface exchange bias phenomenon is con- firmed by taking MSHG measurements after rotating the sample 1808 and observing that the loop is shifted from left to right. In con- trast, MOKE data from the same samples (Fig. 2b,d) exhibit no shift of the hysteresis loop. MOKE and MSHG data for the Fe/MgO sample prepared with low interface oxygen density produced by a high-temperature anneal of the MgO substrate are shown in Fig. 2f,g, respectively, and exhibit no evidence for exchange bias either in the bulk or interface magnetization.

Although the MOKE measurement in principle includes the con- tribution from the interface layer, the interface exchange bias cannot be distinguished due to the much larger contribution from the rest of the film (the ‘bulk’), and the exponential attenuation of the laser probe with distance from the surface, significantly reducing any signal from the buried interface. For a sufficiently thin Fe film com- prising approximately 2–10 monolayers, one would reasonably expect that the interface contribution would dominate. However, we have looked at such films and see no bulk exchange bias with vibrating sample magnetometry or MOKE measurements for Al-capped Fe films on MgO. We attribute this to the fact that Fe does not grow layer by layer on MgO, and thus the large surface roughness present in such thin films introduces ‘orange-peel’ mag- netostatic coupling effects (local fringe fields), which completely mask any interface exchange bias.

By comparing the MOKE and MSHG data of Fig. 2b,c, there is clearly a range of applied field on each side of the zero field axis for which the interface magnetization is antiparallel to that of the bulk. Such an abrupt reorientation of the magnetization has been observed previously at the vacuum surface of magnetic samples35–37, as discussed below. These results imply that the interface magneti- zation is not rigidly coupled to the bulk, but is more strongly coupled by exchange bias to the FeO pinning layer than it is to the bulk. Because exchange coupling is in general anisotropic, strong intralayer coupling maintains magnetic order even though the interlayer coupling may be significantly reduced at the interface due to electron redistribution.

Such anisotropic exchange has been discussed theoretically38 and observed experimentally in several thin-film systems. For example,

several groups have reported large angular deviations and weak exchange coupling between the surface and bulk magnetizations of Co35 and Fe36,37 thin films. Gruyters et al. in fact observed that the reversal of the bulk magnetization of Co films was preceded by a complete reversal of the surface magnetization, demonstrating that the two were largely decoupled35. These different behaviours for the surface and bulk magnetization were attributed to the reduced coordination and site symmetry at the surface, leading to weak interlayer coupling while preserving strong intralayer exchange. For epitaxial films of Fe(001) on AlGaAs(001), both the static and dynamic behaviour of the Fe/AlGaAs interface magnetization were observed to be distinctly different from that of the bulk, result- ing in large angular deviations between the two and attributed to a decoupling of bulk and interface spins39,40. Ferromagnetic exchange coupling originates from Heisenberg exchange of electrons. At the interface, bond formation changes the electron distribution of the interface layer(s), particularly when strongly electronegative species (for example, O and As) are involved, and may reduce the interlayer coupling to the bulk layers.

To further confirm the origin of the exchange bias, Fe/AlOx/MgO(001) reference samples were prepared that included a monolayer of aluminium oxide between the Fe and MgO. Because Al bonds so strongly to oxygen, Fe–O interaction is minimized. These samples exhibited no exchange bias in either the MSHG or MOKE data, demonstrating that Fe–O bond formation plays an essential role. To rule out possible contributions from the air- exposed Fe surface (which can in principle induce a small MSHG signal, as it also breaks inversion symmetry), we made identical measurements on 20-nm-thick Fe films on MgO prepared in the same way. Because the skin depth of 400 nm light is less than 20 nm, the thicker Fe film blocks the MSHG signal from the inter- face, so only a potential surface contribution can be detected. We see no MSHG signal and no exchange bias in these control experiments. These control experiments confirm that the exchange bias signal we observe originates from the Fe/MgO interface.

Atomic-scale model of the interface spin system The shift of the hysteresis loop observed in the MSHG data but absent in the MOKE for the samples with an oxygen-rich interface indicates that the interface magnetization, but not the bulk, is exchange-biased by FeO formation at the Fe/MgO(001) interface. Figure 1b shows an atomic view of the chemical structure of the interface derived from the literature15. The authors of ref. 15 present detailed structural information and conclude that there is an enhanced interlayer spacing between the FeO and adjacent Fe layers, and that there is rumpling within the FeO layers (these struc- tural details are not shown in Fig. 1b). There are two specific details relevant to our model: (i) the starting MgO surface is not perfectly flat and (ii) oxygen diffuses into the Fe film and bonds with the Fe, altering the local electron distribution.

We superpose on this structure our model of the interface spin structure produced by field annealing along Fe[100] as derived from the MSHG data. Oxygen atoms intermix into the first two Fe layers, randomly occupying approximately the fourfold hollow sites within the Fe(001) planes15. Fe atoms in these layers thus bond with the O atoms, effectively forming local areas of FeO. Super exchange coupling through the local Fe–O–Fe bonds, with bond angle close to 1808, orient the coupled Fe atomic moments antiparallel (shown as blue arrows in Fig. 1b), producing compen- sated in-plane AF order characteristic of FeO41. The magnetic field annealing aligns the magnetization of the interface and bulk along the [100] direction. At locations without any intermixed O atoms, the short-range Fe ferromagnetic coupling dominates (shown as red arrows in Fig. 1b). Fe moments shown as open red arrows are exchange biased by the FeO patches at the interface. Fe moments shown as solid red arrows constitute the bulk

NATURE NANOTECHNOLOGY DOI: 10.1038/NNANO.2013.94 ARTICLES

NATURE NANOTECHNOLOGY | VOL 8 | JUNE 2013 | www.nature.com/naturenanotechnology 441

© 2013 Macmillan Publishers Limited. All rights reserved.

magnetization and exhibit no exchange bias. The magnetic moments associated with Fe–O bonds (blue arrows) are nominally co-linear (parallel or antiparallel) with the ferromagnetic magneti- zation (red arrows), although they may be canted slightly42 due to the lattice distortion produced by the AF order (FeO changes from cubic to rhombohedral below the Neel temperature) and magnetostriction43.

According to the model proposed by Meiklejohn and Bean44,45, the effective unidirectional magnetic energy is given by s¼ HEMFetFe. For our high oxygen anion density interface, we find s ≈ 0.001 erg cm22, where HE ¼ 19 Oe, MFe ¼ 1.7 × 10

3 e.m.u. cm23

is the Fe magnetization, and tFe ¼ 0.3 nm (approximately two monolayers) is the thickness of the exchange-biased Fe interface layer probed by MSHG14. For fully compensated AF spins at the Fe/FeO(001) interface, s ≈ 0.05 erg cm22 (ref. 46). Following the extended Meiklejohn and Bean model proposed by Ohldag et al.47, we attribute the decrease in s to the reduction of pinned AF spins. Hence, the percentage of pinned interfacial AF spins in our Fe/MgO(001) samples is �2%. This is consistent with the result reported by Ohldag et al. that only a small fraction (�4%) of pinned AF spins at the interface are responsible for producing an exchange bias in a 2-nm-thick Co layer47.

An estimate of the interlayer exchange coupling between the FeO and interface Fe layer in our samples is given by J ¼sb2/(S1 S2), where b ¼ 0.43 nm is the FeO lattice constant, and S1 ¼ 1 and S2 ¼ 2 are the atomic spin vectors of Fe and FeO, respectively. From this we obtain J ≈ 9.4 × 10219 erg, about two orders of mag- nitude smaller than the value reported for the Fe/FeO(001) inter- face with a full layer of AF spins46.

Controlling the interface exchange bias We control the magnitude of the exchange bias by varying the inter- face oxygen anion density through the oxygen exposure and growth temperatures used to fabricate the different Fe/MgO(001) samples, as described previously. We find that a higher oxygen concentration generates a larger exchange bias of the interface magnetization. Figure 5 plots the strength of the interface exchange bias field HE versus the direction of the in-plane applied magnetic field used for the MSHG measurement relative to the Fe crystal axes for a

variety of concentrations. The filled squares, triangles and circles represent the exchange bias magnitude for interfaces with high, normal and low oxygen anion densities, respectively, for samples grown at room temperature. Open triangles represent data for an interface grown at 473 K with normal oxygen anion density. All samples were first field annealed along [100]. It is important to note that the exchange bias is clearly present even for our normal oxygen anion density sample (MgO heated to 773 K with no oxygen exposure), which is considered a typical clean surface for subsequent deposition of Fe.

The grey curve represents the unidirectional anisotropy for which the magnitude is proportional to |cos u|, where u is the angle between the applied field and the [100] crystallographic axis. The magnitude of the exchange bias has maxima in [100] and [2100], minima in the orthogonal direction, and intermediate values along the k110l axes, consistent with the |cos u| model.

Along [100], the exchange bias field HE of samples grown at room temperature exhibits a monotonic increase with increasing interface oxygen anion density (the strength increases from 0 Oe for the low oxygen density sample to 8 Oe for the normal and 19 Oe for the high-density sample). This behaviour demonstrates that the degree of Fe–O interaction plays a critical role in the inter- face exchange bias. The interface grown at 473 K with normal oxygen anion density shows a slightly smaller exchange bias strength than the corresponding sample grown at room

20

[010]

[100]

10

0

Ex ch

an ge

b ia

s (O

e)

0

10

20

Figure 5 | Magnitude of exchange bias field HE versus applied field

direction. The coordinate axes are those of the Fe(001) film. Note that the

inner circular coordinate contour is HE ¼ 0 to more clearly display the data

points. Symbols represent the measured absolute value of the exchange bias

strength for the interface with high (red filled squares), normal (blue filled

triangles) and low (black filled circles) oxygen anion densities, grown at

room temperature. The open blue triangles represent results for the sample

grown at 473 K with normal oxygen anion density. Error bar of the exchange

bias magnitude, 4 Oe. Grey curve depicts the |cos u| dependence of the exchange bias magnitude.

a

b 5 mW

HE = −19 Hc = 11

HE = −19 Hc = 12

HE = −11 Hc = 12

HE = −11 Hc = 12

HE = −4 Hc = 12

HE = 0 Hc = 7

8 mW

0 25−25 0

H (Oe)H (Oe) H (Oe)

25−25 0 25−25

0 25−25 0

H (Oe)H (Oe)

M SH

G M

SH G

M SH

G M

SH G

M SH

G M

SH G

H (Oe)

25−25 0 25−25

9 mW 10 mW

6 mW

4

0

10

20

6 8 Laser power (mW)

315 324 334

Temperature (K)

343

TIB

Fi el

d m

ag ni

tu de

( O

e)

10

7 mW

Figure 6 | Temperature dependence of the exchange bias. a,b, Exchange

bias field HE (a, black) and coercive field Hc (a, red) from an Fe/MgO(001)

sample grown at room temperature with high interface oxygen density as a

function of temperature (a), as determined from the laser power

dependence of the MSHG hysteresis loops (b). The blocking temperature is

found to be 343 K. Error bars in a are determined by the field increment and

number of sweeps used in the magnetic field scans.

ARTICLES NATURE NANOTECHNOLOGY DOI: 10.1038/NNANO.2013.94

NATURE NANOTECHNOLOGY | VOL 8 | JUNE 2013 | www.nature.com/naturenanotechnology442

© 2013 Macmillan Publishers Limited. All rights reserved.

temperature. EELS reveals that a higher growth temperature sup- presses the formation of FeO at the interface19. Thus, the lower value of HE is consistent with reduced formation of FeO. Both the initial oxygen exposure of the MgO(001) surface and the growth temperature control the magnitude of the interface exchange bias by changing the density of local FeO formation and corresponding AF exchange pinning sites at the Fe/MgO interface.

Figure 6a shows the temperature dependence of the interface exchange bias field HE and coercive field Hc as determined from the MSHG hysteresis loops taken at different incident laser powers (Fig. 6b). The local temperature is determined from the laser power using a steady-state heat rate equation (see Methods). Samples were also field-annealed in a conventional manner using a heater stage, confirming the temperatures achieved with laser annealing. The magnitude of the exchange bias decreases with increasing temperature (increasing power of the incident laser beam) and vanishes at 343 K (10 mW), which we identify as the interface blocking temperature TIB to distinguish it from the con- ventional blocking temperature determined from the exchange bias exhibited by the bulk magnetization (see Methods). The coer- cive field exhibits little temperature dependence up to 343 K, where it decreases from 12 to 7 Oe as the interface magnetization is no longer pinned by the exchange bias. The blocking temperature of the interface magnetization is surprisingly high compared to typical values for bulk exchange bias systems. We do not understand this fully, but believe that several factors may contribute: (i) the magnetostriction plus the lattice mismatch at the interface enhances the AF uniaxial anisotropy energy, which is known to increase TB (ref. 48); (ii) the net magnetization of the interface system is much smaller than that of the bulk, so it remains pinned to higher temperatures by a given AFM/FM exchange coupling inter- action; and (iii) the FeO formed at the interface is likely to be highly defected and the Neel temperature of FeO is known to increase with increasing defect concentration49.

To our knowledge, no evidence for this interface exchange bias has been observed as a field offset in TMR measurements in simple Fe/MgO/Fe thin-film MTJs. We believe this is because MgO films grown by vapour deposition are known to have signifi- cant Mg and O vacancies10,50. Such disorder is likely to broaden the switching field distribution, disrupt or weaken the local interface exchange bias, or even destroy the long-range exchange bias effect we observe. Differences in the stoichiometry, defect structure and morphology of MgO films used in MTJs compared to the bulk single-crystal substrates we use may be of equal or greater impor- tance in determining the TMR. Nevertheless, the interface AF spins associated with FeO patches are present. Although the exchange bias effect may be very small, these AF pinning centres alter the interface spin density of states22, induce spin scattering7, and should be incorporated in the theoretical calculations to accu- rately predict the TMR effect. We have further shown that eliminat- ing excess oxygen from the MgO surface before Fe deposition eliminates any measureable exchange bias of the interface spin system, which we attribute to minimizing the formation of Fe–O AF pinning sites.

Conclusions In conclusion, we have studied the interface spin system of the Fe/MgO(001) bilayer system with MSHG. Previous studies addressed the chemistry of this interface and related it to the band symmetries impacting spin transport14–18, but here we have selec- tively probed the interface magnetization, showed that it exhibits a pronounced exchange bias not detectable in the bulk, thus signalling the presence of an AF interface pinning layer, and derived a model for the spin structure. We demonstrate that this exchange bias orig- inates from the formation of local patches of AF FeO at the interface, produced by super exchange coupling via Fe–O–Fe bonds. These

results are relevant to any ferromagnet/oxide interface in which the elemental constituents are likely to form an antiferromagnetic oxide (for example, FeO, CoO, MnO, NiO), as is the case in many magnetic or multiferroic oxide heterostructures.

Methods For sample preparation we began with MgO(001) bulk single-crystal wafers, which were precleaned with isopropyl alcohol and annealed in situ to 773 K. The oxygen anion density on the surface was altered by treating the substrate in one of three ways: (i) to create a high density, the substrate was oxidized in 200 T partial pressure of oxygen gas at room temperature for 15 min with ultraviolet light from a Hg lamp; (ii) to generate a normal density, the surface was not further treated (this is considered to be a clean starting surface for growth of Fe/MgO/Fe MTJs); (iii) to generate a minimum density, we heated the substrate further to 973 K to desorb any excess oxygen29. Heating to such high temperatures is generally not performed during the growth of MTJs. A 10 nm film of Fe was then deposited using molecular-beam epitaxy either at room temperature or 473 K, and at a growth rate of 0.25 nm min21.

All samples were exposed to air, which oxidized the Fe surface, reducing the free electron density and suppressing surface MSHG30. We field-annealed the samples with a 12 mW pulsed laser beam focused to a spot diameter of 1 mm (Ti:sapphire amplifier system, 800 nm wavelength, 150 fs pulse duration, 1 kHz repetition rate). The average temperature of the annealing spot was 353 K, as described in the following. We then air-cooled the sample to room temperature with a magnetic field of 747 Oe applied along the Fe magnetic easy axis [100]. Samples were also field- annealed in a conventional manner using a heater stage rather than laser-annealed, for comparison. This annealing was performed at 373 K for 15 min in a 750 Oe field applied along the Fe in-plane [100] axis, and the samples were then cooled in the field. The MOKE and MSHG data obtained were similar to those from the samples that were only laser-annealed, confirming the temperatures achieved with laser annealing determined from the model described in the following.

After field annealing, we used the laser beam (reduced to 5 mW) to measure the magnetic properties in two complementary ways, the MSHG technique measuring the magnetic response of the Fe/MgO(001) interface, and MOKE measuring the bulk-averaged magnetization. Data were acquired over a magnetic field range of +366 Oe. All magnetic field and temperature scans were completely reversible. For the case of longitudinal MSHG, we irradiated the sample with s-polarized light and detected the reflected MSHG signal (400 nm wavelength) with a photomultiplier tube, placed after a prism and an analyser, which was set to 68 from s-polarization to measure the interface magnetization. In the longitudinal MOKE studies, we measured the bulk magnetization by irradiating the sample with p-polarized light and detecting the s-component of the reflected light with a photodiode. As a control to isolate the magnetic properties of the air-exposed Fe surface, we grew a 20-nm-thick Fe film on MgO(001) at room temperature; the incident laser light used for the MSHG measurements could not penetrate this film, so any signal thus measured came from the Fe surface. We observed no MSHG signal from the oxidized Fe surface.

We measured the temperature dependence of the magnitude of the exchange bias by increasing the power of the laser beam, raising the temperature of the spot under the laser beam. Temperature T satisfies the steady-state heat equation 2K(∂2T/∂x2 þ∂2T/∂y2) ¼ C/h(Text2T ) þ Q, where K ¼ 76.2 W m

21 K21 is the thermal conductivity of Fe, x and y are the coordinates of the sample surface, C ¼ 10 W m22 K21 is the convective heat transfer coefficient of air, h ¼ 10 nm is the Fe film thickness, Text ¼ 296 K is the air temperature, and Q is the heat source term of laser. Because the laser has a Gaussian spatial distribution of intensity, Q ¼ I(12R)/0.682 × exp[22(x2 þ y2)/r2]/(pr2h), where I is the laser power, R ¼ 87% is the Fe reflectivity at the wavelength of 800 nm, and r ¼ 0.5 mm is the radius of the laser beam. We set the temperature at the boundaries of the sample (1 × 1 cm2) to be 296 K, and calculated the temperature distribution T(x, y) numerically using the finite-element method. The average temperature within the laser beam was obtained by dividing the integral of the temperature distribution by the beam area.

We note that the electrons experience much higher non-equilibrium temperature than the steady-state value. Within the first few picoseconds after the laser pulse, electrons are excited to a higher temperature before transferring energy to the lattice. This process can be described as

n ∫T2

T1

Ce(T)dT = I(1 − R)

where n ¼pr2hr/mA is the amount of Fe (unit, mol) with r¼ 7.874 × 10 3 kg m23

and mA ¼ 56 g mol 21, Ce(T ) ¼ 4.98T mJ mol

21 K21 is the Fe electron thermal capacity, T2 is the maximum electron temperature, and T1 is the average steady-state electron temperature as determined in the preceding paragraph. So T2 ¼ [2I(12R)/4.98n þ T 1

2]1/2. For the laser power of 10 mW, T1 ¼ 343 K and T2 ¼ 768 K. We used the steady-state temperature in our discussion, because the MSHG processes we measure occur on a timescale much shorter than the time required to raise the electron temperature after the incident laser pulse.

NATURE NANOTECHNOLOGY DOI: 10.1038/NNANO.2013.94 ARTICLES

NATURE NANOTECHNOLOGY | VOL 8 | JUNE 2013 | www.nature.com/naturenanotechnology 443

© 2013 Macmillan Publishers Limited. All rights reserved.

The blocking temperature TB is typically determined from the temperature dependence of the exchange bias field (hysteresis loop shift) of the bulk magnetization. As we do not observe any exchange bias phenomenon in the bulk, but only at the interface, we do not involve TB in the discussion to avoid misunderstanding and instead refer to the interface blocking temperature TIB. We note that we cannot directly measure the temperature at which the AF order in the FeO disappears, called the Neel temperature TN , and therefore cannot distinguish between TIB and TN from our data. However, TB is typically less than TN in the systems studied to date.

Received 28 October 2011; accepted 25 April 2013; published online 2 June 2013

References 1. Maruyama, T. et al. Large voltage-induced magnetic anisotropy change in a

few atomic layers of iron. Nature Nanotech. 4, 158–161 (2009). 2. Bibes, M., Villegas, J. E. & Barthelemy, A. Ultrathin oxide films and interfaces

for electronics and spintronics. Adv. Phys. 60, 5–84 (2011). 3. Ikeda, S. et al. A perpendicular anisotropy CoFeB–MgO magnetic tunnel

junction. Nature Mater. 9, 721–724 (2010). 4. Garcia, V. et al. Ferroelectric control of spin polarization. Science 327,

1106–1110 (2010). 5. Li, C. H., van’t Erve, O. M. J. & Jonker, B. T. Electrical injection and detection

of spin accumulation in silicon at 500 K with magnetic metal/silicon dioxide contacts. Nature Commun. 2, 245 (2011).

6. Chu, Y-H. et al. Electric-field control of local ferromagnetism using a magnetoelectric multiferroic. Nature Mater. 7, 478–482 (2008).

7. Fert, A. Nobel Lecture: Origin, development and future of spintronics. Rev. Mod. Phys. 80, 1517–1530 (2008).

8. Chappert, C., Fert, A. & Nguyen van Dau, F. The emergence of spin electronics in data storage. Nature Mater. 6, 813–823 (2007).

9. Parkin, S. S. P. et al. Giant tunneling magnetoresistance at room temperature with MgO (100) tunnel barriers. Nature Mater. 3, 862–867 (2004).

10. Yuasa, S., Nagahama, T., Fukushima, A., Suzuki, Y. & Ando, K. Giant room-temperature magnetoresistence in single crystal Fe/MgO/Fe magnetic tunnel junctions. Nature Mater. 3, 868–871 (2004).

11. Dery, H., Dalal, P., Cywinski, L. & Sham, L. J. Spin-based logic in semiconductors for reconfigurable large-scale circuits. Nature 447, 573–576 (2007).

12. Butler, W. H., Zhang, X-G., Schulthess, T. C. & MacLaren, J. M. Spin- dependent tunneling conductance of Fe/MgO/Fe sandwiches. Phys. Rev. B 63, 054416 (2001).

13. Teixeira, J. M. et al. Resonant tunneling through electronic trapping states in thin MgO magnetic junctions. Phys. Rev. Lett. 106, 196601 (2011).

14. Meyerheim, H. L. et al. Surface X-ray diffraction analysis of the MgO/Fe(001) interface: evidence for an FeO layer. Phys. Rev. B 65, 144433 (2002).

15. Tusche, C. et al. Oxygen-induced symmetrization and structural coherency in Fe/MgO/Fe(001) magnetic tunnel junctions. Phys. Rev. Lett. 95, 176101 (2005).

16. Tusche, C. et al. Growth sequence and interface formation in the Fe/MgO/Fe(001) tunnel junction analyzed by surface X-ray diffraction. Phys. Rev. B 74, 195422 (2006).

17. Wang, S. G. et al. Evidence for FeO formation at the Fe/MgO interface in epitaxial TMR structure by X-ray photoelectron spectroscopy. J. Magn. Magn. Mater. 310, 1935–1936 (2007).

18. Zhang, X-G., Butler, W. H. & Bandyopadhyay, A. Effects of the iron-oxide layer in Fe–FeO–MgO–Fe tunneling junctions. Phys. Rev. B 68, 092402 (2003).

19. Oh, H., Lee, S. B., Seo, J., Min, H. G. & Kim, J-S. Chemical structure of the interface between MgO films and Fe(001). Appl. Phys. Lett. 82, 361–363 (2003).

20. Nogues, J. & Schuller, I. K. Exchange bias. J. Magn. Magn. Mater. 192, 203–232 (1999).

21. Berkowitz, A. E. & Takano, K. Exhange anisotropy—a review. J. Magn. Magn. Mater. 200, 552–570 (1999).

22. Tiusan, C. et al. Interfacial resonance state probed by spin-polarized tunneling in epitaxial Fe/MgO/Fe tunnel junctions. Phys. Rev. Lett. 93, 16602 (2004).

23. He, X. et al. Robust isothermal electric control of exchange bias at room temperature. Nature Mater. 9, 579–585 (2010).

24. Wu, J. et al. Direct measurement of rotatable and frozen CoO spins in exchange bias system of CoO/Fe/Ag(001). Phys. Rev. Lett. 104, 217204 (2010).

25. Skumryev, V. et al. Magnetization reversal by electric field de-coupling of magnetic and ferroelectric domain walls in multiferroic-based heterostructures. Phys. Rev. Lett. 106, 057206 (2010).

26. Wu, S. M. et al. Reversible electric control of exchange bias in a multiferroic field-effect device. Nature Mater. 9, 756–761 (2010).

27. Valev, V. K., Gruyters, M., Kirilyuk, A. & Rasing, Th. Direct observation of exchange bias related uncompensated spins at the CoO/Cu interface. Phys. Rev. Lett. 96, 067206 (2006).

28. Couet, S. et al. Stabilization of antiferromagnetic order in FeO nanolayers, Phys. Rev. Lett. 103, 097201 (2009).

29. Kataoka, M. Thermal desorption of oxygen-containing molecules on (100)MgO substrates for BaKBiO epitaxial growth. Physica C 254, 307–310 (1995).

30. Reif, J., Zink, J. C., Schneider, C-M. & Kirschner, J. Effects of surface magnetism on optical second harmonic generation. Phys. Rev. Lett. 67, 2878–2881 (1991).

31. Colliex, C., Manoubi, T. & Ortiz, C. Electron-energy-loss-spectroscopy near-edge fine structures in the iron-oxide system. Phys. Rev. B 44, 011402 (1991).

32. Pan, R-P., Wei, H. D. & Shen, Y. R. Optical second-harmonic generation from magnetized surfaces. Phys. Rev. B 39, 1229–1234 (1989).

33. Wierenga, H. A., Prins, M. W. J., Abraham, D. L. & Rasing, Th. Magnetization- induced optical second-harmonic generation: a probe for interface magnetism. Phys. Rev. B 50, 1282–1285 (1994).

34. Koopmans, B., Koerkamp, M. G., Rasing, Th. & van den Berg, H. Observation of large Kerr angles in the nonlinear optical response from magnetic multilayers. Phys. Rev. Lett. 74, 3692–3695 (1995).

35. Gruyters, M., Bernhard, T. & Winter, H. Evidence for noncollinearity between surface and bulk magnetization in ultrathin Co films. Phys. Rev. Lett. 94, 227205 (2005).

36. Sirotti, F. et al. Dynamics of surface magnetization on a nanosecond time scale. Phys. Rev. B 61, R9221–R9224 (2000).

37. Allenspach, R., Taborelli, M., Landolt, M. & Siegmann, H. C. Surface precursor to magnetic-domain nucleation observed by secondary-electron spin polarization. Phys. Rev. Lett. 56, 953–956 (1986).

38. Scholl, D. et al. Exchange interactions at the surface of a ferromagnet, Phys. Rev. B 43, 13309–13313 (1991).

39. Zhao, H. B. et al. Interface magnetization reversal and anisotropy in Fe/AlGaAs(001). Phys. Rev. Lett. 95, 137202 (2005).

40. Zhao, H. B. et al. Ultrafast interface magnetization dynamics in Fe/AlGaAs(001) heterostructures. Appl. Phys. Lett. 91, 05211 (2007).

41. Kugel, G. et al. Lattice dynamics of wustite (FeO). Phys. Rev. B 16, 378–385 (1977).

42. Koon, N. C. Calculations of exchange bias in thin films with ferromagnetic/antiferromagnetic interfaces. Phys. Rev. Lett. 78, 4865–4868 (1997).

43. O’Handley, R. C. Modern Magnetics Materials Principles and Application, Ch. 7, 220–224 (Wiley, 1999).

44. Meiklejohn, W. H. & Bean, C. P. New magnetic anisotropy. Phys. Rev. 102, 1413–1414 (1956).

45. Meiklejohn, W. H. Exchange anisotropy—a review. J. Appl. Phys. 33, 1328–1335 (1962).

46. Lin, X., Murthy, A. S., Hadjipanayis, G. C., Swann, C. & Shah, S. I. Magnetic and structural properties of Fe–FeO bilayers. J. Appl. Phys. 76, 6543–6545 (1994).

47. Ohldag, H. et al. Correlation between exchange bias and pinned interfacial spins. Phys. Rev. Lett. 91, 017203 (2003).

48. Xi, H. Theoretical study of the blocking temperature in polycrystalline exchange biased bilayers. J. Magn. Magn. Mater. 288, 66–73 (2005).

49. Koch, F. B. & Fine, M. E. Magnetic properties of FexO as related to the defect structure. J. Appl. Phys. 38, 1470–1471 (1967).

50. Mather, P. G., Read, J. C. & Buhrman, R. A. Disorder, defects and bandgaps in ultrathin (001) MgO tunnel barriers. Phys. Rev. B 73, 205412 (2006).

Acknowledgements The work at College of William and Mary was supported by the Office of Naval Research. The work at NRL was supported by core programmes and the Office of Naval Research. The work at Fudan University was supported by the Natural Science Foundation of China (grant nos 61222407 and 11074044).

Author contributions Y.F., G.L. and B.T.J. conceived the experiment. A.T.H. and C.L. grew the samples and carried out magnetic characterization using vibrating sample magnetometry. R.G. performed the TEM measurements. Y.F. performed the MSHG and MOKE measurements. All authors contributed to interpretation of the data. Y.F., K.J.S., G.L. and B.T.J. wrote the manuscript.

Additional information Reprints and permissions information is available online at www.nature.com/reprints. Correspondence and requests for materials should be addressed to G.L. and B.T.J.

Competing financial interests The authors declare no competing financial interests.

ARTICLES NATURE NANOTECHNOLOGY DOI: 10.1038/NNANO.2013.94

NATURE NANOTECHNOLOGY | VOL 8 | JUNE 2013 | www.nature.com/naturenanotechnology444

© 2013 Macmillan Publishers Limited. All rights reserved.