Introduction

Magnetic heterostructures offer a wide variety of opportunities for engineering materials or devices with novel functionalities. Examples include giant magnetoresistance1, spin-torque oscillators2, and exchange graded films3, among others. Such spintronics devices are typically operated at temperatures low relative to the ordering temperature of the magnetic materials of which they are comprised. Heat assisted magnetic recording is an exception in which a dynamic temperature increase, typically caused by a laser pulse, is used to alter the magnetic properties in order to achieve low field switching of high coercivity ferromagnets). Thermomagnetic switching has also been investigated in magnetic heterostructures in relative static environments by sweeping the sample temperature above and below the Curie temperature (TC) of one of the ferromagnetic layers.

Kravets et al.4,5 showed that the interlayer coupling in F1/Ni1−xCux/F1 structures could be tuned by changing the temperature above and below the TC of the nickel-copper alloy. Given that the TC depends on x, a wide temperature range of functionality was demonstrated.

Poulshkin et al.6 used the elemental specificity of resonant x-ray magnetic reflectometry of CoFe/NiCu/NiFe heterostructures to show that the magnitude of the interfacial moment was related to the magnetization of the adjacent ferromagnetic layer, and thus larger when contacting CoFe than NiFe. They also showed that the exchange coupling across the interface was sufficiently strong that the interfacial moments of the spacer layer tracked the reversal behavior of the adjacent ferromagnetic layer. Given that the CoFe and NiFe had sufficiently different coercive fields, this imposes upon the NiCu a frustration when the CoFe and NiFe magnetizations are not aligned, which could lead to a spin spiral if the spacer were magnetically ordered.

Intrinsic to these studies is a ferromagnetic proximity effect that exists at the interface between the high TC materials and the lower TC spacer layer. Magnetic proximity effects have been studied extensively. For example, Hase et al. used x-ray resonant magnetic scattering to show that for Fe/Pd heterostructures, the Fe layer can induce a large moment in the paramagnetic Pd layer that extends approximately 2 nm from the interface7. Engel et al. observed a similar enhancement in Co/Pd superlattices8. Other groups have looked at proximity effects with a material whose composition is close to that which would make the alloy ferromagnetic, e.g., Fe/Fe0.32V0.68 superlattices9. In such cases, the proximity effect has been shown to boost the ordering temperature of the alloy well above the alloy’s intrinsic TC9,10.

An observation that appears unique to heterostructures that have a low TC ferromagnet as the spacer is the surprisingly long range of the proximity effect. Whereas proximity effects in systems with paramagnetic spacers are often just a few unit cells7, Magnus et al.11 estimated at least 30 unit cells remained ferromagnetic above the TC of Co60(AlZ)40 in a Co85(AlZ)15/Co60(AlZ)40/Sm90Co10 heterostructure. They also showed coupling between the two ferromagnetic layers persisted for Co60(AlZ)40 thicknesses of at least 40 nm. Magnetic depth profiles of these samples obtained by polarized neutron reflectometry confirmed this proximity effect.

Here, we study exchange biased spin valve-like structures with a spacer layer that undergoes a ferromagnetic phase transition. This allows for thermal control of ferromagnetic exchange coupling throughout the device. Our structures are comprised of permalloy, a nickel–iron alloy typically consisting of approximately 80% nickel and 20% iron, and permalloy-copper alloys. The TC of PyxCu1−x decreases linearly by approximately 11 K per atomic-% of Cu from the TC of pure Py (around 900 K) until the dilution pushes the system into a superparamagnetic regime, theoretically at x = 0.83 in the FCC system12. The tunability of the system makes it attractive for thermomagnetic applications. Our approach combines traditional magnetometry, which is sensitive to a sample’s total magnetic moment, with polarized neutron reflectometry, which provides depth resolution of magnetic induction within a sample. Using a variety of temperature and field dependent measurements, we find that the strength of exchange bias and related layer coupling is directly correlated with the order parameter of the spacer layer and that this system has a long-range interfacial proximity effect. We also demonstrate the ability of the structure to spontaneously change its magnetic configuration from the antiparallel to parallel state upon cooling.

Fabrication and methods

We investigated devices of the form SiOx/Py0.8Cu0.2/Py0.4Cu0.6/Py/IrMn/Ta, where Py is permalloy (Ni80Fe20), and PyxCu1−x is permalloy-copper (Py-Cu) alloy. The stoichiometry indicated is that of individual sputtering targets. Samples were grown on thermally oxidized silicon substrates via magnetron sputtering (AJA International ATC Orion-8)13 with a base pressure of 8 nTorr (1 Torr = 133 Pa) in an argon atmosphere with a sputtering pressure of 3 mTorr at ambient temperature. The Py and Py-Cu alloy layers were grown with RF sputtering with rates in the range 0.2–0.4 Å/s, depending on the target; IrMn and Ta were grown at 1.2 Å/s and 0.8 Å/s, respectively, using 100 W DC. All samples were grown in the presence of a small magnetic field on the substrate plate to induce exchange bias14. The Ta was used to prevent oxidation15. Growing directly on oxidized silicon promotes (111) textured growth in the FCC metals, such as Py, Cu, and their alloys16. The substrates were rotated continuously during deposition at about 0.5 Hz.

Figure 1 shows cartoons of the device and the corresponding magnetization profile above and below TC of the spacer layer along with x-ray diffraction data of Py0.8Cu0.2 (14 nm)/Py0.4Cu0.6 (19 nm)/Py (14 nm)/IrMn (7 nm). From top to bottom, the samples are comprised of an antiferromagnetic IrMn layer that pins via exchange bias the magnetization direction of an adjacent high TC ferromagnetic Py layer, followed by a low TC ferromagnetic Py0.4Cu0.6 spacer layer, and a high TC ferromagnetic Py0.8Cu0.2 layer. The properties of this synthetic ferromagnet heterostructure are temperature-dependent: above the TC of Py0.4Cu0.6 (around 160 K for 20 nm thickness), the spacer is paramagnetic and thus unable to mediate any strong coupling between the adjacent ferromagnetic layers. As such, the multilayer stack behaves as a conventional spin valve at high temperatures, with a Py ‘pinned layer’ and a Py0.8Cu0.2 ‘free layer.’ When T falls below TC of the spacer, that layer becomes ferromagnetically ordered, which allows ferromagnetic coupling throughout the three ferromagnetic layers. Thus, at low enough temperature, the stack essentially behaves like a single exchange-biased ferromagnet. X-ray diffraction data shows the (111) peak for Py at 2θ = 44°, as expected17.

Figure 1
figure 1

Cartoon of the heterostructures (left) and magnetization profile (middle) at low (top) and high (bottom) temperatures, as described in the text along with x-ray diffraction results (right) showing a peak at (111) for Permalloy.

Full size image

DC magnetic measurements were taken using a Quantum Design VersaLab system with the vibrating sample magnetometer option. Magnetization vs. externally applied magnetic field (M(H)) curves were taken every 10 K between 50 K and 350 K to study temperature dependencies. The applied magnetic field (H) was swept from 30 kOe → -30 kOe → 30 kOe (10 kOe = 1 T μ0−1). This constitutes a major loop, as 30 kOe is sufficient to saturate the magnetizations of all ferromagnetic layers.

Polarized neutron reflectometry (PNR) measurements were conducted at the NIST Center for Neutron Research using the Polarized Beam Reflectometer. This technique allows for deduction of the depth-dependent scattering length density (ρ) depth profiles in thin films and multilayers18. Non spin-flip reflectivities (+ + and - -) provide information about the nuclear component of ρ (indicative of the sample’s nuclear composition), and a magnetic component of ρ proportional to the in-plane magnetization oriented along the applied field direction. Spin-flip reflectivities (+ - and - +) provide information about the in-plane magnetization component perpendicular to H. Using a 0.475 nm wavelength neutron beam, the specular spin-dependent reflectivities were measured as functions of wavevector transfer along the sample growth direction (Q). The sample was mounted inside a closed-cycle refrigerator, and an electromagnet was used to apply H in the sample plane, perpendicular to both the sample growth direction and the neutron propagation direction; the magnitude of the field was changed for different experiments, as described below. Data were reduced using the Reductus online software suite19. Data modeling was performed using Refl1D20. The temperature-dependent reflectivities measured in a saturating field were simultaneously fitted, using a model with temperature-dependent magnetic profiles and temperature-independent nuclear profiles. That nuclear profile was held fixed for modeling of the data (except for the scattering length density of the Ta capping layer, which could appear different due to time-dependent issues, e.g., progressive oxidation). All data were modeled as an incoherent addition of scattering from the multilayer structure and that of a bare thermally oxidized Si substrate, in order to account for a small fraction of bare spots on the sample surface. The temperature-dependent magnetic profiles are modeled in terms of three uniformly magnetized layers. This is a simplification, as proximity effects will certainly result in some temperature-dependent changes to the shape of the magnetic profile—particularly near the nominal TC of the spacer layer4,21. However, more complex models did not consistently improve fits to the data. As such, the layer-dependent magnetizations reported in this work should be interpreted as layer-wide averages.

Measurements and results

Temperature dependence

Figure 2 shows M(H) for the spin valve above and below TC of the spacer. We see the hysteresis curve of a typical spin valve above TC. There is a clear magnetic saturation and we can differentiate the switching of the two ferromagnetic layers. The pinned layer (Py) shows exchange bias (HEB = 54 Oe at 350 K), and has a coercive field of approximately 16 Oe; the free layer (Py0.8Cu0.2) shows no exchange bias, and has a coercive field of approximately 2 Oe. As we cool the sample below TC of the spacer, we observe several changes in the M(H) loops. There is an increase in total moment, consistent with the spacer undergoing a magnetic phase transition. We see HEB of the pinned layer increase, consistent with expected temperature dependence of an exchange biased bilayer22. We also see a bias develop in the free layer when it becomes coupled to the pinned layer. Related phenomena have been reported elsewhere4,5. We will refer to the shift of the free layer’s hysteresis loop as the coupling field, HF, and retain HEB for the pinned layer. This shift, HF, resembles exchange bias, and has a related explanation. The Py layer is positively saturated throughout the reversal of the free layer. The ferromagnetic coupling between the pinned and the free layer mediated by the ferromagnetic spacer layer imposes a torque that opposes[aids] the free layer reversal for negative[positive] sweeping field. This leads the hysteresis loop of the free layer to shift to negative field by an amount HF.

Figure 2
figure 2

M(H) curves of Py0.8Cu0.2 (14 nm)/Py0.4Cu0.6 (19 nm)/Py (14 nm)/IrMn (7 nm) above TC (top) and below TC (bottom) of the spacer with HEB, and HF shown for both curves.

Full size image

Figure 3 shows the temperature evolution of HF and HEB. Because the spacer magnetization increases with decreasing temperature, the coupling field, HF, does as well. This behavior is directly analogous to that observed for a typical exchange biased bilayer cooled through the blocking temperature of the antiferromagnet. For HEB, the coupling caused by the spacer’s phase transition leads to a reduction of HEB for temperatures below the spacer's Tc, leading to a deviation from a typical exchange bias temperature dependence. Interestingly, the sum of these fields, HF + HEB, shows the smooth temperature dependence expected in exchange biased systems. This is explainable by considering the torques imposed on the pinned Py layer from the IrMn at one interface and the Py0.4Cu0.6 layer at the other interface. The coupling to the antiferromagnet causes the Py hysteresis loop to shift to negative fields by opposing[assisting] the reversal of the magnetization for the negative[positive] sweeping field. Since the Py0.4Cu0.6 will couple ferromagnetically to the Py and is negatively saturated throughout the duration of the Py layer switching, it assists[opposes] the Py layer switching for the negative[positive] field sweeping direction. Thus, the torques acting upon the Py layer by the IrMn and Py0.4Cu0.6 oppose each other once the Py0.4Cu0.6 has long range order. This leads to a reduction of the magnitude of HEB from that of the isolated IrMn/Py behavior. We can then recover the intrinsic temperature dependence by adding the measured HF to the measured HEB.

Figure 3
figure 3

Temperature dependence of the free layer coupling field HF (purple triangles) and exchange bias field HEB (green circles). The sum H F + HEB (solid line) is the temperature dependence of the IrMn/Py in the absence of any coupling to the free layer. Both curves shown are of Py0.8Cu0.2 (14 nm)/Py0.4Cu0.6 (19 nm)/Py (14 nm)/IrMn (7 nm)23.

Full size image

We also studied the temperature dependence of the structure with PNR to explicitly resolve the layer-dependent magnetization. Temperature-dependent PNR measurements were taken with magnetic fields nominally parallel to the exchange anisotropy axis and with magnitudes of 5 kOe (at saturation) and 10 Oe (at remenance). Only non spin-flip reflectivities were measured for these conditions because we expected no appreciable magnetization component perpendicular to H (and thus no spin-flip scattering). Figure 4a shows example fitted data at 100 K and 5 kOe for a sample with a nominal spacer thickness of 19 nm. The model fits closely match the data; the corresponding nuclear and magnetic scattering length density profiles are shown in Fig. 4b. The nuclear and magnetic scattering length densities of Py are larger than that of Cu, thus the increase in both values with Py concentration is consistent with expectations. The thicknesses of the magnetically active outer and spacer layers are 14 nm and 19 nm, respectively, also consistent with expectations. The corresponding saturation and remenant layer-by-layer temperature-dependent magnetizations are shown in Fig. 4c. Bloch-Law fits are shown as lines through the data. The pinned Py and free Py0.8Cu0.2 layers’ magnetizations are essentially field-independent, as their TC values are both well above room temperature. The Bloch-Law fits indicate that with a small applied magnetic field of 10 Oe, the spacer layer has approximately TC = 221 K, with this seemingly enhanced to 283 K under the 5 kOe saturation field. The inflection point (max dM/dT) is dramatically higher for the 5 kOe saturation field than would normally be expected, hence this is a large temperature difference for such a small field change, compared to phase transitions in bulk ferromagnetic materials24.

Figure 4
figure 4

(a) Representative fitted PNR data for Py magnetization parallel to H. (b) Scattering length depth profile used to fit the data in (a). The nuclear profile shown was used for all fitting described in this work. (c) Temperature-dependent layer-by-layer magnetizations for remenant (closed symbols) and saturation (open symbols) field conditions. Note: 1 mTμ0−1 = 10 Oe. All curves of Py0.8Cu0.2 (14 nm)/Py0.4Cu0.6 (19 nm)/Py (14 nm)/IrMn (7 nm).

Full size image

Combining the conventional magnetometry results with those of the PNR gives us some insight into the physics underlying the behavior of these structures. Figure 5 (left axis) shows the temperature dependence of HF, as measured by VSM. Figure 5 (right axis) shows the temperature dependence of the spacer layer’s magnetization, as measured by PNR. The correlation is obvious. Together, these data indicate that the coupling of the free layer to the pinned layer is mediated through the spacer, with the strength of that coupling being dictated by the magnetic order parameter of the spacer layer. The order parameter of the spacer thus serves as a knob that can tune the coupling between the free and pinned layers.

Figure 5
figure 5

The correlation of the temperature dependences of the coupling field of the free layer (triangles, left axis) from VSM and magnetization of the spacer layer (circles, right axis) from PNR depth profiles show that the spacer layer mediates a tunable coupling between the free and pinned layers of Py0.8Cu0.2 (14 nm)/Py0.4Cu0.6 (19 nm)/Py (14 nm)/IrMn (7 nm).

Full size image

Thickness dependences

We also investigated the behavior of this system as a function of the spacer layer thickness with samples of the form SiOx/Py0.8Cu0.2/Py0.4Cu0.6(t)/Py/IrMn/Ta, where t ranged from 3.5 nm to 71 nm with logarithmic spacing. Figure 6 shows example hysteresis loops above and below the Curie temperature for three spacer layer thicknesses. For samples with t = 14 nm or more (right, green in Fig. 6), we see behavior similar to that shown in Fig. 2: there is distinct switching of the pinned and free layers at high temperature, with a smearing of the loops when the spacer layer has long range order at low temperatures. The thinnest sample, t = 3.5 nm (left, purple in Fig. 6), however, shows the layers to be completely coupled even at high temperatures; no distinct loops can be discerned for the pinned and free layers. This strong coupling may imply samples of this thickness are fundamentally different from the others. The t = 7 nm sample (middle, blue in Fig. 6) shows the transition between these two extremes. Other groups4 have made similar thermomagnetic devices and conducted interlayer thickness studies, however they found that coupling diminishes beyond an interlayer thickness of approximately 10 nm. In our studies, we determine that (at least partial) coupling exists at interlayer thicknesses of up to approximately 70 nm. We believe that this long range magnetic coupling interaction could happen for many reasons. For example, it could be that this system has highly enhanced susceptibility, which enables the distance between the ferromagnet layers to be artificially enhanced (longer than typically expected). It’s possible that the enhanced susceptibility could be related to some of the layers being diluted ferromagnets. Another possibility is that there could be clusters of Py within the spacer layer, each of which may again help to propagate the magnetic state to longer length scales than if the spacer were pure copper. We also note that exchange coupling becomes too strong for the device to properly function as a spin valve with interlayer thicknesses of 3.5 nm.

Figure 6
figure 6

Hysteresis loops above (top row) and below (bottom row) the spacer layer’s TC, for Py0.8Cu0.2 (14 nm)/Py0.4Cu0.6 (3.5 nm, 7 nm, and 14 nm)/Py (14 nm)/IrMn (7 nm).

Full size image

From an exchange spring perspective25, the stronger HF for thinner films suggests the spacer is unable to support a domain wall or twist of the magnetization within its thickness. In contrast, the thicker samples may allow such a twist or even full domain wall. That would allow the free layer magnetization to reverse at lower fields, even in the case that the order parameter of the spacer was independent of thickness. This would lead to an inverse relationship between HF and t, ultimately with the thinnest sample showing full coupling within the ferromagnetic layers. The transition between the strong and weak coupling appears to be for a thickness around 7 nm.

The thickness dependencies of HF and the spacer layer TC both scale inversely with thickness, as shown in Figure 7. The coupling deviates from this behavior for the thinnest (3.5 nm) sample; this could be due to roughness/orange-peel coupling becoming dominant for such a thin layer. Fitting the HF(t) = he-t/λ to a decaying exponential for the thicker samples reveals a characteristic length scale of λ = 7.0 ± 1.5 nm. This is a long length scale for a magnetic proximity effect, but consistent with observations in other systems11,21. This may have some enhancement because of the structural coherence of the system, in that all the layers are FCC structure with minor differences in lattice parameters. Doing a similar fit of TC(t) reveals a length scale of 4.7 nm ± 1.0 nm. To further investigate proximity effects on the spacer’s TC, additional temperature-dependent PNR measurements were taken at 500 Oe to obtain M vs. T data (similar to Fig. 5) from samples with different spacer thicknesses. Relative to a TC of 145 K ± 7 K for an isolated 27 nm thick film of the spacer material, we found TC for t = 71 nm and t = 13 nm films to be 158 K ± 7 K and 226 K ± 12 K, respectively. Though we only have two data points from PNR at this time, this inverse relationship between spacer thickness and ordering temperature is a clear signature of a magnetic proximity effect9,11,21,26, which supports the conclusions drawn from conventional magnetometry data.

Figure 7
figure 7

The coupling field (top, purple triangles) and spacer layer Curie Temperature (bottom, blue squares) are inversely related to the spacer layer thickness. The red line is a linear fit. These are Py0.8Cu0.2 (14 nm)/Py0.4Cu0.6 (3.5–71 nm)/Py (14 nm)/IrMn (7 nm).

Full size image

Thermally induced switching

The onset of coupling between the pinned and free layers at low temperatures can lead to a spontaneous reversal of the free layer. As demonstrated in Figure 8 the free layer will rotate its moment to align with that of the pinned layer when the device is set into the antiparallel state then cooled through the phase transition of the spacer layer. SiOx/Py0.8Cu0.2(14 nm)/Py0.4Cu0.6(20 nm)/Py(13 nm)/IrMn(7 nm)/Ta(5 nm). Figure 8 shows the total moment of the sample as a function of temperature, as measured by VSM. The transition is related to the total moment being low in the antiparallel state (the free and pinned magnetizations are pointed in opposite directions and thus effectively cancel each other) and high in the parallel state (free and pinned magnetizations are pointed in the same direction). The strong change of net magnetic moment could be useful for in sensing and possibly energy harvesting.

Figure 8
figure 8

When the device is set to the antiparallel configuration at room temperature, then cooled through the Curie point of the spacer layer, the free layer rotates to align with the pinned layer, resulting in an increase in the net magnetic moment measured by VSM. This is for Py0.8Cu0.2 (14 nm)/Py0.4Cu0.6 (19 nm)/Py (14 nm)/IrMn (7 nm).

Full size image

PNR was used to directly characterize the depth-dependence of the thermal switching phenomenon, as shown in Figure 9. At room temperature, well above the nominal TC of the spacer layer, the sample was oriented with the pinned Py magnetization antiparallel to a 10 Oe applied field used to orient the magnetization of the low anisotropy free layer. Both non spin-flip and spin-flip data were measured as a function of decreasing temperature in order to characterize the layer-dependent magnetization rotation as the spacer layer becomes magnetically ordered and thus capable of mediating direct exchange between the free and pinned layers. Inspection of the data suggest a sharp magnetic switching between 170 and 150 K, as data taken above and below those two temperatures respectively are only weakly temperature-dependent. Fitted non spin-flip data taken near the switching temperature are shown in Fig. 9a, plotted as the difference between +  + and - -, normalized by the theoretical reflectivity of the bare substrate. As the nuclear contribution to +  + and - - is essentially temperature independent, changes in the difference track changes to the component of the in-plane magnetization parallel to H. The corresponding fitted spin-flip data are shown in Fig. 9b, plotted as the average of + - and - + (which are identical under the conditions here), again normalized by the substrate reflectivity. These data track changes to the component of the in-plane magnetization perpendicular to H. The drastic change in non spin-flip reflectivities and the spike in spin-flip reflectivities at 160 K indicate a switching from nominally antiparallel to nominally parallel alignment of the free and pinned layers. The data are well-fit by a highly constrained model parameterized in terms each layer’s magnetization magnitude, and the angle (ɸ) between that magnetization and H. Specifically, the model assumes:

  • An uncoupled to coupled transition corresponding to the changes in scattering at 160 K;

  • ɸ for the pinned Py layer is T-independent, and is treated as a free parameter to account for misalignment from the desired 180 degree orientation with H;

  • At T = 170 K and above, ɸ = 0 degrees for the free and spacer layers;

  • At T = 160 K, ɸ is constrained to be the same for the free and spacer layers, and is treated as a free parameter;

  • At T = 150 K and below, ɸ for all three layers are the same.

Figure 9
figure 9

Summary of PNR results with Py magnetization nominally antiparallel to H (10 Oe) of Py0.8Cu0.2 (14 nm)/Py0.4Cu0.6 (19 nm)/Py (14 nm)/IrMn (7 nm). (a) Fitted non spin-flip data near the thermal switching transition temperature. (b) Fitted spin-flip data near the transition. (c–f) Summary of the temperature-dependent magnetization parameters used to fit the data. (g) Depiction of the sample during thermal switching at 160 K.

Full size image

The good fits demonstrate that the data are consistent with the above constraints. More complex models, or those with more parameter freedom, do not result in significantly better fits to the data. Most notably, the magnetization of the spacer layer is small enough near the 160 K transition, that the direction cannot be reliably determined as an independent parameter. The modeling results are summarized in Fig. 9 (c-f). At high T, the pinned layer (Py) magnetization is pinned at an angle of 8 degrees away from antiparallel with H, the spacer layer (Py0.4Cu0.6) has zero magnetization, and the free layer (Py0.8Cu0.2) magnetization is aligned parallel with H. Below 200 K, the Py0.4Cu0.6 layer begins to magnetically order, and below 170 K the interlayer coupling mediated via that layer becomes strong enough to reorient the free layer. At 160 K we capture the sample in transition as depicted in Fig. 9g. At 150 K, the interlayer coupling is strong enough to have completely oriented the free and spacer layer magnetizations parallel to that of the pinned Py. Notably, the apparent TC of the Py0.4Cu0.6 layer is 195 K in this antiparallel configuration, which is 25 K lower than what was observed for the parallel state to H (Fig. 9c). The TC difference is likely a consequence of dueling proximity effects at the two Py0.4Cu0.6 interfaces; related phenomena have been seen in superconducting heterostructures27. The magnetization profile within the Py0.4Cu0.6 layers may well even be twisted near the transition temperature, but such detail was not possible to distinguish with our measurements.

Conclusions

Exchange biased spin-valve-like structures with a spacer layer that undergoes a magnetic phase transition exhibit novel phenomena. We showed that these phenomena are interface effects with an inverse relationship to thickness. The magnetic proximity effect, an enhancement of the ordering temperature of the spacer, has a relatively long length scale of about 7 nm in this system. The coupling between the free and pinned layers is strongly affected by the magnetic order within the spacer layer. With sufficient order, the system is fully ferromagnetically coupled. With no long-range magnetic order, the system is essentially decoupled, allowing the outer layers to switch independently. Between these two extremes, the coupling is strongly dependent on magnetic order and thus temperature. The nominally free layer acquires a bias due to this coupling, which manifests as a shift of its hysteresis loop toward that of the exchange biased pinned layer until they ultimately merge when sufficient long range magnetic order is present in the spacer. This coupling follows a temperature dependence that is strongly correlated with that of the spacer layer’s order parameter. We demonstrated this by showing that the system prepared in the anti-parallel magnetic configuration spontaneously rotates into the parallel configuration when the temperature crosses a threshold related to the establishment of long-range magnetic order in the spacer. This suggests the potential for novel thermo-magnetic devices in which the spacer layer influences the state of the device.