Perpendicular magnetic tunnel junctions (pMTJs) with synthetic antiferromagnetic (SAF) free layers have attracted much interest for applications on spintronic memory devices with ultrafast speed and ultralow energy. In this work, SAF free layer pMTJs (SAF-pMTJs) were designed and fabricated, in which a Ru/Ta bilayer spacer is used to form the SAF structure. We first characterized the magnetization dynamics of the SAF free layer by using ferromagnetic resonance and found that the Gilbert damping constant of the SAF free layer is around 0.019. Then, in device level studies that span from 900 nm down to 200 nm lateral size, we observed a transition of the SAF free layer from a preferred antiparallel to parallel magnetic configuration at small device sizes, which can be explained by the increased dipole interaction. The impact of the operating current was also investigated. We report an extraordinarily strong dependence, up to 144.1 kOe per A/μm2, of the offset field on the applied current, suggesting an electric-field modulation on the interlayer exchange coupling of the SAF free layer. These results will be instructive to improve the understanding of material properties and device performance of SAF-pMTJs for ultrafast, ultralow-power consumption spintronic devices.

Reducing the switching current density and increasing the switching speed are central to the development of next-generation magnetic tunnel junction (MTJ)-based memory and logic devices. MTJs with perpendicular magnetic anisotropy (PMA), or pMTJs, enable lower switching current density and faster write speed than their in-plane magnetized counterparts, while maintaining sufficient thermal stability (Δ).1–3 Early works, for example, have demonstrated >40% reduction in the switching current density2 and 8.0× faster switching speed3 by using PMA in comparison with in-plane MTJ devices. In recent years, many researchers predicted that ultralow-current density, ultrahigh-speed switching could be achieved with pMTJs that employ synthetic antiferromagnetic (SAF) free layers, or SAF-pMTJs.4–6 Indeed, faster switching speeds of SAF-pMTJs were recently demonstrated in experiments.7 In 2013, Yoshida et al. developed a SAF free layer with CoFeB and Co/Pd multilayers antiferromagnetically coupled through a Ru/Ta bilayer spacer.8 They demonstrated that, by utilizing the bilayer-mediated SAF free layer, the Δ could be improved by ∼64% without increasing the switching current. In 2016, the enhanced Gilbert damping constant (α) was measured in a composite CoFeB-Co/Ni multilayered free layer that was indirectly exchange coupled ferromagnetically.9 Evaluating α in a SAF free layer is similarly consequential for understanding the performance of SAF-pMTJ devices. Nevertheless, such studies are still lacking. Moreover, spintronic devices must be scaled to a certain technical node for practical applications. To date, pMTJs have been experimentally demonstrated to be scalable down to less than 10 nm10 or even 2.3 nm.11 It is desirable to investigate the magnetic properties of SAF-pMTJs at small sizes. In addition, pMTJ devices were reported to show interesting behavior under electric bias.12,13 Evaluating the current- and voltage-induced changes to the magnetic properties of SAF free layers is of significance to the development of SAF-pMTJ devices. However, as SAF-pMTJs attract more attention, a comprehensive investigation on their magnetization switching characteristics, including the impact of the device size and applied bias, is yet to be reported.

In this Letter, we prepared Ru/Ta spacer-mediated SAF-pMTJs and report on film and device level characterization of the magnetic layers comprising the stack. Broadband ferromagnetic resonance (FMR) measurements of the unpatterned film show that the SAF free layer, comprising of a CoFeB layer and Co/Pd multilayers, exhibits an α value moderately higher than what can be achieved in certain isolated CoFeB layers. To explore the magnetization switching characteristics of SAF-pMTJs, the stack was patterned into pillar devices with diameters (d) ranging from 900 nm down to 200 nm. The switching behavior at different device sizes is compared. Surprisingly, a sign reversal of the offset field (HO) of the free layer hysteresis loop—suggesting a transition from antiferromagnetic to ferromagnetic coupling—occurred as the pMTJ diameter was reduced. Furthermore, we applied different currents (I) during observations of the switching field. The coercivity (HC) of the SAF free layer is significantly manipulated because of the voltage (V)-controlled magnetic anisotropy (VCMA) effect. The HO is also observed to be tunable, which can be explained by the voltage-controlled exchange coupling (VCEC). These results can guide the design of SAF-pMTJs at the stack and device level.

The SAF-pMTJ stack was deposited in an ultrahigh vacuum (base pressure better than 5 × 10−8 Torr) system employing magnetron sputtering at room temperature. As depicted in Fig. 1(a), the stack structure is Si/SiO2 subs./Ta (5)/Ru (5)/Pd (5)/[Co (0.3)/Pd (0.7)]4/Co (0.3)/Ru (0.6)/Ta (0.3)/Co20Fe60B20 (1.0)/MgO (2)/Co20Fe60B20 (1.3)/Ta (0.7)/[Pd (0.7)/Co (0.3)]8/Pd (5), in which the MgO layer was deposited by radio frequency sputtering at an Ar pressure of 1.5 mTorr while the remaining layers were deposited by direct current sputtering at an Ar pressure of 2.0 mTorr. The numbers in parentheses are the layer thickness in nanometers, and the subscripted compositions are in atomic percent. After deposition, the samples were annealed at 275–300 °C for 10 min by a rapid thermal annealer to promote the PMA of CoFeB layers and crystallization of the MgO tunneling barrier. The [Co (0.3)/Pd (0.7)]4/Co (0.3)/Ru (0.6)/Ta (0.3)/Co20Fe60B20 (1.0) functions as the free layer. The Co20Fe60B20 (1.3)/Ta (0.7)/[Pd (0.7)/Co (0.3)]8 is the reference layer. The magnetic hysteresis loop was characterized by the vibrating-sample magnetometer module of a physical property measurement system (PPMS)14 and shown in Fig. 1(b). The blue arrow, short red arrow, and long red arrow represent the magnetization of the reference layer, free layer-CoFeB, and free layer-Co/Pd multilayers, respectively. The saturation magnetization (MS) of the full SAF-pMTJ stack, normalized by area, is approximately 7.73×104 emu/cm2. The external field (Hext) is applied perpendicular to the film (see the supplementary material, Fig. S1 for the in-plane M-Hext loop). The stack presents an apparent out-of-plane easy axis for both the free layer and the reference layer with a two-step switching behavior, reflecting the different coercive fields of the layers. After saturation at high fields, the magnetization starts to decrease at about 500 Oe when ramping back, indicating an antiferromagnetic coupling. At this point, the magnetization of free layer-CoFeB (approximately 893 emu/cm3) is reversed. By oppositely ramping up to approximately 1000 Oe, the magnetization decreases again until it oppositely saturates at around 1600 Oe. This corresponds to the switching of both free layer-Co/Pd multilayers and the reference layer, likely due to the dipole interaction and similar HC values. To confirm the magnetic configuration deduced above, a simplified stack with only the free layer structure capped by a 2 nm-thick Ta layer (i.e., Si/SiO2 subs./Ta (5)/Ru (5)/Pd (5)/[Co (0.3)/Pd (0.7)]4/Co (0.3)/Ru (0.6)/Ta (0.3)/Co20Fe60B20 (1.0)/MgO (2)/Ta (2)) was prepared. The M-Hext loop is plotted in Fig. 1(b) as well for comparison. The magnetization decreases before zero field when ramping down, additionally confirming the antiferromagnetic SAF configuration. The interlayer exchange coupling energy (−Jex) is calculated15 to be 0.035 erg/cm2, about one order of magnitude lower than that of the previous study8 by Yoshida et al., which could be explained by the increase in the Ta spacer thickness.

FIG. 1.

(a) Schematic illustration of the stack structure of SAF-pMTJ, which composes of a bottom SAF free layer and a top reference layer. (b) The perpendicular M-Hext loops of the full SAF-pMTJ (black line and circle) and its free layer (red line and square). The magnetic configurations at each stage are presented by the blue arrows (reference layer), short red arrows (free layer-CoFeB), and long red arrows (free layer-Co/Pd multilayers).

FIG. 1.

(a) Schematic illustration of the stack structure of SAF-pMTJ, which composes of a bottom SAF free layer and a top reference layer. (b) The perpendicular M-Hext loops of the full SAF-pMTJ (black line and circle) and its free layer (red line and square). The magnetic configurations at each stage are presented by the blue arrows (reference layer), short red arrows (free layer-CoFeB), and long red arrows (free layer-Co/Pd multilayers).

Close modal

FMR spectroscopy measurements using a signal generator and a microwave diode detector were carried out on the SAF-pMTJ film to further investigate the magnetic properties. To increase the FMR sensitivity, the Hext was modulated, and a lock-in detection scheme was used, whereby FMR spectra were measured at fixed microwave frequencies (16–35 GHz) under a swept Hext. Figure 2(a) shows illustrative FMR measurements for perpendicular Hext from which we extract the frequency-dependent resonance field, Hres, and the linewidth ΔH. The solid lines reflect the best fit of the raw absorption data (unshaded markers) to a derivative Lorentzian absorption line. Expressions for the frequency vs Hext dispersion are given by the Kittel equation, f=μ0γ(Hres+HK,eff), for which f is the FMR frequency, γ is the gyromagnetic ratio, μ0 is the vacuum permeability, and HK,eff is the effective anisotropy field, taking into account both the demagnetizing field and the perpendicular magnetic anisotropy field. Curve-fitting of the Hres vs f graphs [shown in Fig. 2(b)] yields the HK,eff and γ values. The composite SAF free layer has a modest HK,eff of 1.02(2) kOe for which uncertainties reflect the one-sigma variance of the covariance matrix from curve-fitting to the Kittel equation. The total magnetic thickness (t = 5.3 nm) of the Co/Pd multilayers-CoFeB complex is advantageous for lateral scaling as Δ is proportional to HK,eff×t. The present SAF free layer has the potential to deliver similar Δ as a replacement for the single CoFeB layers (∼0.8–1.4 nm) employed in pMTJs with an HK,eff in excess of 5 kOe.1–3 The reference layer exhibited a marked larger HK,eff of 5.40(20) kOe, consistent with the engineering of a composite reference layer with thicker (8 nm) Co/Pd multilayers than the free layer (4 nm). The α was determined from the frequency dependence of the linewidth, μ0ΔH=2α/γf+μ0ΔH0. Figure 2(c) summarizes the ΔH vs f data for both the free and reference layers. From the lines of best fit shown on the figure, we estimate an α of 0.019(2) for the free layer and 0.025(6) for the reference layer, which are between the typical values of CoFeB (∼0.008–0.015, at a thickness of 1 nm)16–18 and those of Co/Pd multilayers (∼0.05–0.2, depending on the structure parameters),19–21 likely reflecting a moderate spin pumping contribution from the thick Pd buffer and Co/Pd multilayers when compared to single CoFeB films with even lower Gilbert damping.22 The inhomogeneous line broadening was 250(30) Oe for the free layer and 1900(100) Oe for the reference layer. These values are also moderately elevated compared to single CoFeB films and reflect the magnetic inhomogeneity that often develops as a consequence of roughness in sputtered, polycrystalline, face-centered-cubic (111) Co-based multilayers.23,24

FIG. 2.

(a) The FMR spectra showing the linear frequency (f ranges from 16 to 35 GHz) dispersion for out-of-plane Hext. The unshaded markers are the raw data. Solid lines reflect their best fit to a derivative Lorentzian absorption line. (b) Out-of-plane Hres vs f, where the lines reflect fitting the Kittel equation to the data (open markers) and (c) out-of-plane ΔH vs f. The uncertainties reflect the one-sigma variance of the covariance matrix from curve-fitting to the Kittel equation.

FIG. 2.

(a) The FMR spectra showing the linear frequency (f ranges from 16 to 35 GHz) dispersion for out-of-plane Hext. The unshaded markers are the raw data. Solid lines reflect their best fit to a derivative Lorentzian absorption line. (b) Out-of-plane Hres vs f, where the lines reflect fitting the Kittel equation to the data (open markers) and (c) out-of-plane ΔH vs f. The uncertainties reflect the one-sigma variance of the covariance matrix from curve-fitting to the Kittel equation.

Close modal

To study the magnetization switching characteristics, we patterned SAF-pMTJs into different sizes, ranging from d = 900 nm to d = 200 nm, by conventional E-beam lithography and Ar+ ion milling. The resistance (R)-Hext loops were tested using a direct current four-probe method by a PPMS at room temperature, as presented in Fig. 3(a). The positive current direction is defined by electrons flowing from the bottom to the top of MTJ devices. Figures 3(b)–3(d) are representative major hysteresis loops at d = 900 nm, 500 nm, and 200 nm, respectively. Red (black) curves indicate sweeping from positive (negative) to negative (positive) fields. At 900 nm, the parallel-to-antiparallel (P AP) switching happens at about 225 Oe when ramping down after positive saturation, showing an antiferromagnetic coupling and is consistent with the M-Hext loop of the unpatterned film. At 500 nm, the P  AP switching happens at nearly zero field, suggesting that the switching field (Hsw) of the CoFeB free layer minor hysteresis loop has been significantly reduced. Finally, at 200 nm, the Hsw is delayed to approximately −400 Oe (i.e., along the opposite direction). In consideration of the large HC of free layer-Co/Pd multilayers, it is reasonable to believe that they are still magnetized along the original direction at zero field. Namely, the SAF magnetic configuration fails in small devices. This is because, at small device sizes, the substantial fringe fields among the free layer-CoFeB, free layer-Co/Pd multilayers, and reference layer generate additional ferromagnetic coupling of a dipolar origin.25,26 If the ferromagnetic coupling is strong enough to overcome the antiferromagnetic interlayer exchange coupling, the ferromagnetic layers in the free layer will be parallelly magnetized. To keep the SAF magnetic configuration while device downscaling, it is desirable to reduce the impact of stray fields. A widely used method to mitigate the stray field is fabricating a SAF reference layer.27,28 Nonetheless, a perfect stray field cancelation of the reference layer remains challenging, particularly at the deep-nanoscale, due to edge defects.29,30 The primary origin of the ferromagnetic coupling is the dipole interaction within the SAF free layer (e.g., between the free layer-CoFeB and free layer-Co/Pd multilayers in this study), to which the dual SAF free layer presented in our previous study31–33 could be a feasible solution.

FIG. 3.

(a) Device footprint and electrical connection employed in the direct current four-probe method, by which all the R-Hext loops were measured. In actual measurements, devices were connected to the PPMS circuit by wire bonding, instead of physical probes. (b)–(d) The major R-Hext loops of SAF-pMTJ devices at d = 900 nm, d = 500 nm, and d = 200 nm, respectively. The black lines and circles show the data from the negative-positive sweeping, meanwhile the red lines and squares represent the data from the positive-negative sweeping. The sweeping directions are also indicated by the black and red arrows in (b). Two regions in (b) are circled, and the corresponding data are replotted in (e) and (f). The magnetic configurations at each stage are presented by the blue arrows (reference layer), short red arrows (free layer-CoFeB), and long red arrows (free layer-Co/Pd multilayers).

FIG. 3.

(a) Device footprint and electrical connection employed in the direct current four-probe method, by which all the R-Hext loops were measured. In actual measurements, devices were connected to the PPMS circuit by wire bonding, instead of physical probes. (b)–(d) The major R-Hext loops of SAF-pMTJ devices at d = 900 nm, d = 500 nm, and d = 200 nm, respectively. The black lines and circles show the data from the negative-positive sweeping, meanwhile the red lines and squares represent the data from the positive-negative sweeping. The sweeping directions are also indicated by the black and red arrows in (b). Two regions in (b) are circled, and the corresponding data are replotted in (e) and (f). The magnetic configurations at each stage are presented by the blue arrows (reference layer), short red arrows (free layer-CoFeB), and long red arrows (free layer-Co/Pd multilayers).

Close modal

A marked feature of the major loops is the long tail of antiparallel-to-parallel (AP  P) switching in contrast with the sharp P  AP switching. Though the AP P switching looks difficult in general, small, sharp switching points can be found as circled in Fig. 3(b). The data of interest are highly symmetric and replotted in Figs. 3(e) and 3(f), respectively. These characteristics can be attributed to the antiferromagnetic coupling within the free layer. Specifically, the magnetization direction of free layer-CoFeB is not stable after the P AP switching. Instead, as Hext ramps up along the opposite direction, the magnetization of free layer-Co/Pd multilayers also tends to reverse. As a result, the magnetization direction of free layer-CoFeB will be off from the AP configuration because of the antiferromagnetic coupling. The junction resistance will decrease accordingly. The free layer-CoFeB, in turn, hinders the switching of free layer-Co/Pd multilayers and causes the extension of the AP P switching. At an elevated Hext of around 6600 Oe, however, the switching of the reference layer reverses a strong stray field and breaks the balance in the free layer. Immediately afterwards, the free layer-CoFeB, free layer-Co/Pd multilayers, and reference layer align parallelly, leading to small, sharp switching points. The magnetization switching behaviors discussed above also exist in conventional pMTJ structures (i.e., single magnet free layer and SAF reference layer). In 2019, Safranski and Sun reported the decrease in the AP state resistance at high Hext.34 Very recently, Han et al. also reported a similar phenomenon.35 In both studies, the decrease in the AP state resistance is ascribed to insufficient PMA of the reference layer. Han et al., in the same work, also observed small, sharp switching points,35 which are explained with the similar mechanism discussed in this study (i.e., simultaneous switching of the free layer-CoFeB and reference layer).

To better understand the magnetization switching behavior of SAF-pMTJ devices, the minor R-Hext loops are measured as well. Prior to the measurements, all the devices are positively saturated. As shown in Fig. 4(a), the black lines and circles present the minor loop at d = 200 nm. RAP and RP represent the resistance at AP and P states, respectively. A HC (given by the half-width of minor loop) of about 92.5 Oe and an HO (defined as the center of minor loop) of approximately −312.5 Oe are obtained. The values of HC and HO at different device sizes are summarized in Fig. 4(b). As the device size decreases, HO monotonously changes from positive to negative. This is because the positive stray fields become stronger as previously discussed. Figure 4(b) also illustrates that HC largely increases as the device size shrinks with the exception of an outlier at d = 300 nm. At d = 900 nm, the HC is only about 25 Oe in contrast with around 92.5 Oe at d = 200 nm. This can be explained by the increase in the demagnetizing factor as well as the transition from domain nucleation and propagation behavior to the Stoner–Wohlfarth coherent rotation.36 

FIG. 4.

(a) The minor R-Hext loops of SAF-pMTJs measured with different I. The device diameter d equals 200 nm. (b) The change of HC and HO with d. The measurements are conducted with an I = 1 μA. (c) The change of HC and HO with I. The device diameter d equals 200 nm. The red dashed line is the linear fitting of HO. (d) The HO as a function of V and the linear fitting.

FIG. 4.

(a) The minor R-Hext loops of SAF-pMTJs measured with different I. The device diameter d equals 200 nm. (b) The change of HC and HO with d. The measurements are conducted with an I = 1 μA. (c) The change of HC and HO with I. The device diameter d equals 200 nm. The red dashed line is the linear fitting of HO. (d) The HO as a function of V and the linear fitting.

Close modal

All of the major and minor R-Hext loops discussed above were measured with an excitation current of 1 μA. To further study the magnetization switching characteristics of SAF-pMTJs under electric bias, different I values were tested at d = 200 nm. The red lines and squares and green lines and nablas in Fig. 4(a), respectively, represent the R-Hext loops measured at I = 0.2 μA and 8 μA. By comparing these two loops with that of I = 1 μA (black lines and circles), it is clear that HC decreases as I increases. There could be two origins of the modification on HC: Joule heating12 and electric-field (E-field),13 as very recently reported by Krizakova et al.37 To figure out the major factor of the HC decrease, we applied a I = −1 μA. The data are shown in Fig. 4(a) by the blue lines and triangles. Though the magnitude is the same, the negative polarity of I leads to an increase in HC compared with the I = 1 μA loop. By further applying a large negative I of −8 μA, the HC becomes larger as represented by the purple lines and stars in Fig. 4(a). A more extensive investigation on the I dependence of HC is shown in Fig. 4(c), in which I ranges from −8 to 8 μA. The odd parity clearly reveals that the E-field effect is a dominant factor. A detailed analysis (see the supplementary material, Fig. S2) shows a linear dependence of HC on the E-field with a slope of 114.3 Oe per V/nm, which is in the same magnitude with the literature values of single layer CoFeB.38 

The change of HO is also evident in Fig. 4(a). The I dependence of HO is summarized and shown in Fig. 4(c) as well. Generally, as we can see, a large I can shift the minor loop toward the positive side. The spin-transfer torque (STT) may contribute to the change of HO, as reported by Krizakova et al.37 and Mihajlović et al.39 As STT is proportional to I, HO should exhibit a linear I dependence. The red dashed line in Fig. 4(c) represents the corresponding fitting result. However, it must be noted that the slope, obtained from linear fitting, is determined to be 144.1 kOe per A/μm2, roughly 8× or even 20× larger compared with literature values.37,39 Although the slope can be attributed to the STT efficiency, it is unlikely to obtain such a large improvement purely from spin transfer effects with a 2 nm-thick MgO barrier. In consideration of the SAF free layer, we believe the change in HO is mainly induced by the VCEC.31–33 To be specific, the E-field can prominently modulate the interlayer exchange coupling strength within a SAF structure or even switch the sign of coupling (i.e., from antiferromagnetic to ferromagnetic or vice versa).31–33,40 As a result, HO can also be tuned or even reversed. In our study, the free layer-CoFeB is antiferromagnetically coupled to the free layer-Co/Pd multilayers, which is assumed to be magnetically stable in minor loops. As the coupling strength changes with E-field, HO changes subsequently. Figure 4(d) shows the HO values as a function of V. A slope of 27.5 Oe/V, which corresponds to a VCEC energy of roughly 1.23 merg/cm2 per V/nm, is obtained by assuming a linear V dependence of HO.

It is worth mentioning that, though the maximum TMR is about 18% for patterned devices, the current-in-plane tunneling (CIPT) measurement41 gives a TMR of about 34% as shown in supplementary material Fig. S3. The TMR decrease in patterned devices may come from the unoptimized nanofabrication process.42,43

In conclusion, we fabricated a SAF-pMTJ stack in which the industry-ready Co/Pd multilayers and mainstream CoFeB/MgO material system are combined. This film structure exhibited apparent PMA and antiferromagnetic coupling through a Ru/Ta spacer. FMR characterizations were conducted on the stack and revealed moderate Gilbert damping of the composite SAF free layer. The magnetization switching characteristics of SAF-pMTJs were investigated and discussed. We compared the HC and HO at different device sizes and reported the transition of the stable interlayer magnetic configuration from antiferromagnetic to ferromagnetic in smaller pMTJ diameters. To take advantage of the SAF configuration for ultrafast and ultralow energy switching, it is necessary to use a device size larger than the transition threshold (e.g., ∼400 nm in this study). We tested and found an unusually strong I dependence of HO, which can be explained by the proposed VCEC effect. The impacts of an electric bias, including STT, VCMA, and VCEC, should be taken into account in the design process of both SAF-pMTJ stacks and devices, as the device performance may change significantly under different current or voltages. These results shed light on the material selection and stack engineering of SAF-pMTJs and can facilitate the application of spintronic devices with ultrahigh switching speed and ultralow energy consumption.

See the supplementary material for the in-plane M-Hext loop, analysis on the E-field dependence of HC, and CIPT results.

This work was supported in part by the Defense Advanced Research Projects Agency (DARPA) via No. HR001117S0056-FP-042 “Advanced MTJs for computation in and near random access memory” and by the National Institute of Standards and Technology. Portions of this work were conducted in the Minnesota Nano Center, which was supported by the National Science Foundation through the National Nanotechnology Coordinated Infrastructure (NNCI) under Award No. ECCS-2025124.

The authors have no conflicts to disclose.

The data that support the findings of this study are available from the corresponding authors upon reasonable request.

1.
S.
Ikeda
,
K.
Miura
,
H.
Yamamoto
,
K.
Mizunuma
,
H. D.
Gan
,
M.
Endo
,
S.
Kanai
,
J.
Hayakawa
,
F.
Matsukura
, and
H.
Ohno
,
Nat. Mater.
9
,
721
724
(
2010
).
2.
P. K.
Amiri
,
Z. M.
Zeng
,
J.
Langer
,
H.
Zhao
,
G.
Rowlands
,
Y.-J.
Chen
,
I. N.
Krivorotov
,
J.-P.
Wang
,
H. W.
Jiang
,
J. A.
Katine
,
Y.
Huai
,
K.
Galatsis
, and
K. L.
Wang
,
Appl. Phys. Lett.
98
,
112507
(
2011
).
3.
D. C.
Worledge
,
G.
Hu
,
D. W.
Abraham
,
J. Z.
Sun
,
P. L.
Trouilloud
,
J.
Nowak
,
S.
Brown
,
M. C.
Gaidis
,
E. J.
O'Sullivan
, and
R. P.
Robertazzi
,
Appl. Phys. Lett.
98
,
022501
(
2011
).
4.
K. Y.
Camsari
,
A. Z.
Pervaiz
,
R.
Faria
,
E. E.
Marinero
, and
S.
Datta
,
IEEE Magn. Lett.
7
,
3107205
(
2016
).
5.
S.
Noh
,
D. H.
Kang
, and
M.
Shin
,
IEEE Trans. Magn.
55
,
3400705
(
2019
).
6.
K.
Yamada
,
K.
Kubota
, and
Y.
Nakatani
,
J. Appl. Phys.
127
,
133906
(
2020
).
7.
E.
Raymenants
,
O.
Bultynck
,
D.
Wan
,
T.
Devolder
,
K.
Garello
,
L.
Souriau
,
A.
Thiam
,
D.
Tsvetanova
,
Y.
Canvel
,
D. E.
Nikonov
,
I. A.
Young
,
M.
Heyns
,
B.
Soree
,
I.
Asselberghs
,
I.
Radu
,
S.
Couet
, and
V. D.
Nguyen
,
Nat. Electron.
4
,
392
398
(
2021
).
8.
C.
Yoshida
,
T.
Takenaga
,
Y.
Iba
,
Y.
Yamazaki
,
H.
Noshiro
,
K.
Tsunoda
,
A.
Hatada
,
M.
Nakabayashi
,
A.
Takahashi
,
M.
Aoki
, and
T.
Sugii
,
IEEE Trans. Magn.
49
,
4363
4366
(
2013
).
9.
T.
Devolder
,
E.
Liu
,
J.
Swerts
,
S.
Couet
,
T.
Lin
,
S.
Mertens
,
A.
Furnemont
,
G. S.
Kar
, and
J.
De Boeck
,
Appl. Phys. Lett.
109
,
142408
(
2016
).
10.
K.
Watanabe
,
B.
Jinnai
,
S.
Fukami
,
H.
Sato
, and
H.
Ohno
,
Nat. Commun.
9
,
663
(
2018
).
11.
B.
Jinnai
,
J.
Igarashi
,
K.
Watanabe
,
T.
Funatsu
,
H.
Sato
,
S.
Fukami
, and
H.
Ohno
, in
IEEE International Electron Devices Meeting (IEDM)
(
IEEE
,
2020
), pp.
24.6. 1
24.6. 4
.
12.
N.
Strelkov
,
A.
Chavent
,
A.
Timopheev
,
R. C.
Sousa
,
I. L.
Prejbeanu
,
L. D.
Buda-Prejbeanu
, and
B.
Dieny
,
Phys. Rev. B
98
,
214410
(
2018
).
13.
W.-G.
Wang
,
M.
Li
,
S.
Hageman
, and
C. L.
Chien
,
Nat. Mater.
11
,
64
68
(
2012
).
14.
Certain commercial equipment, instruments, or materials are identified in this paper in order to specify the experimental procedure adequately. Such identification is not intended to imply recommendation or endorsement by NIST, nor is it intended to imply that the materials or equipment identified are necessarily the best available for the purpose.
15.
P. J. H.
Bloemen
,
H. W.
van Kesteren
,
H. J. M.
Swagten
, and
W. J. M.
de Jonge
,
Phys. Rev. B
50
,
13505
(
1994
).
16.
X.
Liu
,
W.
Zhang
,
M. J.
Carter
, and
G.
Xiao
,
J. Appl. Phys.
110
,
033910
(
2011
).
17.
S.
Iihama
,
S.
Mizukami
,
H.
Naganuma
,
M.
Oogane
,
Y.
Ando
, and
T.
Miyazaki
,
Phys. Rev. B
89
,
174416
(
2014
).
18.
E. C. I.
Enobio
,
H.
Sato
,
S.
Fukami
,
F.
Matsukura
, and
H.
Ohno
,
IEEE Magn. Lett.
6
,
5700303
(
2015
).
19.
S.
Pal
,
B.
Rana
,
O.
Hellwig
,
T.
Thomson
, and
A.
Barman
,
Appl. Phys. Lett.
98
,
082501
(
2011
).
20.
T.
Kato
,
Y.
Matsumoto
,
S.
Kashima
,
S.
Okamoto
,
N.
Kikuchi
,
S.
Iwata
,
O.
Kitakami
, and
S.
Tsunashima
,
IEEE Trans. Magn.
48
,
3288
3291
(
2012
).
21.
K.
Yamamoto
,
T.
Matsuda
,
K.
Nishibayashi
,
Y.
Kitamoto
, and
H.
Munekata
,
IEEE Trans. Magn.
49
,
3155
3158
(
2013
).
22.
D. B.
Gopman
,
C. L.
Dennis
,
R. D.
McMichael
,
X.
Hao
,
Z.
Wang
,
X.
Wang
,
H.
Gan
,
Y.
Zhou
,
J.
Zhang
, and
Y.
Huai
,
AIP Adv.
7
,
055932
(
2017
).
23.
J. M.
Shaw
,
H. T.
Nembach
, and
T. J.
Silva
,
J. Appl. Phys.
108
,
093922
(
2010
).
24.
J. M.
Shaw
,
H. T.
Nembach
, and
T. J.
Silva
,
Phys. Rev. B
85
,
054412
(
2012
).
25.
E.
Liu
,
A.
Vaysset
,
J.
Swerts
,
T.
Devolder
,
S.
Couet
,
S.
Mertens
,
T.
Lin
,
S.
Van Elshocht
,
J. De
Boeck
, and
G. S.
Kar
,
IEEE Trans. Magn.
53
,
3401305
(
2017
).
26.
Y.
H.
Wang
,
S.-H.
Huang
,
D.-Y.
Wang
,
K.-H.
Shen
,
C.-W.
Chien
,
K.-M.
Kuo
,
S.-Y.
Yang
, and
D.-L.
Deng
, in
International Electron Devices Meeting (IEDM)
(
IEEE
,
2012
), pp.
29.2.1
29.2.4
.
27.
S.
Bandiera
,
R. C.
Sousa
,
Y.
Dahmane
,
C.
Ducruet
,
C.
Portemont
,
V.
Baltz
,
S.
Auffret
,
I. L.
Prejbeanu
, and
B.
Dieny
,
IEEE Magn. Lett.
1
,
3000204
(
2010
).
28.
K.
Yakushiji
,
A.
Fukushima
,
H.
Kubota
,
M.
Konoto
, and
S.
Yuasa
,
Appl. Phys. Express
6
,
113006
(
2013
).
29.
S.
Jenkins
,
A.
Meo
,
L. E.
Elliott
,
S. K.
Piotrowski
,
M.
Bapna
,
R. W.
Chantrell
,
S. A.
Majetich
, and
R. F. L.
Evans
,
J. Phys. D: Appl. Phys.
53
,
044001
(
2020
).
30.
J.
Huang
,
C. H.
Sim
,
V. B.
Naik
,
M.
Tran
,
S. T.
Lim
, and
G.
Han
,
J. Appl. Phys.
117
,
17B721
(
2015
).
31.
D.
Zhang
,
M.
Bapna
,
W.
Jiang
,
D. P.
de Sousa
,
Y.-C.
Liao
,
Z.
Zhao
,
Y.
Lv
,
A.
Naeemi
,
T.
Low
,
S. A.
Majetich
, and
J.-P.
Wang
, arXiv:1912.10289 (
2019
).
32.
J.-P.
Wang
,
D.
Zhang
,
S. A.
Majetich
, and
M.
Bapna
,
U.S. patent 10,586,579
(March 10,
2020
).
33.
J.-P.
Wang
,
D.
Zhang
, and
P.
Sahu
,
U.S. patent 11,183,227
(November 23,
2021
).
34.
C.
Safranski
and
J. Z.
Sun
,
Phys. Rev. B
100
,
014435
(
2019
).
35.
G.
Han
,
K.
Zhang
,
X.
Yang
,
L.
Ai
,
M.
Wang
, and
B.
Liu
,
J. Magn. Magn. Mater.
527
,
167789
(
2021
).
36.
W.
Stefanowicz
,
L. E.
Nistor
,
S.
Pizzini
,
W.
Kuch
,
L. D.
Buda-Prejbeanu
,
G.
Gaudin
,
S.
Auffret
,
B.
Rodmacq
, and
J.
Vogel
,
Appl. Phys. Lett.
104
,
012404
(
2014
).
37.
V.
Krizakova
,
E.
Grimaldi
,
K.
Garello
,
G.
Sala
,
S.
Couet
,
G. S.
Kar
, and
P.
Gambardella
,
Phys. Rev. Appl.
15
,
054055
(
2021
).
38.
H.
Meng
,
V. B.
Naik
,
R.
Liu
, and
G.
Han
,
Appl. Phys. Lett.
105
,
042410
(
2014
).
39.
G.
Mihajlović
,
N.
Smith
,
T.
Santos
,
J.
Li
,
M.
Tran
,
M.
Carey
,
B. D.
Terris
, and
J. A.
Katine
,
Phys. Rev. Appl.
13
,
024004
(
2020
).
40.
T.
Newhouse-Illige
,
Y.
Liu
,
M.
Xu
,
D. R.
Hickey
,
A.
Kundu
,
H.
Almasi
,
C.
Bi
,
X.
Wang
,
J. W.
Freeland
,
D. J.
Keavney
,
C. J.
Sun
,
Y. H.
Xu
,
M.
Rosales
,
X. M.
Cheng
,
S.
Zhang
,
K. A.
Mkhoyan
, and
W. G.
Wang
,
Nat. Commun.
8
,
15232
(
2017
).
41.
D. C.
Worledge
and
P. L.
Trouilloud
,
Appl. Phys. Lett.
83
,
84
86
(
2003
).
42.
S.-W.
Chun
,
D.
Kim
,
J.
Kwon
,
B.
Kim
, and
H.
Lee
,
J. Vac. Sci. Technol. B
30
,
06FA01
(
2012
).
43.
S. V.
Beek
,
S.
Rao
,
S.
Kundu
,
W.
Kim
,
B. J.
O'Sullivan
,
S.
Cosemans
,
F.
Yasin
,
R.
Carpenter
,
S.
Couet
,
S. H.
Sharifi
,
N.
Jossart
,
D.
Crotti
, and
G.
Kar
, in
IEEE International Reliability Physics Symposium (IRPS)
(
IEEE
,
2021
), pp.
1
5
.

Supplementary Material