Field-free switching of magnetic tunnel junctions driven by spin-orbit torques at sub-ns timescales
Viola Krizakova, Kevin Garello, Eva Grimaldi, Gouri Sankar Kar, Pietro Gambardella
FField-free switching of magnetic tunnel junctions driven by spin-orbittorques at sub-ns timescales
Viola Krizakova, a) Kevin Garello, Eva Grimaldi, Gouri Sankar Kar, and Pietro Gambardella b) Department of Materials, ETH Zurich, 8093 Zurich, Switzerland. imec, Kapeldreef 75, 3001 Leuven, Belgium (Dated: 12 June 2020) We report time-resolved measurements of magnetization switching by spin-orbit torques in the absence of an externalmagnetic field in perpendicularly magnetized magnetic tunnel junctions (MTJ). Field-free switching is enabled by thedipolar field of an in-plane magnetized layer integrated above the MTJ stack, the orientation of which determinesthe switching polarity. Real-time single-shot measurements provide direct evidence of magnetization reversal andswitching distributions. Close to the critical switching voltage we observe stochastic reversal events due to a finiteincubation delay preceding the magnetization reversal. Upon increasing the pulse amplitude to twice the critical voltagethe reversal becomes quasi-deterministic, leading to reliable bipolar switching at sub-ns timescales in zero externalfield. We further investigate the switching probability as a function of dc bias of the MTJ and external magnetic field,providing insight on the parameters that determine the critical switching voltage.Current-induced spin-orbit torques (SOT) allow for ma-nipulating the magnetization of diverse classes of mag-netic materials and devices . Recent studies on ferromag-netic nanodots and magnetic tunnel junctions (MTJ) haveshown that SOT-induced switching can overcome spin trans-fer torque (STT) switching in terms of speed, reliability, andendurance . Fast and reliable deterministic switching ofMTJs is especially important for the development of non-volatile magnetic random access memories . However,when switching a perpendicular magnetization by SOT, astatic in-plane magnetic field is required to break the torquesymmetry, which otherwise does not discriminate betweenup and down magnetic states . As such field is detrimen-tal for memory applications, various approaches have beenproposed to achieve field-free SOT switching, including ex-change coupling to an antiferromagnet , RKKY andDzyaloshinskii-Moriya coupling to a reference ferromagnet,tilted magnetic anisotropy , geometrical asymmetry ,and two-pulse schemes . An alternative approach, com-patible with back-end-of-line integration of MTJs on CMOSwafers, is based on embedding a ferromagnet in the hard maskthat is used to pattern the SOT current line . In such devices,the rectangular shape of the magnetic hard mask (MHM) pro-vides strong shape anisotropy along the current direction, thusgenerating an in-plane field on the free layer without impos-ing additional overheads on the processing and power require-ments of the MTJs [Fig. 1(a)].In this letter, we report on real-time measurements of field-free magnetization switching in three-terminal MTJs includ-ing a MHM. Whereas the switching dynamics has been exten-sively studied in two-terminal MTJs operated by STT ,there are only few studies addressing the transient dynam-ics and real-time reversal speed of individual SOT-inducedswitching events in three-terminal MTJs . In particular,the SOT-induced dynamics, reversal speed, and critical volt- a) Electronic mail: [email protected] b) Electronic mail: [email protected] pulse scopeMTJ
50 Ω+25 dB V SOT V dc (b) (d) (a) (c) V MTJ
150 nm
SOT-line
MTJ
MHM V SOT H MHM V MTJ V AP -V P ∆ t V o l t age ( m V ) Time V sw t FIG. 1. (a) Schematic and (b) TEM cross-sectional view of a field-free switching MTJ device. The field H MHM produced by the MHMis represented by gray arrows. (c) Simplified schematic of the mea-surement circuit for after-pulse and real-time detection of SOT and/orSTT switching. (d) Voltage difference between P and AP states mea-sured by the scope and averaged over 300 switching events (blackline). Single-shot switching voltage trace (blue line) prior to normal-ization (see text), and definition of the characteristic times t and ∆ t .The positive direction of the bias voltage V MTJ is indicated in (a) and(c). age in MTJs with perpendicular magnetization have not beeninvestigated in the absence of an external magnetic field.Our devices are top-pinned MTJs with 108% tunnelingmagnetoresistance (TMR) patterned into a circular pillar witha diameter of 80 nm. The pillars are grown on top of a 190 nmwide β -W SOT-current line with resistivity of 160 µ Ω cm andresistance of 370 Ω . The MTJ is formed by free and refer-ence layers made of CoFeB with a thicknesses of 0.9 and1 nm, respectively, separated by an MgO tunnel barrier withresistance-area product of 20 Ω µ m , as shown in Fig. 1(a).The reference layer (yellow) is pinned to a synthetic anti-ferromagnet (black, SAF). The reference layer and the SAFgenerate an out-of-plane dipolar field ≈
13 mT that favors a r X i v : . [ c ond - m a t . m e s - h a ll ] J un V s w ( no r m . ) Time (ns) V s w ( no r m . ) Time (ns) -571 mV-515 mV-460 mV-404 mV C oun t s t (ns) D t (ns) mV C oun t s t (ns) D t (ns) P-APAP-P 430 490 5500246810 P-AP AP-P t ( n s )
430 490 550 / D t ( n s - ) | V SOT | (mV) (d) (f) (a)(b) (c) (e)
FIG. 2. Representative single-shot time traces for different pulse amplitudes of V SOT acquired at zero external field for (a) P-AP and (b) AP-Pswitching at V MTJ = -0.5 V SOT . The characteristic times t and ∆ t are extracted by fitting the data to a linear ramp (black lines). (c) Statisticaldistribution of t and ∆ t for P-AP and (d) AP-P switching. (e) Dependence of t and (f) 1 / ∆ t on V SOT extracted from the distributions. Symbolsdenote the median values, shaded areas represent the lower and upper quartiles of the distributions. The bottom curves in (e) show t measuredwith V MTJ = − . V SOT . The dashed line in (f) is a linear fit to 1 / ∆ t . the anti-parallel (AP) over the parallel (P) state of the MTJ.The hard mask used to pattern the SOT line incorporates a50 nm thick Co magnet , which provides an in-plane field( µ H MHM ≈
40 mT) parallel to the SOT line [Figs. 1(a) and(b)]. Due to its high aspect ratio of 130 ×
410 nm , the magne-tization direction of the Co hard mask remains constant aftersaturation. Moreover, the magnetization of the hard mask isnot influenced by pulsing the current either through the SOTline or the MTJ pillar.Electrical measurements are performed using a setup thatcombines real-time and after-pulse readout of the MTJ resis-tance, as depicted in Fig. 1(c) and reported in more detail inRef. 16. After-pulse switching measurements consist of aninitialization pulse that sets the free layer magnetization in thedesired state, which is verified by a dc measurement of theMTJ resistance, and a switching pulse, also followed by a dcresistance measurement. In the time-resolved measurements,a driving voltage supplied by a pulse generator with 0.15 nsrise time is split in two pulses with a well-defined amplitudethat are simultaneously fed to the input electrodes of the three-terminal MTJ device. The pulse applied to the bottom elec-trode ( V SOT ) drives the SOT reversal. Together, the pulse ap-plied to the top electrode and V SOT determine the potentialdifference across the MTJ pillar ( V MTJ ). By adjusting the ratiobetween these pulse amplitudes, we control the instantaneousvalue of V MTJ , which allows for studying phenomena emerg-ing from the bias during SOT-driven reversal, such as volt-age control of magnetic anisotropy (VCMA) as well as STTswitching . In this work, we restrict ourselves to switching at zero, low, and strong bias ( V MTJ = 0, -0.5 and -1.65 V SOT ). Thepulse transmitted through the device is amplified and acquiredon a 20 GHz sampling oscilloscope, which allows for monitor-ing the MTJ resistance in real-time in a time window definedby the width of the driving pulse. To facilitate the analysis ofdifferent switching events, a reference trace is subtracted toeach voltage trace recorded during a switching pulse. The ref-erence trace is obtained by maintaining the MTJ in its initialstate, either P or AP, by application of an external magneticfield opposing H MHM . The resultant trace is then divided bythe voltage difference between the P and AP states [Fig. 1(d)]in order to obtain the normalized switching signal V sw .Time-resolved studies of current-induced switching can beperformed either by pump-probe measurements , whichyield the average magnetization dynamics, or single-shotmeasurements . Although average time-resolved mea-surements provide information on the reproducible dynamicbehavior of the magnetization and afford a higher signal-to-noise ratio compared to single-shot measurements, stochas-tic dynamical processes can only be revealed in studies car-ried out on individual switching events. In this study we per-form both types of measurements to highlight different as-pects of the reversal dynamics. In order to provide a measur-able tunneling magnetoresistance reading in single-shot mea-surements, we apply a small voltage bias V MTJ on the MTJto allow for current flow through the pillar. Depending onthe sign of V MTJ relative to V SOT , this bias can either assist orhinder the SOT switching. Two effects are induced by V MTJ ,namely the STT and VCMA. As the sign of the STT dependson the orientation of the reference layer and the VCMA doesnot, these two effects can be disentangled from each other .Here, we set the sign of V MTJ such that STT always opposesSOT switching for the chosen orientation of the referencelayer and H MHM . We thus focus primarily on SOT-inducedswitching, unlike previous work in which STT was used topromote switching . Moreover, in our configuration, theVCMA balances the effect of the SAF dipolar field.Figures 2(a) and 2(b) show representative time traces of in-dividual P-AP and AP-P switching events obtained in zeroexternal field for different pulse amplitudes of V SOT and V MTJ = -0.5 V SOT , which corresponds to a current density j MTJ <
30% of the critical STT switching current. The switch-ing traces reveal that the reversal of the free layer starts aftera finite incubation time ( t ) followed by a single jump of theresistance during a relatively short transition time ( ∆ t ), afterwhich the magnetization remains quiescent in the final stateuntil the pulse ends. The reversal dynamics is thus qualita-tively similar to that observed in the presence of an externalmagnetic field . Accordingly, we attribute t to the time re-quired to nucleate a reversed domain and ∆ t to the time topropagate a domain wall across the free layer . Noise inthe time traces noticeably increases upon increasing V SOT andwith the time elapsed from the pulse onset, which we asso-ciate with the rise of the device temperature during the pulse.Each reversal trace is fit by a linear ramp and the character-istic times t and ∆ t , defined in Fig. 1(d), are extracted fromthe line breakpoints. The distributions of t and ∆ t represent-ing statistics over 200 single-shot measurements are plotted inFigs. 2(c) and 2(d). At low pulse amplitude, typical values of t significantly exceed ∆ t . However, both the center and thewidth of the t distribution can be reduced by more than oneorder of magnitude by increasing V SOT .To investigate the characteristic times in more detail, weextract the median as well as the lower and upper quartiles ofeach distribution and plot them as a function of | V SOT | . Fig-ure 2(e) shows that the median t decreases to below 1 ns forboth AP-P and P-AP switching when increasing V SOT from400 to 560 mV. At the lowest pulse amplitudes, the median t of the AP-P switching configuration is twice as long com-pared to P-AP switching. This asymmetry, which is attributedto the dipolar field of the SAF, gradually reduces upon increas-ing V SOT . Additionally, our measurements show that suchan asymmetry can be strengthened or eliminated by tuning V MTJ . To demonstrate the potential of SOT-driven switchingat increased V MTJ , we report in the same plot t obtained at V MTJ = -1.65 V SOT (bottom curve). In this case, the differ-ence between both configurations is minimized by VCMA,which favors AP-P at the expense of P-AP switching, whereasthe median value and its dispersion are reduced by the bias-induced temperature rise in the device. Note that STT has thesame hindering effect on both switching configurations, as itfavors the orientation of the free layer opposite to the finalstate defined by V SOT . Moreover, despite the presence of astrong opposing STT when V MTJ = -1.65 V SOT , which corre-sponds to j MTJ <
96% of the critical STT switching current inthe absence of SOT, we do not observe writing errors withinour data set.
Time (ns) V c V c V s w ( no r m . ) Time (ns) V c V c FIG. 3. Averaged time traces of AP-P switching for 1.5-ns (left) and15-ns-long (right) pulses at different V SOT pulse amplitudes close to V c acquired at V MTJ = . V SOT and zero external field. P-AP switch-ing traces (not shown) yield similar results.
In contrast to t , ∆ t has a much weaker dependence on V SOT .Figure 2(f) shows that 1 / ∆ t increases linearly with V SOT , witha moderate slope of 1.5 ns − V − . Linear scaling with currentis indeed expected for the speed of SOT-driven domain wallsin the flow regime . Supposing that the reversal initiateswith the nucleation of a domain wall at one edge of the freelayer, as shown in previous work , our data imply an aver-age domain-wall propagation speed of 100 m s − induced byan SOT current density of 1 . × A m − at V SOT = 0.48 V.To demonstrate the reliability of switching for repeatedevents, we performed time-resolved measurements averagedover 1000 switching trials, as shown in Fig. 3. Averaging theacquired waveforms allows us to decrease the bias down to V MTJ = 0.1 V SOT , which corresponds to a current density j MTJ that is <
15% of the STT switching threshold. The time tracescompare AP-P switching for different values of V SOT given inmultiples of the critical switching voltage ( V c ), correspondingto 50% switching probability. Each trace comprises an initialdelay and a smooth transition part without noticeable inter-mediate states. Shortening of the delay and transition part ofthe averaged time traces indicates that the switching processchanges from stochastic to almost deterministic upon increas-ing V SOT . We further observe a striking reduction of the totalswitching time from 15 ns to less than 1 ns at V SOT ≥ . V c ,which can be interpreted as switching at sub-ns timescale inthe great majority of the 1000 trials, with the confidence givenby the signal-to-noise ratio.Our measurements also allow us to compare the criticalswitching voltage obtained by after-pulse, single-shot, and av-eraged time-resolved measurements. This comparison is rel-evant to assess the reliability of different methods employedto measure V c , particularly for the more common after-pulseresistance measurements, in which t or ∆ t cannot be ac-cessed. We define the critical switching time ( t c ) as the pulsewidth corresponding to 50% switching probability in after-pulse measurements, and as t + ∆ t / V sw = 0.5, in single-shot and averaged time-resolvedmeasurements, respectively. Likewise, we define V c as thecorresponding SOT pulse amplitude. Figure 4(a) shows that,irrespective of the measurement method, all values of V c fallon the same curve and scale inversely with the critical time t c for pulses shorter than 4 ns. Such a scaling is expected forthe intrinsic regime, in which conservation of angular momen-tum gives 1 / t c ∝ ( V c − V c0 ) . Here V c0 is the intrinsiccritical voltage that reflects the minimum amount of angu-lar momentum required to achieve switching in the absenceof thermal effects. A linear fit of the data in Fig. 4(a) gives V c0 = 390 mV, which corresponds to an intrinsic critical cur-rent of 1 . × A m − . For pulses longer than 4 ns, devi-ations from the linear behavior are attributed to the onset ofthermally-activated switching .Figure 4(b) shows that in both switching configurations, V c reduces considerably upon increasing V MTJ . The reductionof V c is largest for the shorter pulses and for AP-P switch-ing (bottom panel) compared to P-AP switching (top panel).These observations can be explained by the combined impactof VCMA and heat generated by the bias. V MTJ > V SOT is negative (AP-P switching), both VCMA and temperature lead to a reduc-tion of V c ; however, when V SOT is positive (P-AP switching),the VCMA opposes the temperature-induced decrease of theswitching barrier, resulting in a smaller reduction of V c . Con-sequently, V c in both switching configurations equalizes for V MTJ ≈ − . V SOT . This result shows that V MTJ can be ef-ficiently used to realize symmetric switching conditions [seealso Fig. 2(e)].Last, we address the functionality of our devices underan external field ( H x ) by measuring the after-pulse switchingprobability ( P sw ) for different pulse amplitude and field con-ditions. We start by considering SOT switching at V MTJ = 0. | V c | ( V ) V MTJ / V SOT
0 -0.5 -1.4 -1.9
P-APAP-P | V c | ( V ) t c (ns -1 )0.40.50.60.70.8 t c (ns -1 ) | V c | ( V ) Measurement: after-pulse ave. time res. single-shot (a) (b)
FIG. 4. (a) V c obtained from after-pulse probability (squares),real-time single-shot measurements (diamonds), and averaged time-resolved measurements (circles). The gray line represents a fit tothe data by a 1 / t c function for t c < V c for SOT-dominatedswitching at different values of V MTJ obtained from after-pulse prob-ability measurements for P-AP (top panel) and AP-P (bottom panel)switching. H x H MHM P s w ( AP - P ) P s w ( P - AP ) -120 -60 0 60 120-0.9-0.6-0.30.00.30.60.9 V S O T ( V ) m H x (mT) P AP P AP FIG. 5. SOT switching probability as a function of V MTJ and µ H x for 0.5 ns-long pulses at V MTJ =
0. The color of each point repre-sents the after-pulse switching probability P sw (out of 50 trials). Themagnetization of the MHM was initially set to be negative (positive)for H x < > H MHM > <
0) as indicated by thearrows above the diagram. The AP/P labels denote the final state.
Each point in the switching phase diagram in Fig. 5 illus-trates the statistical result of 50 trials for a fixed pulse width(0.5 ns). The black boundary defines P sw = 0.5, which dividesthe diagrams into under- (gray) and over-critical (blue and yel-low) regions. Since H x polarizes the MHM, which has a co-ercivity of about 20 mT, H MHM is always antiparallel to H x [Fig. 1(a)]. As a consequence, the switching polarity dependson the sign of | H MHM − H x | , i.e., of the total in-plane fieldacting on the free layer. The diagram also allows for evalu-ating the strength of H MHM from the difference between two H x values resulting in the same V c . In this manner, we esti-mate that µ H MHM ≈
40 mT. The diagram shows that bipolarswitching is possible in a wide range of external fields withthe exception of narrow intervals, in which | H MHM − H x | ap-proaches zero. A typical transition of P sw from 0.01 to 0.99occurs upon increasing V SOT by less than 80 mV, in contrastwith STT switching, for which a 150 mV increase of V MTJ isrequired using the same device and 15 ns-long pulses.Before concluding, we discuss a few possibilities to im-prove the switching speed and the design of field-free SOTdevices. As shown in Ref. 16, increasing the magnitude of thein-plane field significantly reduces the switching time for agiven V SOT . In general, H MHM can be increased by i) optimiz-ing the aspect ratio and thickness of the magnetic layer, ii) re-placing the Co layer by a material with higher saturation mag-netization, such as CoFe, and iii) bringing the MHM closerto the MTJ. The optimal strength of H MHM will ultimately de-pend on the critical current, switching rate, and thermal stabil-ity of the free layer set by the target application. Static mea-surements show that the TMR is not affected by the MHMand that the overall device properties are more influenced bythe design of the MTJ pillar rather than by the hard maskitself . Better compensation of the out-of-plane stray fieldproduced by the SAF would lead to a more symmetric switch-ing behavior between the P and AP configurations. This canbe achieved, e.g., by changing the thickness of the referencelayer or the thickness and number of repetitions of the SAFmultilayer. Alternatively, as shown here, V MTJ can be used tobalance the SAF field. The MHM approach is also compati-ble with dense designs, as shown in Ref. 32. Micromagneticsimulations further show that surrounding magnets have a sta-bilizing effect on the magnetization of the hard mask. MHMsdown-scaled to a volume of 50 × ×
25 nm and a pitchof 100 ×
150 nm have more uniform magnetization patternsthan MHMs with a volume of 110 × ×
50 nm and a pitchof 260 ×
540 nm . Future studies might establish if the MHMapproach based on a single mask per MTJ is compatible withsharing the same SOT write line between multiple MTJs, asproposed in Ref. 15.In summary, we have demonstrated field-free switchingof perpendicularly magnetized MTJs by SOT in real time.Single-shot time-resolved measurements show that thestochastic incubation delay near the critical voltage threshold( V SOT ≈ V c ) is several ns long for both the AP-P and P-APswitching configurations, whereas the actual transition timeis about 1 ns. Upon increasing V SOT or V MTJ , the switchingdistributions narrow down leading to reduced latency andquasi-deterministic switching. Averaged time-resolved mea-surements show that the total switching time can be reducedto 0.7 ns by increasing V SOT up to 1 . V c with negligibleassistance of either STT or VCMA. At timescales shorterthan 4 ns, the critical switching voltage is found to scalelinearly with inverse of the switching time, as expected inthe intrinsic regime. Real-time measurements and after-pulseswitching statistics as a function of pulse length are found toprovide a consistent estimate of the critical switching time.Measurements of the switching probability as a function of V MTJ and external field indicate that further improvements ofthe switching dynamics and reduction of V c can be obtainedby VCMA, increase of the dipolar field of the MHM, andcompensating the dipolar field due to the SAF.This research was supported by the Swiss National ScienceFoundation (Grant no. 200020-172775), the Swiss Govern-ment Excellence Scholarship (ESKAS-Nr. 2018.0056), theETH Zurich (Career Seed Grant SEED-14 16-2) and imec’sIndustrial Affiliation Program on MRAM devices.This article may be downloaded for personal use only.Any other use requires prior permission of the authorand AIP Publishing. This article appeared in Appl.Phys. Lett. , 232406 (2020) and may be found athttps://doi.org/10.1063/5.0011433. A. Manchon, J. Železný, I. M. Miron, T. Jungwirth, J. Sinova, A. Thiaville,K. Garello, and P. Gambardella, Rev. Mod. Phys. , 035004 (2019). I. M. Miron, K. Garello, G. Gaudin, P. J. Zermatten, M. V. Costache, S. Auf-fret, S. Bandiera, B. Rodmacq, A. Schuhl, and P. Gambardella, Nature ,189 (2011). L. Liu, C.-F. Pai, Y. Li, H. W. Tseng, D. C. Ralph, and R. A. Buhrman,Science , 555 (2012). M. Cubukcu, O. Boulle, M. Drouard, K. Garello, C. Onur Avci, I. MihaiMiron, J. Langer, B. Ocker, P. Gambardella, and G. Gaudin, Appl. Phys.Lett. , 042406 (2014). S. Fukami, T. Anekawa, C. Zhang, and H. Ohno, Nat. Nanotechnol. ,621 (2016). S. W. Lee and K. J. Lee, Proc. IEEE , 1831 (2016). Z. Luo, A. Hrabec, T. P. Dao, G. Sala, S. Finizio, J. Feng, S. Mayr, J. Raabe,P. Gambardella, and L. J. Heyderman, Nature , 214 (2020). K. Garello, C. O. Avci, I. M. Miron, M. Baumgartner, A. Ghosh, S. Auffret,O. Boulle, G. Gaudin, and P. Gambardella, Appl. Phys. Lett. , 212402(2014). C. Zhang, S. Fukami, H. Sato, F. Matsukura, and H. Ohno, Appl. Phys.Lett. , 1 (2015). S. V. Aradhya, G. E. Rowlands, J. Oh, D. C. Ralph, and R. A. Buhrman,Nano Lett. , 5987 (2016). M. Baumgartner, K. Garello, J. Mendil, C. O. Avci, E. Grimaldi, C. Murer,J. Feng, M. Gabureac, C. Stamm, Y. Acremann, S. Finizio, S. Wintz,J. Raabe, and P. Gambardella, Nat. Nanotechnol. , 980 (2017). J. Yoon, S. W. Lee, J. H. Kwon, J. M. Lee, J. Son, X. Qiu, K. J. Lee, andH. Yang, Sci. Adv. , e1603099 (2017). M. M. Decker, M. S. Wörnle, A. Meisinger, M. Vogel, H. S. Körner, G. Y.Shi, C. Song, M. Kronseder, and C. H. Back, Phys. Rev. Lett. , 257201(2017). M. Cubukcu, O. Boulle, N. Mikuszeit, C. Hamelin, T. Brächer, N. Lamard,M.-C. Cyrille, L. Buda-Prejbeanu, K. Garello, I. M. Miron, O. Klein,G. D. Loubens, V. V. Naletov, J. Langer, B. Ocker, P. Gambardella, andG. Gaudin, IEEE Trans. Magn. , 9300204 (2018). Y. Kato, Y. Saito, H. Yoda, T. Inokuchi, S. Shirotori, N. Shimo-mura, S. Oikawa, A. Tiwari, M. Ishikawa, M. Shimizu, B. Altansargai,H. Sugiyama, K. Koi, Y. Ohsawa, and A. Kurobe, Phys. Rev. Appl. ,044011 (2018). E. Grimaldi, V. Krizakova, G. Sala, F. Yasin, S. Couet, G. S. Kar,K. Garello, and P. Gambardella, Nat. Nanotechnol. , 111 (2020). A. D. Kent and D. C. Worledge, Nat. Nanotechnol. , 187 (2015). D. Apalkov, B. Dieny, and J. M. Slaughter, Proc. IEEE , 1796 (2016). G. Prenat, K. Jabeur, P. Vanhauwaert, G. D. Pendina, F. Oboril, R. Bishnoi,M. Ebrahimi, N. Lamard, O. Boulle, K. Garello, J. Langer, B. Ocker, M. C.Cyrille, P. Gambardella, M. Tahoori, and G. Gaudin, IEEE Trans. Multi-Scale Comput. Syst. , 49 (2016). S. Fukami, C. Zhang, S. Duttagupta, A. Kurenkov, and H. Ohno, Nat.Mater. , 535 (2016). Y. W. Oh, S. H. C. Baek, Y. M. Kim, H. Y. Lee, K. D. Lee, C. G. Yang,E. S. Park, K. S. Lee, K. W. Kim, G. Go, J. R. Jeong, B. C. Min, H. W. Lee,K. J. Lee, and B. G. Park, Nat. Nanotechnol. , 878 (2016). A. van den Brink, G. Vermijs, A. Solignac, J. Koo, J. T. Kohlhepp, H. J.Swagten, and B. Koopmans, Nat. Commun. , 10854 (2016). Y. C. Lau, D. Betto, K. Rode, J. M. Coey, and P. Stamenov, Nat. Nanotech-nol. , 758 (2016). Z. Luo, T. P. Dao, A. Hrabec, J. Vijayakumar, A. Kleibert, M. Baumgartner,E. Kirk, J. Cui, T. Savchenko, G. Krishnaswamy, L. J. Heyderman, andP. Gambardella, Science , 1435 (2019). L. You, O. J. Lee, D. Bhowmik, D. Labanowski, J. Hong, J. Bokor, andS. Salahuddin, Proc. Natl. Acad. Sci. U. S. A. , 10310 (2015). G. Yu, P. Upadhyaya, Y. Fan, J. G. Alzate, W. Jiang, K. L. Wong,S. Takei, S. A. Bender, L. T. Chang, Y. Jiang, M. Lang, J. Tang, Y. Wang,Y. Tserkovnyak, P. K. Amiri, and K. L. Wang, Nat. Nanotechnol. , 548(2014). C. K. Safeer, E. Jué, A. Lopez, L. Buda-Prejbeanu, S. Auffret, S. Pizzini,O. Boulle, I. M. Miron, and G. Gaudin, Nat. Nanotechnol. , 143 (2016). T. Y. Chen, H. I. Chan, W. B. Liao, and C. F. Pai, Phys. Rev. Appl. ,044038 (2018). A. van den Brink, S. Cosemans, S. Cornelissen, M. Manfrini, A. Vaysset,W. Van Roy, T. Min, H. J. Swagten, and B. Koopmans, Appl. Phys. Lett. , 012403 (2014). M. Wang, W. Cai, D. Zhu, J. Kan, Z. Zhao, K. Cao, Z. Wang, Z. Wang,Y. Zhang, T. Zhang, C. Park, J.-P. Wang, A. Fert, and W. Zhao, Nat. Elec-tron. , 582 (2018). R. L. de Orio, A. Makarov, S. Selberherr, W. Goes, J. Ender, S. Fiorentini,and V. Sverdlov, Solid State Electron. , 107730 (2019). K. Garello, F. Yasin, H. Hody, S. Couet, L. Souriau, S. H. Sharifi, J. Swerts,R. Carpenter, S. Rao, W. Kim, J. Wu, K. Sethu, M. Pak, N. Jossart,D. Crotti, A. Furnemont, and G. S. Kar, IEEE Symp. VLSI Technol. ,T194 (2019). T. Devolder, J. Hayakawa, K. Ito, H. Takahashi, S. Ikeda, P. Crozat, N. Zer-ounian, J. V. Kim, C. Chappert, and H. Ohno, Phys. Rev. Lett. , 057206 (2008). H. Tomita, K. Konishi, T. Nozaki, H. Kubota, A. Fukushima, K. Yakushiji,S. Yuasa, T. Shinjo, Y. Nakatani, M. Shiraishi, and Y. Suzuki, Appl. Phys.Express , 061303 (2008). Y.-T. Cui, G. Finocchio, C. Wang, J. A. Katine, R. A. Buhrman, and D. C.Ralph, Phys. Rev. Lett. , 097201 (2010). C. Hahn, G. Wolf, B. Kardasz, S. Watts, M. Pinarbasi, and A. D. Kent,Phys. Rev. B , 214432 (2016). T. Devolder, A. Le Goff, and V. Nikitin, Phys. Rev. B , 224432 (2016). T. Devolder, J. V. Kim, F. Garcia-Sanchez, J. Swerts, W. Kim, S. Couet,G. Kar, and A. Furnemont, Phys. Rev. B , 024420 (2016). T. Inokuchi, H. Yoda, K. Koi, N. Shimomura, Y. Ohsawa, Y. Kato, S. Shiro-tori, M. Shimizu, H. Sugiyama, S. Oikawa, B. Altansargai, and A. Kurobe,Appl. Phys. Lett. , 192404 (2019). I. M. Miron, T. Moore, H. Szambolics, L. D. Buda-prejbeanu, S. Auffret,B. Rodmacq, S. Pizzini, J. Vogel, M. Bonfim, A. Schuhl, and G. Gaudin,Nat. Mater. , 419 (2011). A. Thiaville, S. Rohart, É. Jué, V. Cros, and A. Fert, Europhys. Lett. ,57002 (2012). S. Emori, U. Bauer, S.-M. Ahn, E. Martinez, and G. S. D. Beach, Nat.Mater. , 611 (2013). E. Martinez, S. Emori, G. S. D. Beach, E. Martinez, S. Emori, and G. S. D.Beach, Appl. Phys. Lett. , 072406 (2013). N. Mikuszeit, O. Boulle, I. M. Miron, K. Garello, P. Gambardella,G. Gaudin, and L. D. Buda-Prejbeanu, Phys. Rev. B , 144424 (2015). D. Bedau, H. Liu, J. Z. Sun, J. A. Katine, E. E. Fullerton, S. Mangin, andA. D. Kent, Appl. Phys. Lett. , 262502 (2010). H. Liu, D. Bedau, J. Z. Sun, S. Mangin, E. E. Fullerton, J. A. Katine, andA. D. Kent, J. Magn. Magn. Mater.358-359