Magnetic tunnel junctions (MTJs) with perpendicular magnetic anisotropy (PMA) attract much interest since they allow to scale down the dimension of spintronic devices to the latest technological node (<45 nm), while keeping a sufficient thermal stability (*E*/*k*_{B} *T* > 60) and lowering the current density needed to switch the free layer (FL) by spin transfer torque [Mangin 2006]. It is also pointed out that circular pillars can be used instead of elliptical ones in the case of MTJs with PMA (p-MTJs), since shape anisotropy is not required for stability. This allows simplifying the fabrication process and gives the maximum areal bit density. Currently, p-MTJs use a wide variety of perpendicular materials such as Co/Pt [Carvello 2008] or CoFe/Pd [Mizunuma 2009] multilayers, L1_{0}FePt ordered alloys [Yoshikawa 2008], rare earth elements-transition metal (RE-TM) ferrimagnets [Nakayama 2008] or Pt/Co/oxide stacks [Nistor 2009]. In these systems, the reference layer consists of a single hard magnetic layer. However, except for the case of compensated RE-TM ferrimagnets, dipolar interaction between the two magnetic electrodes results in a highly asymmetrical reversal of the FL with respect to the applied field or to the current density. Moreover, if the induced shift of the loop (*H*_{coupl}) is greater than the FL coercive field (*H*_{C}), only one resistance state remains stable at zero field, which is not suitable for applications. For in-plane MTJ, this problem has been solved by replacing the pinned layer by a synthetic antiferromagnet (SAF) consisting of two ferromagnetic layers antiferromagnetically coupled through a nonmagnetic spacer layer (such as Ru) by the Ruderman–Kittel–Kasuya–Yosida coupling (*J*_{Ru}) [Parkin 1990]. In this system, when the antiferromagnetic configuration is stabilized because of the well-chosen Ru layer thickness, stray fields are considerably decreased [Leal 1998] and the shift of the FL hysteresis loop is greatly reduced. In the case of systems with PMA, such compensation has been previously studied for the purpose of multilevel recording [Baltz 2007, 2009]. In this letter, we investigate the effects of these stray fields in the context of MTJs and compare two types of p-MTJs: the first one using a single hard reference layer (HL), the other one using a perpendicular SAF. We find that SAF structures are highly suitable for magnetoresistive random access memories (MRAM) cells with PMA, especially when the diameter of the cells (Φ) is scaled down.

SECTION III

## RESULTS AND DISCUSSION

The magnetic properties of SAF and HL have been first investigated on unpatterned samples by VSM and EHE. VSM measurements indicate that both magnetizations of SAF have a saturation magnetization of *M*_{S} = 1200 ± 50 emu·cm^{−3}. The two moments are compensated, the macroscopic magnetization of SAF is equal to zero without applied field. The saturation magnetization of HL is also *M*_{S} = 1200 ± 50 emu·cm^{−3}. Hysteresis loops measured by EHE are shown in Fig. 1. All magnetizations are perfectly aligned along the thin film plane normal. However, SAF presents much higher reversal fields. Such behavior can be explained by the different nucleation process in antiferromagnetically coupled perpendicular layers [Hellwig 2003]. The SAF presents two magnetic contributions: the jump at 2500 Oe corresponds to the bottom (Co/Pt) multilayer reversal, and the minor loop denoted by the black arrows in Fig. 1 corresponds to the top (Co/Pt)/CoFeB layers reversal. By measuring the shift *H*_{Ru} of the loop of the softest magnetic layer in SAF, we can deduce that *J*_{Ru} = *H*_{Ru} *tM*_{S} = 0.8 erg·cm^{−2} here. *H*_{Ru} is thus much larger than the coercive field of the minor loop, so that the antiferromagnetic configuration is the only stable one in zero field. EHE measurements show different level of resistance at zero field for SAF, whereas VSM measurements showed that both magnetizations where equal. The difference of Hall resistances contributions for the two magnetizations can be explained by the dependence in such systems of the skew scattering and quantum side jump factors on the composition of the magnetic layer and of its adjacent layers, as well as its thickness [Canedy 1997]. The Hall coefficient of the thick CoFeB layer is thus lower than the Co ones. The measurement of anisotropy fields indicates that the effective anisotropy is 4.6 × 10^{6} erg·cm^{−3} for HL, 6 × 10^{6} erg·cm^{−3} for the bottom SAF layer, and 3.6 × 10^{6} erg·cm^{−3} for the top SAF layer. For a 90-nm-diameter cell, these correspond to 2820*k*_{B}T for HL and 4400*k*_{B}*T* for SAF. Such a high stability can be explained by the high PMA induced at the Co/Pt and CoFeB/AlOx interfaces [Nistor 2009].

For unpatterned samples, when the FL is added on top of the barrier, we obtain a complete p-MTJ. After annealing, the FL (Co/Pt) multilayer remains perpendicular to plane. The FL minor hysteresis loop is shifted by 40 Oe in a direction indicating antiferromagnetic coupling across the alumina barrier. This coupling of 40 Oe is probably due to magnetostatic coupling induced by roughness [Moritz 2004], since AlO_{x} is amorphous and pinholes would induce a ferromagnetic interlayer coupling. The magnetic anisotropy of this FL electrode is 2.6 × 10^{5} erg·cm^{−3}, which is much smaller than the PMA of bottom electrodes (HL and SAF), due to poor growth conditions on the oxide, and the lack of (111) texture. Nevertheless, this corresponds to 95*k*_{B}T for a 90-nm-diameter cell, which fulfils the stability condition for the 90-nm technological node. CIPT measurements indicate that this junction presents a TMR ratio of 6% and a resistance–area (RA) product of 21 Ω·μm^{2}. Such a low TMR ratio is probably due to small thickness of the top magnetic electrode and the absence of a reasonably long annealing [Sousa 1998], which would greatly improve barrier/magnetic layers interfaces. Moreover, it is probable that the Pt atoms diffuse to the barrier, enhancing nonspin-dependent tunneling conduction.

Fig. 2(a) and (b) shows calculation of the perpendicular to plane component of the stray fields (H_{z}) in a middle section of FL along the normalized lateral dimension *x*/*r*, *r* being the radius of the pillar, using a simple macrospin model that considers facing charged surfaces [Hubert 1998]. Both cases of HL and SAF have been investigated for different pillar diameters, assuming that all magnetic layers of HL and SAF have perfectly rigid perpendicularly oriented moments. Magnetization of each magnetic layer (Co or CoFeB) has been taken equal to 1200 emu·cm^{−3}, following VSM measurements. These calculations show that stray fields are greatly reduced when HL is replaced by the compensated SAF, due to its reduced effective magnetization. However, since the center of the two magnetic layers of SAF does not coincide, it cannot be cancelled everywhere, edges presenting higher fields due to the higher contribution given essentially by the closest layers.

Fig. 3 shows the evolution of the calculated mean stray field *H*_{coupl} = ∫*H*_{z} (*x*,*y*)*dxdy*/(π *r*^{2}) for different dot diameters as a function of the ratio α = *m*_{2}/*m*_{1}, *m*_{1}, and *m*_{2} being, respectively, the magnetic moments of the top (Co/Pt)/CoFeB and the bottom (Co/Pt) multilayer. In this study, we only changed the number of repeats in the bottom (Co/Pt) multilayer in order to increase the contribution of this layer. For α equal to 1 (vertical dotted line), we have a perfectly compensated SAF giving the increased dipolar interaction when reducing the diameter of the pillar, as shown in Fig. 2(c). Fig. 3 (inset) gives the optimal ratio α_{opt} at which *H*_{coupl} is 0 (horizontal dotted line) as a function of the diameter. It is noticeable that when the diameter of the pillar decreases, the slope of *H*_{coupl}(α) increases, which means that the optimization becomes more sensitive to deviations from the perfect ratio. In summary, contributions of both layers have to be adjusted for each diameter to minimize dipolar interaction, by varying the magnetic layers thickness and/or the number of repeats of the (Co/Pt) multilayers on both sides of the Ru spacer.

Patterned samples show resistance hysteresis loops exhibiting TMR ratio varying from 0.5% to 6% and RA products varying from 10 to 28 Ω·μm^{2}. Optimal TMR ratios are obtained along with the nominal RA product of 21 Ω·μm^{2}. The spread of electrical properties from the CIPT measurements are essentially due to fabrications issues. No significant TMR or RA difference can be observed between HL-MTJ and SAF-MTJ junctions in terms of electrical properties. Fig. 4 presents hysteresis loops measured on one pillar of HL-MTJ and SAF-MTJ. It shows several steps on the 540-nm diameter pillars, probably due to the nonuniform reversal of the different magnetic grains of the FL, edge roughness on the dot or pinning and depinning of domain walls on defects in the pillar. Reducing the diameter to 140 nm makes the steps of FL reversal disappear, indicating a more coherent magnetization switching within individual pillars. However, the pillar to pillar fluctuations in *H*_{C} are larger in the smaller pillars due to a reduced number of grains and defects per pillar, as can be seen from the larger σ_{Hc} values on Table 1. The increase of σ_{Hc} when using a SAF reference layer for the 140-nm diameter pillars instead of single hard layer seems to be mainly due to fabrication reproducibility issues and should not be a problem when an optimized process for this kind of stacks will be achieved.

In these pillars, *H*_{coupl} contains different contributions: Néel coupling, pinholes, and stray field effects. Even if these different contributions cannot be separated in the present measurements, only the dipolar interactions should vary by changing the pillar size. The average of Néel and pinhole coupling can be estimated to be 40 Oe, as measured on the unpatterned sample. Nevertheless, the influence of the SAF is clearly seen on these loops: SAF-MTJ presents smaller shifts than HL-MTJ at the same diameter. Without SAF, only the parallel state remains stable at zero field for the smaller dots, whereas *H*_{coupl} is lower than *H*_{C} in SAF-MTJ. We can verify that stray fields greatly depend on the size of the MTJ, as predicted by calculations. Table 1 shows that 140 nm pillars of SAF-MTJ present a coupling field of 80 Oe since α = 1 is not the optimal ratio for this diameter (calculations give α_{opt} = 1.23), whereas it is only 5 Oe for the 540-nm diameter pillars in which α_{opt} = 1.16, closer to the experimental one. The shift of the hysteresis loops of HL-MTJ is also increased up to 170 Oe for this diameter, significantly larger than in SAF-MTJ. However, the model does not fit our experimental data well, particularly for the smaller dots of HL-MTJ, which should present much higher *H*_{coupl}. This could be attributed to several factors, which are not taken into account in the model: possible oxidation of the sidewalls during the etching, which can decrease the magnetic moments [Yoshikawa 2006], geometrical imperfections of the dots, roughness, defects, and other inhomogeneities.