Understanding stability diagram of perpendicular magnetic tunnel junctions

Perpendicular magnetic tunnel junctions (MTJ) with a bottom pinned reference layer and a composite free layer (FL) are investigated. Different thicknesses of the FL were tested to obtain an optimal balance between tunneling magnetoresistance (TMR) ratio and perpendicular magnetic anisotropy. After annealing at 400 °C, the TMR ratio for 1.5 nm thick CoFeB sublayer reached 180% at room temperature and 280% at 20 K with an MgO tunnel barrier thickness corresponding to the resistance area product RA = 10 Ohmμm2. The voltage vs. magnetic field stability diagrams measured in pillar-shaped MTJs with 130 nm diameter indicate the competition between spin transfer torque (STT), voltage controlled magnetic anisotropy (VCMA) and temperature effects in the switching process. An extended stability phase diagram model that takes into account all three effects and the effective damping measured independently using broadband ferromagnetic resonance technique enabled the determination of both STT and VCMA coefficients that are responsible for the FL magnetization switching.

Magnetic tunnel junctions (MTJs) have become a basic building block for various types of spintronics devices, such as magnetic random access memory (MRAM) cells, magnetic field sensors and microwave generators or detectors 1 . The properties of spintronics devices, such as thermal stability of an MRAM cell 2 or sensitivity of microwave detectors 3 utilizing MTJs can be greatly improved by using magnetic layers with perpendicular anisotropy 4 . Among a few ways to realize such a perpendicular MTJ, taking advantage of the interface anisotropy component 5 yields the best results so far, especially in terms of high tunneling magnetoresistance (TMR) ratio, which is measured typically in MTJs with CoFeB/MgO/CoFeB trilayer. Recent studies on perpendicular MTJ showed the TMR ratio exceeding 200% 6 thanks to careful optimization of both the free layer (FL) and reference layer (RL) structure 7 . In addition, one of the key challenges for the commercial development of spin transfer torque (STT)-MRAM is to optimize perpendicular MTJ to withstand the temperature budget introduced at the back end of line CMOS fabrication process with temperatures up to 400 °C. To achieve this a precise design of the layer stack, taking into account all constituent layers as well as the properties and the treatment of the bottom electrode, has to be performed.
In this letter, we report on the perpendicular MTJ with a composite CoFeB/W/CoFeB FL 8,9 which is characterized by high perpendicular magnetic anisotropy and spin polarization resulting in up to 180% TMR measured at room temperature and above 280% TMR at low temperature. The RL is pinned to a synthetic ferromagnet (SyF) consisting of Co/Pt super-latices 10 coupled by a thin Ru spacer. Electrical transport measurements were performed in MTJs patterned into 130-nm diameter pillar. Voltage vs. perpendicular magnetic field switching diagrams 11,12 are measured in order to separate between STT, voltage control of magnetic anisotropy (VCMA) and temperature effects. An analytic model based on work by Bernert et al. 13 was extended to reproduce the experimental results. ranging from 1.1 up to 1.6 nm. The schematics of the multilayer stack is presented in Fig. 1. The bottom Co/Pt super-lattices coupled by a thin Ru spacer are characterized by high perpendicular magnetic anisotropy (PMA). The W-based separator between the superlatices and the reference layer ensures high ferromagnetic coupling between the top super-lattice and the RL. In addition, it provides structural transition from a face center cubic SyF 14 to a body center cubic CoFeB and contributes to the absorption of B atoms from CoFeB during annealing and crystallization processes. Damping parameter was measured using ferromagnetic resonance (FMR) (see Methods section). Figure 1 presents the dependence of the ΔH on the excitation frequency.
Vibrating sample magnetometry (VSM) measurements of a representative sample with t FL = 1.1 nm presented in Fig. 2 reveal independent switching of the FL (at small magnetic fields below 50 kA/m) and RL (at high magnetic fields between 150 and 300 kA/m), which ensures bistable parallel (P) and antiparallel (AP) state. The FL magnetization was calculated and yielded μ 0 M S = 1.12 T. An inset of Fig. 2 depicts the TMR ratio for different t FL measured on the wafer-level using current in-plane tunneling (CIPT) method. An increase of the TMR from 140% for t FL = 1.1 nm up to TMR = 180% for t FL = 1.5 nm is explained by an increase of the spin polarization for the thicker ferromagnetic layer. A rapid reduction of the TMR for t FL = 1.6 nm is caused by the transition of the FL magnetization vector to the sample plane. After the patterning process the TMR ratio measured in the nano-pillar was about 5-15% lower comparing to the result obtained from CIPT method, which can be explained by the appearance of the small serial parasitic resistance. This small resistance, however, has negligible influence on the MTJ parameters derived afterwards. Figure 3 presents the TMR vs. magnetic field dependence measured in the MTJ nanopatterned into pillas of 130-nm in diamater with different t FL . Increase in t FL leads to an increase in TMR ratio and decrease in the

Stability diagram.
To further elucidate the properties of the fabricated MTJ, current (voltage)-induced switching loops were measured in the presence of the perpendicular magnetic field. An inset of Fig. 3 presents a representative resistance vs. voltage loop measured in a magnetic field compensating the offset field in the MTJ with t FL = 1.1 nm. It has been already established that apart from the conventional STT effect observed in MTJs with relatively thin MgO barriers, the switching process can be also affected by the VCMA effect in devices with a thin FL 15,16 . To investigate the switching process in more detail, we repeated the R(V) loop measurements, with different constant magnetic field. The stability diagram obtained in this way both at room temperature (T = 300 K) and low temperature (T = 20 K) for the MTJ with t FL = 1.1 nm is presented in Fig. 4. To understand these diagrams, the following fitting procedure was used. First, low-temperature data were modeled using Eq. (4) to obtain switching field H W , H S (being the offset field measured at low bias voltage) and STT coefficients, which are little affected by heat. The slope of V(H) depends mostly on damping-like torque -a DL (see modelling details in Methods section), whereas the vertical offset is adjusted by H W -solid lines in Fig. 4. Next, to compensate the offset between AP-P and P-AP switching voltages, (which take place at opposite electric field applied to the MgO/ CoFeB interface) VCMA coefficient k V was introduced according to the Eq. (5), without temperature influence so far (k t = 0). Finally, thermal reduction of H W was introduced by adjusting k t to fit the stability diagram obtained at room temperature. Independently, we simulated the Joule-heating effect in our MTJ pillar and we found negligible temperature increase, therefore, we assumed constant temperature in our analysis.  In addition, for the precise derivation of the STT coefficients, the magnetization damping was calculated based on an independent FMR measurement presented in Methods section and included in Table 1.

Discussion
Fitting the experimental stability diagram to the Eqs (4) and (5) yielded the temperature coefficients of k t = 0.0014 1/K. This parameter was kept constant for MTJs with different t FL . Remaining parameters of the stability diagrams for each t FL were modeled independently. For t FL = 1.1 nm, the following STT components were obtained a DL = 0.024 T/V and a FL = 0.02 T/V 2 , however, we note that the modeled stability diagram is only little sensitive to a FL , which agrees with another macrospin approach based on LLG equation presented in Ref. 17. Damping-like torque τ DL was thereafter recalculated using Eq. (1):  Table 1.
The damping-like torque component obtained from the stability diagram is almost constant as a function of t FL , which is explained by little dependence of the TMR ratio, and thus the spin polarization, on the ferromagnetic layer thickness in the investigated regime. The VCMA coefficient is comparable for MTJs with t FL = 1.3 nm and 1.5 nm and greater than in MTJ with t FL = 1.1 nm. This behavior is expected, as for thicker t FL the absolute value of the effective magnetization is reduced and it is more susceptible to the anisotropy change induced by the electric field 23 . Moreover, in the same thickness regime, where the transition between perpendicular and in-plane anisotropy occurs, the effective damping increases, which may be attributed to an increase in the level of magnetization disorder 24 .
In conclusion, we investigated perpendicular MTJs with composite CoFeB/W/CoFeB FL of different thickness and SyF Co/Pt/Ru-pinned RL. In the investigated FL thickness range we observed an increase of the effective damping extracted from the broadband FMR measurements with increasing FL thickness, which is mainly caused by the reduction of the effective anisotropy. After patterning MTJs into nano-meter scale pillars, we measured the resistance vs. voltage loops for different external magnetic field amplitudes and created the stability diagrams for each FL thickness. To model the experimental data, we included the thermal and VCMA terms into the theoretical STT-switching phase diagram. Based on the fitting procedure, we obtained STT components together with the VCMA coefficient. Our findings shine more light on the switching process of MTJs applied in future MRAM technologies.

Methods
Sample deposition and nano-fabrication. Multilayer samples were deposited using Singulus TIMARIS sputtering system on chemically-mechanically polished 4-inch Si wafers. After the deposition, the samples were annealed at 400 °C to induce proper crystallographic orientation of Fe-rich CoFeB and PMA of the CoFeB/MgO interfaces. Wafer-level parameters of the deposited multilayers were investigated by CIPT 25 , VSM and broadband FMR methods 26 . The latter was performed by measuring the complex transmission coefficient (S 21 ) in a dedicated coplanar waveguide with a 10 × 8 mm unpatterned sample placed face down. The frequency of the vector network analyzer is kept between 4 and 22 GHz, while sweeping the perpendicular magnetic field in ±550 kA/m range.
Selected MTJs were patterned into circular cross-section pillars with diameter ranging from 130 up to 980 nm by means of electron-beam lithography, ion-beam etching and lift-off process.
The transport properties presented in this work were measured for the smallest devices with the area of A = 0.013 μm 2 in a dedicated probe station equipped with magnetic field source. Four-probe method with a voltage source was used to apply 1-ms long pulses and measure the resistance during this voltage-pulse application. The stability diagrams were determined by sweeping the voltage pulses amplitude in the presence of a given magnetic field. Selected devices were characterized at low temperatures of T = 20 K in order to determine the temperature influence on the magnetization switching properties.
Modelling. Magnetization damping was calculated based on linewidth ΔH, which was measured using FMR technique and fitted by the Eq. (2): where γ 0 = γμ 0 , with the gyromagnetic ratio γ = (gμ B )/ħ = 28 GHz/T, μ 0 is the permeability of the free space, g is the Lande spectroscopic splitting factor, μ B is the Bohr magneton, ħ is the reduced Planck's constant, a DL and a FL are the damping-like and field-like STT coefficients expressed in T/V and T/V 2 units, respectively, α is the magnetization damping, R and R p are the MTJ resistance in a given state and minimal (parallel state) resistance, H eff is the effective magnetic field: where, V C is the switching voltage. It was assumed that damping-like (field-like) torque component is a linear (quadratic) function of the applied current 20 . To account for the additional physical effects that contribute to the stability diagram, namely VCMA and temperature, H W is scaled by the factor: where V is the applied voltage, H C is the coercive field, k V is the VCMA coefficient 27,28 and T is the ambient temperature. The dependence of the switching field on the temperature is represented by k t , which in the first approximation is a square-root function 29 . We note that for the discussed device size, the switching process may not be entirely uniform, however in Ref. 30 the authors found substantial deviations from the macrospin evolution only for MTJs of diameter greater than 150 nm.
The damping factor was measured independently by the broadband FMR technique. For each microwave frequency f, the complex magnetic susceptibility vs. magnetic field χ(H) is extracted from S 21 measurement by subtracting the magnetic independent offset and time-dependent drift 31 :