Enhanced voltage-controlled magnetic anisotropy in magnetic tunnel junctions with an MgO/PZT/MgO tunnel barrier

There is a fast-growing need in the semiconductor industry for alternative memory technologies which can combine nonvolatile operation, high speed, high endurance, and high density in a single silicon-compatible device. Magnetic random access memory (MRAM) is an emerging candidate providing potential advantages in a range of standalone and embedded memory applications. Present MRAM devices typically utilize current-controlled switching of magnetization via the spin transfer torque (STT)^1,2 or spin-orbit torque (SOT)^3,4 effects to write information into magnetic bits. However, the use of currents results in a memory cell size (i.e., bit density) limitation due to the large size of the required access transistors,^5,6 and large dynamic switching energy due to Ohmic power dissipation. Therefore, there has been a great interest in using an applied voltage (instead of current) to manipulate the magnetization of nanoscale magnetic tunnel junctions (MTJs). The electric-field effect, or the voltage-controlled magnetic anisotropy (VCMA) effect, is utilized to temporarily lower the interfacial perpendicular magnetic anisotropy (PMA) of the free layer during the writing operation, thus reducing the writing energy required to overcome the energy barrier between the two stable magnetization states.^7,8

A promising type of electric-field-controlled memory device has been realized in perpendicular magnetic tunnel junctions using the Ta/CoFeB/MgO material system, where both high tunneling magnetoresistance (TMR)^9,10 and VCMA-induced magnetization switching^11–14 have been demonstrated. For large memory array (>1 Gb) with scaled CMOS below 14 nm, VCMA coefficients larger than 200 fJ/V m may be needed.^15,16 However, the traditional Ta/CoFeB/MgO system offers limited VCMA in the range of 10–60 fJ/V m.^17–23 

To achieve a larger VCMA effect, multiple approaches have been explored, such as using different seed and cap layers adjacent to the ferromagnetic layer.^17,24–27 Ab initio electronic structure calculations have revealed that epitaxial strain has a dramatic effect on increasing the VCMA.²⁸ Another promising method is by utilizing different dielectrics. As the VCMA effect originates from the charge accumulated at the CoFeB/oxide interface when voltage is applied,²⁹ it has been demonstrated theoretically that using a single oxide or multiple layers of oxides with higher dielectric constant(s) (⁠$ϵ$⁠) can induce a higher VCMA coefficient, thus a reduction in voltage for magnetization switching.³⁰ In past experimental works, enhanced VCMA effect was measured in CoFeB/oxide structures using MgO/Al₂O₃ and MgO/HfO₂/Al₂O₃ as the gate oxide,²³ but lacked an electrical readout because full MTJ was not fabricated. Moreover, there has been intensive research on MTJs using barrier materials other than MgO. However, MTJs using SrTiO₃ with CoFe electrodes had a rather low TMR around 10%;³¹ likewise, multiferroic tunnel junctions with ferroelectric barriers such as PbZr_0.2Ti_0.8O₃ and BaTiO₃ with Co/Fe and La_0.7Sr_0.3MnO₃ electrodes only demonstrated a reasonable TMR below room temperature.^32,33 Therefore, to achieve better writing efficiency with reliable readout for voltage-controlled MRAM, it is critical to have a sizeable room temperature TMR in addition to VCMA enhancement after integration of high-$ϵ$ oxide(s) into the stack.

In this work, an ultra-thin layer of high-$ϵ$ lead zirconate titanate (PZT or Pb(Zr_xTi_1-x)O₃) was integrated into the MgO tunnel barrier in order to enhance the VCMA effect while maintaining a sizeable TMR. A combination of sputtering and atomic layer deposition (ALD) techniques was used to grow MTJ stacks with an MgO/PZT/MgO tunnel barrier. Based on measurements on an ensemble of MTJ devices with the MgO/PZT/MgO barrier, the VCMA coefficients were improved by about 40%, and the room-temperature TMR values were comparable—only slightly lower than in those of MgO barrier MTJs.

PZT has been commonly used in Ferroelectric Random Access Memory (FeRAM) devices^34–36 and has been used in multiferroic tunnel junctions.³² In this work, PZT thin film was integrated into the tunnel barrier because it has one of the largest dielectric constants (i.e., 300–1300 for 1–3 μm PZT thin films^37,38). Due to the fact that the PZT was interfaced with MgO on both sides, the interfacial dead layer that is intrinsic to the electrode/dielectric boundary is expected to be negligible in our film.³⁹ PZT deposition was performed via ALD,^40,41 which has been previously shown to provide conformal atomically smooth ultra-thin films with precise control over composition and thickness.⁴² 

MTJs with a pure MgO tunnel barrier were used as the reference sample and compared to the MTJs with the MgO/PZT/MgO tunnel barrier (hereafter referred to as MgO MTJ and PZT MTJ, respectively). Sample structures are schematically illustrated in Figure 1, with the following structures: Ta(18 nm)/Co₂₀Fe₆₀B₂₀(0.9 nm)/MgO(2.5 nm)/Co₂₀Fe₆₀B₂₀(2.0 nm)/Ta(4 nm)/Pt(2 nm) for the MgO MTJ, and Ta(18 nm)/Co₂₀Fe₆₀B₂₀(0.9 nm)/MgO(1.0 nm)/PZT(1.5 nm)/MgO(1.0 nm)/Co₂₀Fe₆₀B₂₀(2.0 nm)/Ta(4 nm)/Pt(2 nm) for the PZT MTJ.

The stacks were deposited on thermally oxidized Si substrates using an AJA magnetron sputtering system and thermal ALD. All metallic layers were DC sputtered. The Co₂₀Fe₆₀B₂₀ bottom free layer has a thickness of 0.9 nm and the top fixed layer has a thickness of 2.0 nm; they were out-of-plane and in-plane magnetically anisotropic, respectively.⁹ For the MgO MTJ, a 2.5 nm thick MgO tunnel barrier was grown by RF sputtering, while for the PZT MTJ, a 1.0 nm thick MgO layer was first sputtered, then a 1.5 nm thick PZT film was deposited via ALD at a substrate temperature of 250 °C, and finally, a 1.0 nm thick MgO was sputtered to form the MgO/PZT/MgO tunnel barrier. The synthesis of PZT thin film has been outlined in previous papers.^40,41 The PZT MTJ film stack was annealed at 200 °C under vacuum both before the PZT deposition and after depositing the whole film stack. Since the PZT MTJ film stack was also in-situ annealed during the ALD process under 250 °C, the MgO MTJs were annealed at 250 °C for a fair comparison. MTJ devices with elliptical diameters of 4 × 16 μm and 4 × 12 μm were subsequently fabricated using standard photolithography and dry etching techniques.

First, material properties of the MTJ stacks were characterized using Kratos AXIS X-ray photoelectron spectroscopy (XPS) and an FEI Titan scanning transmission electron microscope (STEM). XPS confirmed the composition ratio Zr:Ti = 52:48 of the PZT thin film deposited on the bottom layers of a film stack, as shown in Figure 2(a). Note that it has been shown that PZT exhibits enhanced properties (e.g., dielectric constant) at the morphotropic phase boundary composition of Zr:Ti = 52:48.⁴³ The XPS survey scan also showed the Mg KLL, Co 2p, Fe 2p, and Ta 4d elemental peaks. Note that the B 1s peak was not observed because the estimated XPS penetration depth is limited to 10 nm and due to the fact that most of the boron has diffused far into the Ta layer due to the annealing process.^44,45 Cross-sectional TEM was performed on the fabricated MgO MTJ and PZT MTJ devices, as shown in Figures 2(b) and 2(c), respectively, in which the arrows indicate the general location of layer interfaces, spaced per Figure 1. Nano-diffraction patterns were collected for both cross-sections, as shown in the insets of Figures 2(b) and 2(c). A selected-area aperture was used for the MgO MTJ, but in order to maximize diffracted intensity from the ∼3 nm thick MgO/PZT/MgO layers-of-interest in the PZT MTJ, a highly condensed probe was employed, elongated along the in-plane direction of the film, which provided informative results due to the FEI Titan's parallel beam nearly all the way to the crossover point. The inset diffraction patterns clearly showed that the MgO had crystallized; however, indexing of the remaining spots to either CoFeB or PZT was not possible due to resolution limitations. Next, unpatterned MgO and PZT MTJ stacks were characterized for their magnetic properties using superconducting quantum interference device (SQUID) magnetometry. The saturation magnetizations (M_s) were measured to be 1017 $±$ 22 emu/cm³ and 932 $±$ 41 emu/cm³ for MgO and PZT MTJ stacks, respectively, indicating that the ALD PZT deposition had not significantly affected the magnetic properties of the CoFeB layers.

The MTJs were then measured electrically to investigate the VCMA effect via the TMR readout at room temperature.^17,46 The resistance was measured as the in-plane magnetic field was swept while voltages were applied between −300 to +300 mV, as shown in Figure 3(a). At zero magnetic field, the magnetic moment of the bottom CoFeB free layer was perpendicular and that of the top fixed CoFeB layer was in-plane, while at the maximum in-plane magnetic field, the two CoFeB layers were both in-plane magnetized. Hence, the resistance decreased as the magnetic field was increased. The resistance-area (RA) products of the PZT and MgO MTJ in Figure 3(a) were 98 $kΩ μm2$ and 14 $kΩ μm2$⁠, respectively, which are typical for voltage-controlled MRAM.^11,15 Using the equation $G=GS(1+PF2 cos θ)$⁠, the measured conductance G of the MTJ was related to the relative angle between the two CoFeB layers, where $GS$ was the mean surface conductance, $θ$ was the angle between two CoFeB layers, and $PF$ was the effective spin polarization.⁴⁷ As the top 2.0 nm thick CoFeB layer was fixed at an in-plane direction, the in-plane magnetization component M_x of the bottom free layer CoFeB can be obtained by $MxMS=cos θ=G(H)−G(0)G(Hmax)−G(0)$⁠, where G(H), G(H_max), and G(0) are, respectively, the MTJ conductances at in-plane magnetic field H, at the maximum in-plane magnetic field measured, and at zero external field.^23,46 Note that here H_max was determined by saturation of the free layer magnetization to the in-plane orientation, and the PZT MTJ demonstrated a higher saturation field than the MgO MTJ. The perpendicular magnetic anisotropy energy E_perp can then be calculated by conducting the following integration for the free layer from the perpendicular easy axis (at zero external field), to the in-plane hard axis (at H_max): $Eperp=MS∫01Hd(MxMS)$⁠.^17,18

Next, using the equation $Ki=(2πMS2+Eperp)tCoFeB$⁠, the value of interfacial PMA (K_i) was obtained,⁹ where t_CoFeB is the thickness of the CoFeB free layer. Finally, the VCMA coefficient $ξ$ was determined by $ξ=ΔKi/ΔEeff$⁠, where the effective electric field E_eff was calculated by dividing the applied voltage $V$ with the total thickness d of the tunnel barrier. All measurements were performed at room temperature.

The VCMA coefficients $ξ$ (i.e., the slope of K_i versus E_eff plot) are shown in Figure 3(b) for two representative MgO and PZT MTJ devices. A total of six devices were measured for each MTJ stack. The average VCMA coefficients were $ξ$_average = 14.3 $±$ 2.7 fJ/V m for MgO MTJs, and $ξ$_average = 19.8 $±$ 1.3 fJ/V m for PZT MTJs, as shown in Figure 4(a). Therefore, by incorporating the PZT film into the MgO barrier, the VCMA effect was shown to be enhanced by about 40%.

From the physics point of view, this enhanced VCMA effect could be understood as follows. As indicated from ab initio calculations, the K_i stems from the hybridization of Fe/Co 3d orbitals and O 2p orbitals at the CoFeB/MgO interface.^48,49 The application of a positive electric field (i.e., top electrode of the MTJ at a higher electric potential) across the MgO barrier induces accumulation of electrons at the bottom CoFeB/MgO interface, which in turn affects the hybridization of Fe/Co and O orbitals, thus decreasing the value of $Ki$⁠,^29,30 which is consistent with the data shown in Figure 3(b). Hence, if the interface charge density $σq$ increases for the same applied electric field $Eeff$⁠, a larger VCMA coefficient (⁠$ξ=ΔKi/ΔEeff$⁠) can be effectively achieved. The interface charge density $σq$ can be expressed as $σq=ϵ0ϵeffV/d=ϵ0ϵeffEeff$⁠, where $ϵ0$ is the permittivity of free space, and $ϵ$_eff is the effective dielectric constant of the tunnel barrier.^23,30 Thereby, for the same tunnel barrier thickness and applied voltage, the increase in the effective dielectric constant $ϵ$_eff by incorporating PZT in the tunnel barrier gives rise to a larger interface charge density change at the CoFeB/MgO interface, thus resulting in a larger overall VCMA.

From the obtained VCMA ratio between the PZT MTJ and the MgO MTJ, the dielectric constant for the PZT ultra-thin film could also be calculated using a serial capacitor assumption. For the PZT MTJ, the effective dielectric constant was $ϵPZT−MTJ=(dMgO+dPZT)/(dMgO/ϵMgO+dPZT/ϵPZT)$⁠, while for the MgO MTJ, the dielectric constant was assumed to be $ϵMgO−MTJ=ϵMgO=10$⁠.⁵⁰ As the change of interfacial PMA is proportional to the change of interface charge density, i.e., $ΔKi∝Δσq$⁠, it is deduced that $ξ∝ϵeff$⁠.²³ Thus, based on the VCMA coefficients obtained for PZT and MgO MTJ, the dielectric constant of the PZT ultra-thin film was estimated to be 28.4, a plausible value taking into account the 1.5 nm PZT thickness, as well as existing literature values for an ultra-thin ALD PZT film.^39,41

The VCMA coefficients were also plotted against K_i (Figure 4(a)) and TMR ratio (Figure 4(b)) for all measured MgO and PZT MTJ devices. The PZT MTJs were observed to have a larger VCMA effect and a slightly smaller TMR ratio compared to the MgO MTJs. The VCMA_average was 14.3 $±$ 2.7 fJ/V m for MgO MTJs, and 19.8 $±$ 1.3 fJ/V m for PZT MTJs. The TMR_average was 61.4 $±$ 11.5% for MgO MTJs, and 53.1 $±$ 1.7% for PZT MTJs. Note that the TMR ratio here was defined by $TMR=(Rap−Rp)/Rp$⁠, where the anti-parallel resistance $Rap$ was calculated according to equation $1/Rap=2/Rort−1/Rp$⁠,⁴⁷ where the parallel resistance $Rp$ was the resistance at the maximum magnetic field or $1/G(Hmax)$⁠, and the orthogonal CoFeB configuration resistance $Rort$ was the resistance at zero external magnetic field or $1/G(0)$⁠.

Compared with other works on Ta/CoFeB/MgO in the literature with the VCMA coefficients ranging from 10 to 60 fJ/V m,^17–23 the VCMA coefficient values in our PZT and MgO MTJs are at the lower bound, but the VCMA effect can be improved by optimizing a number of parameters, including annealing conditions,²⁷ surface roughness,⁵¹ and intrinsic strain²⁸ of the layers. Nevertheless, a 40% enhancement in the VCMA coefficient was achieved by using the MgO/PZT/MgO tunnel barrier while a relatively high TMR was still preserved.

In conclusion, by combining atomic layer deposition and magnetron sputtering techniques, an ultrathin PZT layer was incorporated into the MgO tunnel barrier of a magnetic tunnel junction. The resulting magnetic tunnel junctions using a high-$ϵ$ tunnel barrier were shown to have both large tunneling magnetoresistance (>50%) and an enhanced VCMA effect (by 40%) at room temperature. This high-$ϵ$ tunnel barrier MTJ is a potential candidate for future voltage-controlled, ultralow-power, high-density MRAM devices.

This work was supported by the NSF Nanosystems Engineering Research Center for Translational Applications of Nanoscale Multiferroic Systems (TANMS). The authors are grateful to the UCLA Nanoelectronics Research Facility (NRF), California NanoSystems Institute (CNSI), specifically the Materials Lab at the Molecular Instrumentation Center (MIC), Nano and Pico Characterization Lab (NPC), and Electron Imaging Center for NanoMachines (EICN), for their assistance and use of lab equipment. The authors would like to acknowledge the collaboration of this research with King Abdul-Aziz City for Science and Technology (KACST) via The Center of Excellence for Green Nanotechnologies (CEGN). The authors would also like to thank Professor Greg Carman and Professor Chris Lynch for fruitful discussions.