Current-induced Magnetization Switching by a generated Spin-Orbit Torque in the 3D Topological Insulator Material HgTe

Abstract

Magnetic random access memory (MRAM) technology aims to replace dynamic RAM (DRAM) due to its significantly lower power consumption and non-volatility [Dong08]. During the last couple of years the commercial focus was set on spin-transfer torque MRAM (STT-MRAM) systems, where a current is pushed through a ferromagnetic (FM) free layer and a reference layer which are separated by an insulator. The free layer can be set to parallel or anti-parallel depending on the current direction [Kim11]. Unfortunately these currents have to be quite high which could lead to damages of the tunnel barrier of the magnetic tunnel junction resulting in higher power consumption as well as reliability issues. At this point a new effect, where the current is passed below the ferromagnetic layer stack, can be exploited to change the direction of the free layer magnetization. The effect is known as spin-orbit torque (SOT) and describes the transfer of angular momentum onto an adjacent magnetization either by the spin Hall effect (SHE) or inverse spin galvanic effect (iSGE) [Manchon19]. The latter describes a spin accumulation due to a current. This is similar to the process of spin accumulation in TIs, where a current corresponds to an effective spin due to spin-momentum locking [Qi11]. Thus TIs exhibit a high current-to-spin conversion rate, which makes them a promising material system for SOT experiments. Among all TIs it is HgTe, which can be reliably grown as an insulator. This thesis covers the development of a working device for SOT measurements (SOT-device) in a CdTe/CdHgTe/HgTe/CdHgTe heterostructure. It involves the development of a tunnel barrier (ZrOx) as well as the investigation of the behavior of a ferromagnetic layer stack on top of etched HgTe. The main result of this work is the successful construction and evaluation of a working SOT-device, which exhibits the up to date most efficient switching of in-plane magnetized ferromagnetic layer stacks. \n \nIn order to avoid hybridization between HgTe and the adjacent ferromagnetic atoms, which would cause a breakdown of the topological surface state, it is necessary to implement a thin tunnel barrier in between the TI and free layer [Zhang16]. Aside from hybridization a tunnel barrier avoids shunting of the current, that is pushed on the surface of the HgTe/CdHgTe interface. Thus a bigger part of the current can be used for spin accumulation and, at the same time, the resistance measurement of the ferromagnetic layer stack is not perturbed. In chapter 3 the focus is set on investigating the tunneling characteristics of ZrOx on top of dry etched HgTe. Thin barriers are used as the interaction of the current generated spin and the adjacent magnetization decreases with distance. On the other hand too small insulator thicknesses lead to leakage currents which disturb heavily the measurement of the resistance of the ferromagnetic layer stack. Thus an optimum thickness of 10 ALD cycles (\\(d\\approx 1.6\\rm\\, nm\\)) is determined which yields a resistance area product of \\(R\\cdot A \\approx 3\\rm\\, k\\Omega\\mu m^{2}\\). This corresponds to a tunneling resistance of \\(R_{T}\\approx 20\\rm\\, k\\Omega\\) over a structure surface of \\(A_{T} = 0.12\\rm\\, \\mu m^2\\). Multiple samples with different thicknesses have been produced. All samples have been examined on their tunneling behavior. The resistance area product as a function of thickness shows a linear behavior on a logarithmic scale. Furthermore all working samples show non-linear I-V curves as well as parabolic dI/dV-curves. Additionally the tunneling resistance \\(R_{T}\\) increases with decreasing temperature. All above mentioned properties are typical for tunnel barriers which do not include pinholes [Jonsson00]. The last part of chapter 3 deals with thermal properties of HgTe. By measuring the second harmonic of a biasing AC current in the channel below the tunnel barrier it is attempted to extract the diffusion thermopower of the heated electrons. Unfortunately the measured signal showed a far superior contribution of the first harmonic. According to electric circuit simulations a small asymmetry in the barrier (penetration and leaving point of electrons) could be responsible for this behavior. \n \nA ferromagnetic layer stack, consisting of PY/Cu/CoFe, serves as a sensor for magnetization changes due to external fields and current induced spin accumulations. The layer stack exhibits a giant magnetoresistance (GMR) which has been measured by a resistance bridge. The biggest peculiarity in depositing a GMR stack on top of HgTe is that its easy axis forms along only one of the crystal axes (\\((110)\\) or \\((1\\overline{1}0)\\)). The reason for this anisotropy is still unclear. Sources such as an influence of the terminating material, miscut, furrows during IBE or sputter ripples have been ruled out. It can be speculated that the surface states due to HgTe might have an influence on the development of this easy axis but this would need further investigation. A consequence of this unexpected anisotropy is that every CdTe/CdHgTe/HgTe/CdHgTe wafer has first to be characterized in SQUID in order to find the easy axis. A ferromagnetic resonance (FMR) measurement confirmed this observation. The shape of the ferromagnetic layer stack is chosen to be an ellipse in order to support the easy axis direction by shape anisotropy. Over 8 million ellipses are used to generate a SQUID signal of \\(m > 10^{-5}\\rm\\, emu\\). This is sufficient to extract the main characteristics of an average nano pillar under the influence of an external magnetic field. As in the case of bigger structures the ellipse shaped structure shows a step-like behavior. A measured minor loop confirms the existence of the irreversible anti-parallel stable magnetic state. Furthermore this state persists for both directions at \\(m=0\\) resulting in an anti-ferromagnetic coupling between Py and CoFe. \n \nThe geometry of the SOT-device is chosen in such a way that the current induced spin aligns either parallel or anti-parallel to the effective magnetic field \\(\\vec{B}_{eff}=\\vec{B}_{ext}+\\vec{B}_{aniso}+\\vec{B}_{shape}\\), which acts on the pillar. Due to interaction of the spin with the adjacent magnetization of Py the magnetization direction gets changed by a torque \\(\\vec{T}\\). In general this torque can be decomposed into two components a field-like torque \\(\\vec{\\tau}_{FL}\\) and a damping-like torque \\(\\vec{\\tau}_{DL}\\) [Manchon19]. In the case of TIs \\(\\vec{T}\\) is additionally depending on the z-component of \\(\\vec{m}\\) [Ndiaye17]. In our case the magnetization is lying in the sample plane (\\(m_{z}=0\\)) which results in \\(\\vec{\\tau}_{DL}=0\\). Thus, in the case of \\(\\vec{S}\\parallel\\left(\\vec{\\hat{z}}\\times\\vec{j}\\right)\\) and \\(\\vec{j}\\parallel\\vec{\\hat{y}}\\), the only spin dependent effective magnetic field is \\(\\vec{B}_{FL}=\\tau_{FL}\\cdot\\vec{\\hat{x}}\\) which is lying parallel or anti-parallel to \\(\\vec{B}_{eff}\\). The evaluation of \\(\\vec{B}_{FL}\\) can therefore be done in the following manner. First a high \\(B_{ext}\\) has to be set along the easy axis of the pillar. Then \\(B_{ext}\\) has to be reduced just a few \\(\\rm\\, Oe\\) before the switching occurs at the magnetic field \\(B_{ext,0}\\). At the magnetic field \\(\\Delta B = B_{ext}-B_{ext,0}\\approx 0.5\\rm\\, Oe\\) the lower resistive state should be stable over a longer time range (\\(10-30\\rm\\, min\\)) in order to exclude switching due to fluctuations. Now a positive or negative current can be pushed through the channel below the pillar. For one of the two current directions the magnetization of Py switches. It is therefore not a thermal effect that drives the change of \\(\\vec{m}\\). Current densities that are able to switch \\(\\vec{m}\\) at small \\(\\Delta B\\neq 0\\) lie in the range of \\(j\\approx 10^{4}\\rm\\, A/cm^{2}\\). In all experiments the switching efficiency \\(\\Delta B/j\\) decreases with rising \\(j\\). Furthermore the efficiency as a function of \\(j\\) depends on the temperature as \\(\\Delta B/j\\) values tend to be up to 20 times higher at \\(T=1.8\\rm\\, K\\) and \\(j\\approx 0\\) than at \\(T=4.2\\rm\\, K\\). This temperature dependence suggests that switching occurs not due to Oersted fields. Furthermore the Biot-Savart fields had been calculated for four different models: an infinite long rectangular wire, two infinite planes, a full volume and two thin volume planes. Every model shows an efficiency, which is at least three times lower than the observation. \n \nThe highest efficiencies in our samples show up to 10 times higher values than in heavy-metal/ferromagnets heterostructures. In contrast to measurement procedures of most other groups our method leads to direct determination of SOT parameters like the effective magnetic field \\(\\vec{B}_{FL}\\). Other groups make use of spin-transfer FMR (ST-FMR) where they AC bias their structure and extract SOT parameters (like \\(\\tau_{FL}\\) and \\(\\tau_{DL}\\)) from second harmonics by fitting theoretical models. Material systems consisting of TIs and magnetic insulators (MIs) on the other hand show 10 times higher efficiencies [Khang18,Li19]. In those cases the magnetization points out of the sample plane which is conceptually different from in-plane magnetic anisotropy geometries like in our case. The greatest benefit in-plane magnetic anisotropy systems is its easy realisation [Bhatti17]. Here only an elliptical shape has to be lithographically implemented instead of conducting research on the appropriate combination of material systems that result in perpendicular magnetic anisotropies [Apalkov16]. Despite the fact that in our case only \\(\\vec{\\tau}_{FL}\\) acts as the driving force for changing \\(m\\) our device still exhibits the up to date highest efficiencies in the class of in-plane magnetized anisotropies of all material classes ever recorded.