Ultrafast Magnetic Switching of Nanoelements with Spin Currents.doc

此文档收集于网络，如有侵权，请联系网站删除Ultrafast Magnetic Switching of Nanoelements with Spin Currents The goal of this project is the detailed understanding of the processes involved in switching the magnetization of a thin nanoscale magnetic element by a spin polarized current. A charge current of random spin polarization is sent through a lithographically manufactured pillar as illustrated in Fig.1, containing two magnetic layers shown in blue. The “polarizer” (darker blue), whose magnetization direction is in practice fixed by exchange bias (see Fig. 3 below), serves to create a spin polarized current either in transmission (middle of Fig. 3) or in reflection (bottom of Fig. 3), depending on the direction of current flow through the pillar. The polarizer is separated from a second ferromagnetic layer, the “sensor” (light blue), by a non-magnetic spacer layer typically made of Cu. The goal is to reliably switch the magnetization of the sensor layer by reversing the direction of current flow. Fig. 1 Schematic of a spin injection structure and the associated switching effects arising from current flow. The basic structure consists of a lithographically manufactured pillar composed of different thin layers. A spin polarizing ferromagnetic layer, the “polarizer”, shown in dark blue, has a fixed magnetization direction, in practice accomplished through exchange biasing (see Fig. 4 below). When the current flows through it toward a second ferromagnetic layer, the “sensor”, shown in light blue, the spin polarized transmitted current can switch the sensor layer by spin torque, as shown in the middle. When the current direction is reversed, a spin accumulation builds up in front of the polarizer through reflection of spins on the polarizer. The reflected spins have the opposite direction from the transmitted spins and can switch the sensor back to the original state.Spin injection utilizes the strong, short-range quantum mechanical exchange interaction of the injected spin polarized electrons with the atomic spins on the atoms in the FM sensor layer. In order to achieve a sufficient density of the injected spins one uses pillar diameters that are of the order of 100 nm. The current densities required for switching are of order 107 - 108 A/cm2. As proposed by Berger 1 and Slonczewski 2 the injected spin current exerts a torque on the magnetization of the sensor layer which can lead to switching of its direction. The existence of switching has been experimentally verified by GMR transport measurements by Myers et al. 3. In such measurement only the existence of two different magnetization directions of the sensor layer can be detected, while the details of the magnetic structure in the sensor layer are not directly revealed. Our studies are aimed at elucidating the details of the spin injection process that remain hidden in GMR measurements which compare the resistance between two switched states. There is good reason to believe that the switching process and even the final switched states are more complicated than expected from spin injection alone. One of the intriguing questions associated with the spin injection process is the effect of Oersted fields that necessarily accompany the flow of a (charge) current. Such Oersted fields are present inside and outside the pillar, as shown in Fig. 2. The fields are actually zero at the center of the pillar but have maximum value at its periphery. The curl associated with these fields may actually play an important role in initiating the spin injection switching process and it may result in a non-uniform switching process and a resulting curled magnetization in the sensor, as has been suggested based on simulations 4. The latter is, of course, highly undesirable, since curled states reduce or even eliminate GMR.Fig. 2 Schematic of a spin injection structure as in Fig. 1, and the associated spin and charge effects arising from current flow. The spin current is accompanied by a charge current which gives rise to Oersted fields inside and outside the pillar, as shown. The Oersted fields have a closed-loop form and are maximum at the perimeter of the pillar. The curl associated with these fields may lead to non-uniform switching with a resulting curled magnetization in the sensor.The most difficult part of our project was the manufacture of the nanoscale spin injection samples. Our sample preparation approach is based on optical and e-beam lithography techniques and is carried out by graduate students using facilities on Stanford campus. This approach has the educational benefit that students are being trained in nanofabrication. We have successfully fabricated a variety of structures that are based on the concept of current flow through a metallic pillar embedded in an insulator. The pillar always contains the magnetic polarizing and sensor layers. Leads fabricated by optical lithography are connected to the two ends of the pillar. An example of a structure fabricated at Stanford is shown in the plan view image in Fig. 3, recorded by scanning transmission x-ray microscopy. The caption explains the current flow through the structure. The small oval area which is shown enlarged in the inset indicates the pillar (topological contrast) through which the current flows from the bottom to the top of the structure. The pillar axis is oriented perpendicular to the shown view. Fig. 3 Scanning x-ray microscopy image of a lithographically manufactured spin injection sample. The voltage is applied between the two vertical and the two horizontal leads. The vertical leads go to the bottom of the pillar and the horizontal ones to the top of the pillar. The pillar location is seen as a small oval structure in the middle of the intersection and is enlarged in the upper left. In the case shown it had a dimension of 100 nm x 300 nm. The current flows up through the pillar (perpendicular to the picture plane). The details of a typical pillar structure are shown in Fig. 4. The detailed structure of the spin injection pillar, the central part of the experimental sample, is shown in Fig. 4. The figure caption gives the details of the various layers. We used a pillar cross section of 100 nm x 150 nm and employed exchange bias pinning to fix the direction of the polarizer, as shown in the figure. The current direction through the pillar can be varied. Fig. 4 Detailed structure of the nano-pillar used in our spin injection experiments. The sample was fabricated by J. Katine at Hitachi Global Storage systems, and the pillar had an oval cross section of 100 nm x 150 nm. A 17.5 nm thich PtMn AFM layer (green) was used to pin the lowest 1.8 nm thick Co0.86Fe0.14 FM (blue) which was coupled antiferromagnetically through a 0.8 nm thick Ru spacer layer to a second 2.0 nm thick Co0.86Fe0.14 layer (blue). This AFM coupling between the two FM layers reduces stray fields. The spin polarization of the injected current flowing through the pillar (direction variable) is determined by the magnetization (red arrow) of the second FM, referred to as the “polarizer”. It then enters a 4.0 nm thick Co0.86Fe0.14 FM layer, referred to as the “sensor” (light blue), whose magnetization is imaged by x-rays. We also indicate that the electric current gives rise to an Oersted field inside the pillar (see Fig. 2). If the current flows from bottom to top, as shown, the current gets polarized by the second dark blue layer in the direction of the red arrow. The spin polarized current then enters the light blue sensor layer and excites its magnetization. If the current flows from top to bottom the current becomes spin polarized by the light blue layer. It then tries to enter the dark blue layer below, which has a fixed magnetization direction. Only spins with the proper polarization are allowed to enter the ferromagnet while the others are reflected. These reflected spins now act back on the light blue layer, and since it is not pinned, can switch it. Note that the directions of the transmitted spins and reflected spins are opposite, so that the sensor layer can be switched in opposite directions by inverting the direction of current flow. GMR transport measurements performed on two pillars of different dimensions are shown in Fig. 5. In (a) we show GMR results for a 100 nm x 50 nm pillar of the structure in Fig. 4. The resistance is shown as a function of the current in the absence of an external field, and in the inset, as a function of an applied external field. Both spin injection and external field switching take the sample from one to another well defined state, which are associated with opposite magnetization directions of the sensor layer. In contrast, as shown in (c), the larger 100 nm x 150 nm pillar shows the existence of the same two well-defined states in an external magnetic field but the spin-injection loop has an unexpected transition from the parallel to some intermediate configuration at the point labeled (3) in (c). Fig. 5. (a) and (c): Current induced magnetization reversal in a 100x50nm2 and 100x150nm2 pillar structures, respectively, seen as a change in the magnetoresistance as a function of current. Insets show the magnetoresistance as a function of an external magnetic field. The points labeled (1) and (2) correspond to the transition from parallel to anti-parallel and anti-parallel to parallel configurations, respectively. Point (3) in (c) is an unexpected transition from the parallel to some intermediate configuration. (b) and (d): Switching behavior as a function of a field or spin injection pulse. (A) and (B) show the parallel and anti-parallel configurations obtained by applying an external field. Subsequent changes in the magnetoresistance are induced by current pulses. The “set pulse” causes the transition from (B) to (C) and the “reset pulse” causes the transition from (C) to (D). For the 100x50nm2 device the switching occurs between states that are reached by either external field pulses or spin injection pulses. In contrast, for the 100x150nm2 sample, spin injection set pulses lead to an intermediate configuration, while the reset pulse drives the magnetization to the same state as an external field. The behavior of the loops in (a) and (c) is confirmed by switching cycles plotted as a function of time in (b) and (d) for the same samples. The 100 nm x 50 nm pillar switches equally well between two states under application of field and spin injection pulses. The 100 nm x 150 nm pillar switches fully under application of an external field but only partially to an intermediate state when a spin injection “set pulse” (B to C) is applied. The “reset pulse” (C to D) drives the magnetization to the same state as an external field. The nature of the intermediate state cannot be determined by transport measurements. The only way to obtain information on the magnetic structure of this intermediate state is by means of imaging of the magnetization in the sensor layer. X-rays provide this capability because they offer the necessary spatial resolution and have the ability to “see” a buried layer. In principle, it would be advantageous to use a different FM for the polarizer and the sensor layers because one could then distinguish them directly through the element specificity of x-rays associated with the different energies of absorption edges. In our case, all FM layers consisted of the same material Co0.86Fe0.14, as shown in Fig. 4. The reason was the difficulty of having enough sputter targets in the deposition of the various layers in the devices. However, this fact did not lead to difficulties in distinguishing the magnetic contrast from the sensor layer. First, the presence of two polarizing layers of nearly equal thickness and opposite magnetization directions approximately cancelled the magnetic signals from the two dark blue layers. Second, the reliable pinning of the layers by exchange bias prevented any change in their magnetic orientation as a function of current, so that any observed change in magnetization was due to the sensor layer. We fabricated oval pillar structures of 100 x 150 nm2 dimension and composed of the layer structure shown in Fig. 4 on a SiNx coated Si wafer. In the last lithography step, a 100 x 100 mm2 window was etched from the back through the Si wafer so that the pillar sample was supported only by the remaining SiNx membrane. Thus x-rays could be sent through the entire pillar structure, as illustrated in Fig. 6. X-ray images of the sample were recorded by means of scanning transmission x-ray microscopy (STXM) on beam line 11 at the ALS with a spatial resolution of about 30 nm, limited by the zone plate. The beam line is equipped with a variable polarization undulator and we used circular polarized x-rays that were incident at an angle of 30 from the surface normal of the sample. This geometry resulted in a finite projection of the photon angular momentum along the in-plane magnetization direction of the FM sensor layer in the pillar. Magnetic contrast was obtained by tuning the x-ray energy to the Co L3 edge (780 eV). By recording images for two orthogonal azimuthal sample orientations relative to the fixed x-ray direction, we could construct arrow plots for the in-plane magnetization directions in the 4 nm thick Co0.86Fe0.14 sensor layer. In order to isolate the small magnetic signal from the 4 nm thick Co0.86Fe0.14 sensor layer from the large background from the other layers, with a combined thickness of about 230 nm, we used a pump-probe technique. Opposite polarity current “pump” pulses of 4 ns duration and separation were synchronized with the x-ray pulses from the storage ring, as illustrated in Fig. 6. The x-ray “probe” pulses of about 100 ps FWHM were separated by 2 ns, as shown in the figure. For a given sample position in the beam, we recorded the transmitted x-ray intensity separately for eight consecutive x-ray probe pulses which could be arbitrary shifted relative to the current pump pulses. This was made possible by use of ultrafast electronics and a fast avalanche photodiode detector. Such pump-probe cycles were repeated to accumulate adequate signal to noise ratios. The sample was then scanned to a new position and the procedure was repeated until a complete set of magnetic images of the sensor layer were recorded as a function of pump-probe delay times. Fig. 6. Left: Schematic of the pillar structure, showing the ferromagnetic layers in blue, the antiferromagnetic pinning layer in green and the Cu leads and spacer layers in orange. The bottom two FM layers are coupled into a fixed antiferromagnetic arrangement by a Ru spacer layer and their magnetization direction is pinned by exchange coupling to the green antiferromagnet shown at the bottom. The incident x-ray beam is incident 30 from the surface normal and is focused by a zone plate to a size of about 30 nm. The transmission through the structure as a function of sample position is monitored by an x-ray detector. Right: Timing scheme used for the pump-probe spin injection experiments. A 4 ns pump current pulse, “set pulse”, is followed by a “reset pulse” of the same length but opposite polarity and this scheme is repeated. The x-ray probe pulses are synchronized to the pump pulses and for each pump-probe cycle the intensities of eight adjacent x-ray pulses (2 ns separation) are measured by 8 separate photon counters. This procedure is repeated for a given sample position in the beam until the signal-to-noise ratio is adequate. The sample is then scanned to a new position in the beam and the procedure is repeated. The pump-probe delay times can be arbitrarily chosen to obtain complete time dependent movies of the magnetization dynamics. The reconstructed arrow plots for the magnetization of the sensor layer at nine specific times in the pump-probe cycle are shown in Fig. 7. These results indicate that only the reset spin injection pulse leads to a uniform magnetization direction, which is shown in (i) and (a). This is in agreement with the GMR transport results shown in Fig. 5 (d), which show that the “reset pulse”, causing the transition from (C) to (D), leads to a well defined magnetic state which is also reached by magnetizing the sensor layer uniformly in an external magnetic field. In all other cases shown in Fig. 7 we find a curled state, indicative of the influence of the Oersted field. The sense of the curled state observed during the set and reset pulses is opposite, as expected. The curled states observed during the set pulse, shown in (c), and after the pulse, shown in (d), are almost identical. This shows that the magnetization remains in the same curled state when the field is switched off. This state corresponds to the intermediate state (C) in Fig. 5 (d). The importance of our results lies in the fact that spin injection is clearly accompanied by Oersted fields for samples of size near 100 nm. This is an unexpected result. The Oersted fields are also found to have important consequences. They not only dominate the dynamical evolution as indicated by the images taken during the current pulse but they can also lead to unique non-uniform intermediate states. In particular, the observed intermediate curled state is quite unique. It has never been seen in field-induced magnetization states, and arises from the interplay between the uniform spin-injection and the curled Oersted contributions. Fig. 7. Measured time dependent magnetization directions indicated by arrows within the 4 nm thick Co0.86Fe0.14 sensor layer inside the 100