The pinning of domain walls in ferromagnetic (F) wires is one possible technique for the creation of a solid state magnetic memory. Such a system has been under consideration for some time but one of the main limitations is the control of, and non-uniformity of the domain wall pinning. Techniques such as the lithographic definition of notches and steps in the substrate have had some success in creating local pins but have the disadvantage of being expensive to fabricate and the reproducibility of the domain wall pinning strength is limited. In this letter, we report on an alternative strategy to create pins of reproducible strength using crossed ferromagnetic and antiferromagnetic (AF) wires such that exchange bias can be introduced at the crossing points. Such a system has the advantage of ease of fabrication and creating domain wall pins of controlled strength by varying the width of the AF wire. We have achieved domain wall pinning field strengths of up to 37 Oe in a system where the AF wire is deposited above the F wire which is comparable to the values achieved using notches.
Magnetic recording has been the most successful method for data storage over the last decades. In particular, magnetic tapes are the most reliable and economic form of archival storage, followed by hard disk drives (HDD) for large-capacity storage in personal computers and cloud servers. Magnetic random access memory (MRAM) has been proposed and commercialised with the advantages of non-volatility, nanosecond access time, reliability in severe environments, and compatibility with current Si-based technology.1 However, the unit cell of an MRAM device requires a capacitor or transistor for reliable operation. Generally, the applications of MRAM are restricted to severe environments, e.g., high temperatures and high levels of electromagnetic radiation.
In 2008, Parkin et al. proposed a racetrack memory which has all the advantages of MRAM and all metallic semiconductor free structure.2 Racetrack memory consists of an F wire, typically 0.1–1 μm wide, where a magnetic domain wall can be injected and detected. A 180° domain wall carries a data bit via its configuration of either north to north or south to south poles. The domain wall is shifted along the wire by a pulse of electric current due to the spin-transfer torque effect theoretically proposed by Slonczewski3 and demonstrated by Yamaguchi et al.4 Parkin et al. reported the successful operation of a 3 bit chip in 2008 (Ref. 5) and arrays of 256 fully operational racetrack cells in 2011.6 A key issue in the reliable operation of a domain wall memory is the uniformity of the pinning of the domain walls. The pinning sites in the devices of Parkin et al. were created via a notch cut in the F wire using electron beam or photolithography.2,5 The notches suffered from variations in shape leading to non-uniform domain wall pinning strengths. They also add complexity to the fabrication process and hence the system becomes expensive to produce.6
In this letter, we propose and demonstrate an alternative simple pinning scheme in straight wires using exchange bias. We have patterned a series of AF wires above and below F wires and perpendicular to them.7 The AF wires have different widths. When field cooled the AF wires induce exchange bias at the AF/F crossing points which act as pinning sites. Due to its simplicity of design and fabrication, our racetrack memory concept has significant advantages over conventional designs and may accelerate the development of this technology.
Samples were fabricated by first creating a mask using photolithographic patterning and subsequently depositing the wires by sputtering, i.e., a lift-off process using acetone as a solvent. A thermally oxidised Si substrate was cleaned in acetone and isopropanol for 3 min in an ultrasonic bath. Photoresist (Shipley, S1805) was then spin-coated onto the surface of the substrate at a speed of 4000 rpm for 60 s giving a resist thickness of 500 nm. A series of 30 μm long AF wires with widths of 1.0, 1.5, 2.0, and 2.5 μm were patterned by photolithography (S.E.T., MA-750). A Hg lamp (405 nm wavelength) was used with an exposure time of 2.8–3.5 s to activate the resist. After exposure the patterns were developed by dipping into Microposit MF-319 developer for 90 s.
Polycrystalline seed and AF layers consisting of NiCr(6 nm)/IrMn(5 nm) were then grown on the patterned Si substrate by high vacuum sputtering (Plasma Quest, HiTUS).8 The base pressure of the system was 3 × 10−7 mbars and the growth pressure was 3.77 mbars. From previous studies on the NiCr/IrMn system, the grain size in the films will be approximately 5 nm and the anisotropy of the AF layer will be of the order of 107 erg/cc.9 The coercivity of unpatterned films of CoFe grown in our system but without an AF layer lies in the range of 50 Oe–100 Oe.
The same process was repeated to fabricate a 20 nm thick, 1 μm wide, and 55 μm long CoFe wire bridging across the AF wires. At one end of the F wire, a 15 μm × 15 μm square pad was defined for the introduction of domain walls. At the other end, a sharp point was defined to allow for domain wall annihilation. Electrical contact pads of Ti(10 nm)/Au(100 nm) were fabricated above the ends of all AF and F wires.
By reversing the order of the AF and F layers deposition, a racetrack memory with a top-bias configuration was also fabricated with the same layer thicknesses. The samples were field cooled from 225 °C in a field of 20 kOe either along the direction of the AF or the F wire to set the order in the AF grains and thereby induce the exchange bias. It should be noted that IrMn has a Néel temperature of 730 K and hence cannot be set above TN without damage to the film. Hence, the setting process is by thermal activation.10
Unpatterned continuous films with layers of the same thickness were measured using a vibrating sample magnetometer (VSM, ADE, Model 10). Systems with both top and bottom bias structures were grown and field cooled in 20 kOe from a temperature of 225 °C.
A magneto-optical Kerr effect (MOKE) magnetometer was used for the measurements of the magnetic behaviour of the cross wire systems. A 532 nm wavelength continuous wave laser beam was linearly polarised and focussed onto the sample at an angle of incidence of approximately 45° out of the plane of the F wire but along the same direction. This angle gives an elliptical laser spot with dimensions 7 μm × 4 μm. Effectively this is the spatial resolution of the MOKE but the sample stage can be adjusted in the x and y directions in the plane to within 100 nm. The optical polarisation of the reflected light was analysed to convert polarisation changes to intensity changes for detection. The longitudinal MOKE configuration was used with the F wire long axis parallel to the sensing direction. Magnetic fields of maximum amplitude 300 Oe were applied to the samples along the direction of the F wire at a frequency of 27 Hz.
The magnetisation curves for the unpatterned films shown in Figure 1 exhibit a loop shift of Hex = 46 Oe induced by the exchange bias at the AF/F interface. This result is somewhat unexpected as the presence of the seed layer beneath the AF for the bottom bias structure was expected to increase the value of the loop shift (Hex) due to the improved (111) in-plane texture of the IrMn.9
Figures 2(a) and 2(b) show a schematic diagram of the racetrack memory together with an optical image of one of the devices fabricated. Domain walls were generated in the square pad and injected into the F wire by applying a magnetic field along the wire direction. The domain walls became pinned at junctions where the AF wires cross the F wire.
Figure 3 shows the hysteresis loops for the bottom bias device at the points along the F wire indicated in Figure 2. From the data, the field required to inject a domain into the pad and from there into the wire can be seen. The field values are shown in Figure 4. The data in Figures 3 and 4 clearly show an increase in the coercivity with the increasing width of the AF wires. This indicates that the domain wall pinning strength is dependent on the width of the AF wire. It should be noted that for an identical device which had not been field cooled a very small domain wall pinning was observed at the crossing points of the F and AF wires. Typically, this was of the order of 5 Oe which is almost within error of the field measurement. This confirms that it is the exchange bias effect that gives rise to the domain wall pin and not simply the presence of the AF wire. Any domain wall pinning due to the unaligned AF wire would therefore be of a similar strength to the inherent domain wall pinning strength in the wire.
Hysteresis loops for the bottom bias device measured at the points indicated in Figure 2.
Reproducibility of the coercivity at the points labelled in the racetrack memory with bottom bias (full circles) and top bias (empty circles) configurations. Also shown is data (triangles) for a device where the AF was not set.
Reproducibility of the coercivity at the points labelled in the racetrack memory with bottom bias (full circles) and top bias (empty circles) configurations. Also shown is data (triangles) for a device where the AF was not set.
Figure 4 shows the AF wire width dependence of the coercivity (Hc) for the 3 μm wide F wire. For the case of the bottom bias configuration, the domain wall nucleation field in the square pad was 107 Oe and the domain wall injection field into the wire was 110 Oe. The coercivity of the F wire adjacent to the 1 μm wide AF (point c in Figure 2(a)) wire was then measured to be 113 Oe giving a pinning field for the 1 μm wide AF wire of 3 Oe. The pinning fields for the 1.5, 2.0, and 2.5 μm AF wires were measured to be 6, 9, and 12 Oe, respectively. This is significant for future device development as it shows that the strength of pins can be controlled via the AF wire width.
In order to remove any domain wall pinning effects due to the F wire crossing the AF wires and therefore not being fully planar, a top-bias device was also measured with identical layer and wire dimensions. The domain wall nucleation and injection fields were measured to be 65 and 110 Oe, respectively, which agrees with the values for the bottom bias device. The pinning fields in this device for the 1.0, 1.5, 2.0, and 2.5 μm AF wires were measured to be 15, 19, 17, and 37 Oe, respectively. These values are significantly larger than those for the bottom bias device. Again these results verify that such a structure is capable of producing domain wall pins of controlled strength. The value for the 2.5 μm wide wire is particularly significant as it is comparable to the values achieved by Parkin et al. for a notched wire system.2
Figure 4 shows the values of Hc measured repeatedly at the points on the F wire indicated in the schematic diagram included in the figure. It should be noted here that the MOKE magnetometer operates at a frequency of 27 Hz and hence each value measured is therefore an average over many loops. Each point on the graph is a separate measurement of Hc. The data in Figure 4 show that the measured values of Hc are remarkably consistent. This indicates that the strength of the induced domain wall pins is reproducible. Also shown in the figure are data measured on a sample that was not field cooled showing that an unset AF induces only a very small degree of pinning which increases only slightly with the wire width.
We have also measured the domain wall pinning with the AF wire set along the direction of the F wire. This data is shown in Figure 5. In this case no reproducibility study was undertaken. Similar levels of pinning are observed, in particular, for the top bias device, which lie within the range of variability that we observe for different devices. We believe this variability is the result of our use of a relatively low resolution optical lithography system. For this alignment of the AF along the F wire direction, the pinning of the domain walls is again much smaller indicating that a top bias configuration is preferable as it keeps the F wire in the plane of the film.
We have also undertaken some initial measurements of the current density required to move a domain wall past an exchange bias pin. A typical current density of 5 × 107 A cm−2 is sufficient to overcome a pin compared with values of circa 108 A cm−2 reported by Parkin et al. for a notched system.2 In our measurements, a continuous current was used whereas Parkin et al.2 used a pulsed current system.
In our studies to date, the strength of the domain wall pins is modest but, importantly, controllable. The strength of the exchange bias effect, represented by the loop shift Hex, varies as the inverse of the thickness of the F layer tF (Hex ∼ 1/tF). Hence stronger pinning is expected in thinner F wires.
It is also known that in general the domain walls in narrow wires lie at an angle to the direction of the wire. In our system, it is possible to vary the direction of the exchange bias at the crossing points of the F and AF wires by changing the angle of the field applied during the annealing step. Our initial data indicates that for 3 μm wide F wires the direction of the setting field gives no significant effect.
To date, the development of racetrack memory has been limited by the inability to produce domain wall pinning sites in the nanowires. Using the method presented in this letter, domain wall pins of well defined and reproducible strength can be created and controlled. Further studies will follow on devices with F wire widths down to 100 nm and having thinner F wires so as to vary the exchange bias.
This work was partially supported by EPSRC (EP/I000933/1). I.P. is grateful to JEOL UK Ltd., for support of his studentship. A.H. thanks the Royal Society and Hitachi Cambridge Laboratory for his Industrial Fellowship. T.J.H. acknowledges EPSRC for his Career Acceleration Fellowship (EP/J002275/1). We acknowledge the contribution of R. Carpenter to the device design.
