Spintronics: Poised for Next Great Memory Breakthrough?

In 1997, IBM incorporated the spin-value sensor based on “spintronics” to the disk drive, which went on to increase the storage capacity of disk drives a thousand fold.¹ A decade later, the spin-valve sensor has been supplanted by magnetic tunnel junction (MTJ) devices, which provide even greater sensitivity for magnetic recording read heads. This article reviews the basics of spintronics, and briefly discusses the basic structure and physics of magnetic tunnel junctions and their application to field sensors and high-performance non-volatile magnetic random access memory (MRAM).

Over the past two decades, a new field of condensed matter physics — colloquially termed “spintronics” — has arisen in which the properties of nanodevices depend on the controlled passage of spin-polarized electrical current. While conventional microelectronic devices rely on the passage or storage of electrical charge, it is electron's quantum-mechanical property of spin that is harnessed in spintronics. Spin has only two values, “up” or “down.” When the spins of many individual electrons become oriented parallel to one another, either all up or all down, materials exhibit magnetism.

Some common metals, such as iron, cobalt and nickel, are magnetic at room temperature (and at much higher temperatures), thereby making them useful for device applications. Although the magnetism of these metals and their alloys has been known for centuries, the development of advanced thin-film deposition techniques about 20 years ago led to the exploration of magnetic heterostructures formed from atomically thin layers of magnetic and non-magnetic materials. Such materials possessed a range of unusual properties, the most important being magnetoresistance changes in resistance exhibited in response to magnetic fields.

Many materials exhibit large magnetoresistance effects but usually at very low temperatures (~4.2 K) and very large magnetic fields (several tesla). Moreover, these first structures were deposited using the elaborate deposition technique of molecular beam epitaxy (MBE).

The magnetoresistance effect found in single-crystalline superlattices formed from alternating layers of iron and chromium was much larger than that exhibited by iron layers themselves, and so was termed giant magnetoresistance (GMR).² However, soon afterwards it was discovered that the same phenomenon was observed in polycrystalline multilayers of iron and chromium prepared by the much simpler technique of magnetron sputter deposition.³

Exploration of large numbers of magnetic multilayers led to the discovery of GMR effects at room temperature in cobalt/copper multilayers,⁴ and the oscillatory dependence of GMR magnitude on the thickness of the non-magnetic spacer layer.^3,5,6 This oscillatory variation was shown to be related to a long-range oscillatory interlayer exchange coupling of the magnetic layers through the spacer layers, caused by the formation of a spin-density wave in the metallic spacer layers.

By taking advantage of spacer layer thicknesses for which the oscillatory interlayer coupling is antiferromagnetic in character — that is, the magnetization of the magnetic layers on either side of the spacer layer naturally aligned anti-parallel to one another — useful structures could be spin-engineered to build extraordinarily sensitive sensors of small magnetic fields.^7,8

Another important technique used to magnetically engineer magnetic heterostructures is the use of exchange-bias, a phenomenon observed more than 50 years ago.⁹ A magnetic material displays a magnetic moment in the absence of magnetic field, but the direction along which the magnetic moment is pointed can be reversed by applying a sufficiently large magnetic field — the magnetic coercive field. Usually, the strength of the field required to reverse the magnetization of the material is independent of the direction along which the magnetic field is applied (whether positive or negative), but in some cases this strength can be different for positive and for negative field directions so that the system exhibits a unidirectional magnetic anisotropy.

Exchange-bias results when a magnetic material is in atomic proximity to an antiferromagnetic material. Further discussion of spin-engineering is beyond the scope of this article, but we illustrate basic concepts and structures in Figure 1.

1. Composed of two magnetic layers, pinned and free elements, spintronic devices have been engineered for memory (a), for sensors (b), in its simplest form (c), for stability and reduced magnetic coupling (d), without an exchange-bias layer (e), where both elements consist of AF-coupled pairs (f), and as a double tunnel junction (g).

Fixing magnetization's direction

All useful spintronic nanodevices currently use the basic spin-engineering concepts of exchange anisotropy to fix the direction of the magnetization of one or more magnetic layers in a multilayered structure so that their moments are little affected by magnetic fields applied during operation of the device and the artificial antiferromagnetic concept.^1,7,10-12 The idea is to eliminate the otherwise very large magnetic dipolar fields arising from magnetic poles formed at the edges of spintronic nanodevice structures. The latter fields increase approximately inversely with the size of the nanodevice so that, for devices in the deep submicron dimension regime, these dipolar fields become extremely large.

In ferromagnetic metals, the electrical current is carried largely by electrons of one spin orientation, either up or down, depending on the magnetic material. This is due to the fact that electrons whose spin is oriented parallel to the local atomic magnetic moments of the material are scattered differently from those whose spins are anti-parallel to these moments. The current in ferromagnetic metals is thus intrinsically spin-polarized.

When one forms a sandwich of two magnetic layers separated by, for example, a thin layer of non-magnetic and highly conducting (i.e., little electron scattering) copper, the current flowing through the sandwich depends on the relative magnetic orientation of the moments of the magnetic layers. When these moments are parallel to one another, current can flow easily because electrons of one spin orientation are little scattered anywhere within the sandwich. Conversely, when the moments are aligned anti-parallel, electrons of either spin orientation are scattered in one or other of the magnetic layers, increasing the resistance of the sandwich. This is the basic origin of GMR. In useful spin-engineered structures, often called spin-valves, it turns out the GMR effect is quite modest, ~10% at room temperature.

A closely related structure to the spin-valve, the MTJ device,¹³ is fundamentally a sandwich of two magnetic layers separated by an ultrathin dielectric layer (Fig. 2). When a voltage is applied across the insulating dielectric layer, current flows via quantum mechanical tunneling. When the magnetic moments of the two magnetic layers are parallel to one another, the current is spin-polarized.

2. Current of spin-up and spin-down electrons tunneling across a barrier is spin-polarized when the magnetic moments of the magnetic electrodes are parallel to one another (left) and unpolarized when the moments are anti-parallel (right).

The origin of the spin polarization and, in particular, what determines the magnitude of the degree of spin polarization, and how it depends on the magnitude of the voltage across the MTJ, is currently a subject of considerable interest. When the magnetic moment direction of one of the magnetic layers of the MTJ is reversed, the current through the device is reduced. This gives rise to tunneling magnetoresistance (TMR). The amount of the reduction in current flow depends on the degree of spin polarization of the current. Indeed, if the current were perfectly spin-polarized, composed only of spin-up or spin-down electrons, no current would flow through the device when the magnetic moments in the MTJ were aligned exactly anti-parallel. Such a device would be a perfect electrical switch.

Although such a switch has not yet been achieved, there have been remarkable developments in the understanding of spin-dependent tunneling. Over the past decade, the magnitude of the TMR effect has increased from just a few percent at room temperature to several hundred percent.¹³

The biggest breakthrough during this period was the discovery that the spin polarization of the tunneling current depends not solely on the electronic properties of the magnetic electrodes, as long thought, but rather is strongly influenced by the dielectric material that forms the tunnel barrier of the MTJ.^14,15 Indeed, by changing the dielectric material from a thin amorphous layer of aluminum oxide (Al2O3) to a crystalline layer of magnesium oxide (MgO), the degree of spin polarization of the current increases from ~50% to ~90% for the same magnetic electrode material (an alloy of cobalt and iron).¹⁴ At room temperature, the TMR has correspondingly increased from ~70% to >350% in useful spin-engineered devices. The TMR effects are so large that MTJ devices have replaced spin-valve field sensing devices in magnetic hard disk drives over the past 18 months so that nearly all disk drives manufactured today use MTJ read heads.

Another important development has been the demonstration that spin-polarized current can itself be used to reverse the direction of a magnetized layer without the need for any magnetic field. This effect, which was first proposed in 1996,^16,17 is now commonly called spin torque reversal or spin momentum transfer (SMT).

The origin is, in principle, very simple. Associated with the electron's spin is a spin angular momentum, which, just like the angular momentum of a rotating mass, must be conserved. If a large enough current of spin-polarized electrons is injected from one rigid magnetic layer across a tunnel barrier (or a metal spacer layer in a spin-valve device) into a second magnetic layer, the moment of this second layer can be excited and caused to undergo precessional motion, and even can be completely reversed if a large enough current is applied for a sufficiently long time. This effect was first observed in a spin-valve structure,¹⁸ but its most useful application is likely to be in MTJ-based cells.

The current-induced switching of an MTJ device can be compared to changing the state of the same device using conventional means, namely by using an external magnetic field (Fig. 3). The device is switched between a high resistance state (magnetic moments on either side of the MgO tunnel barrier are anti-parallel) to a low resistance state (parallel moments) either by field or by current.

3. Switching of the magnetic state of an MTJ is shown with an induced external magnetic field (left) and a spin-polarized current (right), and resistances are compared. The TMR is ~103% with device area ~80 x 160 nm2 (inset).

Figure 3 also shows that the MTJ device can exhibit both a low and a high resistance state in zero magnetic field (and current). This demonstrates that the MTJ can be used as a nonvolatile memory element where the data is encoded in the resistance of the device — i.e., as the direction of magnetization of the upper magnetic layer.

In the original proposal for an MTJ-based MRAM, the MTJ memory elements were written by applying local magnetic fields (Fig. 4). The basic structure of the MRAM is a cross-point array of copper wires (write lines and bit lines), where the MTJ elements are inserted at each cross point between the lower bit line and the upper write line. When a current is passed along one of the bit lines (without passing through the MTJs), it generates an Oersted magnetic field that all the MTJs along that bit line will be subjected to. A particular MTJ is written by simultaneously passing a current along the corresponding write line. The combination of the bit and write line fields at the chosen cross-point junction is used to set the magnetic state of the MTJ. By changing the direction of the write current, the MTJ can be written to its low or its high resistance state.

4. The schematic diagram shows a cross-point MRAM device (field written).

The MTJ elements can be read by passing small currents through the MTJ itself wherein the larger the TMR effect, the easier it is to distinguish the two resistive states of the MTJ device. However, the reading performance is lowered by loss of current to the array of neighboring MTJs, which are all connected in parallel. The reading performance is improved by connecting a transistor in series with each MTJ device to ensure the read current passes only through one device. This is the basic architecture of an MTJ-based MRAM (Fig. 5). The first MRAM prototype was built by IBM in 1999.¹² Today this technology is being pursued by several major companies and institutions in the United States, Japan, Europe, Korea and Taiwan.

5. A cross-section of a 16 Mb MRAM chip ~7.9 x 10 mm2 in area and fabricated using IBM’s 180 nm CMOS 7sf technology, shows metal layers M1, M2 and M3. The area corresponding to one MTJ memory cell is 1.42 mm2. (Source: IBM)

Larger transistors would increase the memory cell size and cost. To be competitive, the SMT current must be reduced. Given the enormous strides made in our understanding of SMT writing in the past five years, it seems highly likely that new structures and materials will emerge that will allow for SMT writing with a small-footprint MTJ memory cell. If this is accomplished, MTJ MRAM promises to offer a unique combination of nonvolatility, high performance and high endurance at modest cost compared with other emerging nonvolatile RAM technologies.

The future is very bright for the field of spintronics. Spintronic sensing devices have revolutionized our ability to store massive amounts of digital data at very low cost, effectively providing the modern world with instant access to information no matter where we are. Magnetic disk drives, however, are slow, energy-consuming, and not very reliable. A new technology, racetrack memory, which uses some of the spintronic concepts, promises to provide a data storage device that combines the low cost of a magnetic disk drive with the high performance and reliability of conventional RAM within a few years.¹⁹