There are times when research progress is incremental or serendipitous. Then there are those few occasions when a leap is made by one or more groups of clever and motivated researchers. This is the case for the recent development of magnetic tunnel junctions (MTJs) that use a single-crystal MgO tunnel barrier.

One of the most active areas in materials science is magnetoelectronics. This field focuses on the development of integrated devices that use the spin orientation of conduction electrons in order to achieve new functionalities. Typical structures incorporate patterned thin films of transition metal ferromagnetic elements. One success is the spin-valve sensor used in read heads. These have been used in hard disk drives since 1998 and are largely responsible for the great increase in storage density ever since. Spin valves have a resistance of ∼50 Ω that changes by a fractional amount (the magnetoresistance, or MR) of ∼12% in the modest fields associated with bits recorded in magnetic media. Read heads are packaged, sold, and used individually, and there is a large research effort aimed at more highly integrated applications, such as nonvolatile magnetic random access memory (MRAM). The development of MTJs with AlO barriers has enabled MRAM prototypes of a few megabits. These MTJs show spin-dependent tunneling (SDT) with room-temperature MR values of 50-70%. Prototype MRAM chips show very high performance with metrics that are superior to static random access memory and floating gate memories. But scaling problems associated with cell size and ferromagnetic hysteresis stand in the way of high-density MRAM.

The accepted model for SDT is named after the man who, in 1975, made the first MTJ and analyzed its tunnel magnetoresistance (TMR). The Jullière model is approximately correct for MTJs that use amorphous tunnel barriers such as AlO. The conduction band of each ferromagnetic electrode has an up-spin and a down-spin subband density of states at the Fermi level. Because they differ, an electric current in the ferromagnet has a fractional spin polarization P < 1. SDT is given by a simple calculation that depends on the up- and down-spin densities of states of the two electrodes. For transition metal ferromagnetic materials commonly used in MTJs, P ∼ 40-50% and the TMR is 50-70%.

Motivated by preliminary experiments using MgO barriers, a group of theorists challenged the simple Jullière model. In their analysis [Butler et al., Phys. Rev. B (2001) 63, 054416; Mathon and Umerski, Phys. Rev. B(2001) 63, 220403R], each ferromagnetic electrode has conduction electrons that are described by Bloch waves, and each of these Bloch waves has a unique evanescent wave (and decay length) in the single-crystal MgO barrier. The polarization of the conduction electrons that cross the barrier depends on the symmetry of the wavefunction and the characteristics of the evanescent wave. Bloch waves with momenta perpendicular to the barrier may be characterized by high spin polarization and relatively high transmission, whereas Bloch waves with transverse momentum components may have low spin polarization and be preferentially reflected by the barrier. In this way, the TMR can be much larger than 70% and the spin polarization larger than 50%.

At about the same time, a group at IBM was working on MgO barriers. In a beautiful set of experiments [Parkin et al., Nat. Mater. (2004) 3 (12), 862], they fabricated CoFe/Fe/MgO/CoFe and CoFe/MgO/CoFe MTJs by room-temperature sputtering and measured room-temperature TMR values as large as 220%. The MTJ stack was highly textured but not single crystalline. This is important because their growth process is compatible with semiconductor processing. Furthermore, they found P = 85%, much larger than expected in a simple Jullière model but consistent with the predictions of theory. A Japanese group independently confirmed this picture with excellent experimental results [Yuasa et al., Nat. Mater. (2004) 3 (12), 868]. Single-crystal Fe(001)/MgO(001)/Fe(001) MTJs were grown epitaxially and room-temperature TMR values as large as 180% were measured. They studied MTJs with a wide range of barrier thicknesses, measuring the barrier height to be about 0.39 eV and demonstrating high TMR for MTJs with a product of resistance and area as small as 300 Ω μm. These are excellent characteristics for device applications.

These extraordinarily high values of TMR translate to high values of device output voltage, and IBM has already made submicron devices. The implications may be very important for next-generation read heads in the magnetic recording industry, and for MRAM, which will benefit in the scaling of device impedance as well as output. More important is the new understanding of the fundamental physical principles of spin transport in ferromagnetic materials and across tunnel barriers. This will have an impact on a variety of spin-injection devices.

[1] Mark Johnson is a research physicist at the US Naval Research Laboratory, Washington, DC.

Read full text on ScienceDirect

DOI: 10.1016/S1369-7021(05)71103-2