In this brief review, we explain the theoretical basis for the notion that spin-transfer torques (STTs) and giant-magnetoresistance effects can, in principle, occur in circuits containing only normal and antiferromagnetic (AFM) materials, and for the notion that antiferromagnets can play a role in STT phenomena in circuits containing both ferromagnetic and AFM elements. We review the experimental literature that provides partial evidence for these AFM spintronic effects but demonstrates that, like exchange-bias effects, they are sensitive to details of interface structure that are not always under experimental control. Finally, we speculate briefly on some strategies that might advance progress.
Among the ordered electronic states that occur in solid-state materials, magnetism is uniquely robust, persisting to well above room temperature in a wide variety of materials. The idea that transport currents could influence the ordered moments of ferromagnetic (F) metals first arose as a theoretical notion [1,2], and was independently discovered in experiments shortly thereafter [3–5]. Thanks to a decade of steady progress in deepening understanding (see the related review volume on this topic ) of this theoretically interesting effect, there is now a genuine prospect that it will dramatically extend the reach of spintronics, a technology that has so far been based mainly on the electrical and magnetic properties of F metals.
The influence of transport currents on F metal magnetic configurations is normally understood in terms of conservation of spin angular momentum. When current flows through an electrical circuit containing non-collinear magnetic elements, upstream and downstream transport electrons are spin-polarized in different directions. If the electronic system is invariant under rotation of electron spin directions, the total spin angular momentum must be conserved. The torque that causes the spin direction of the transport electrons to precess must be accompanied by a reaction torque, which is then identified as acting on the collective magnetization. This notion survives in real systems, even though spin is never perfectly conserved. Because it arises from the approximate conservation of total spin angular momentum, this torque is generally referred to as a spin-transfer torque (STT); i.e. spin angular momentum is transferred from the transport electrons to the collective magnetization. Most current-induced changes in the magnetic configurations of circuits containing F metals can be understood at least qualitatively in these terms.
2. Theoretical concepts for current-induced torques in antiferromagnets
The notion that transport currents could also influence the ordered moments of antiferromagnetic (AFM) metals also arose theoretically , in this case out of an effort [8,9] to develop a more microscopic picture of current-induced torques. The need for a microscopic theory of STTs was motivated by some limitations of the macroscopic spin-transfer picture, the most obvious of which is that (because of spin–orbit coupling and crystal field effects) spin angular momentum is not even approximately conserved in some systems. (The sum of spin and electron orbital angular momentum is not conserved either. Angular momentum conservation is recovered only when the mechanical contribution from nuclear motion is included .)
A microscopic theory of current-induced torques requires both a theory of transport and a theory of exchange interactions within the magnetic material. The picture developed in Núñez & MacDonald  combines a self-consistent field theory of metallic magnetism, imagined as ab initio spin-density-functional theory (SDFT) when chemical realism is desirable, with a transport theory that is able to calculate not only currents but also other local observables like the spin and charge densities. For nanoscale electrical circuits, the non-equilibrium Green function formalism (e.g. ) is convenient when the magnetization is time independent. In SDFT, the (Kohn–Sham) electronic quasi-particles experience a spin-dependent effective potential, which is also time dependent whenever the order is time dependent, and is determined by the instantaneous total electronic charge and spin densities. The role of the transport theory is to predict the Kohn–Sham quasi-particle charge and spin densities in the non-equilibrium situation in which the transport system is connected to source and drain reservoirs with different chemical potentials. In the microscopic theory, current-induced torques simply reflect the property that the relationship between the Hamiltonian and electronic charge and spin densities is altered when the quasi-particles are not in electrical equilibrium. In systems with magnetic order, large changes in global moment orientations, which are normally fixed only by weak magnetostatic, magnetocrystalline or external magnetic field couplings, can be produced by the transport-induced torques, even though they are normally relatively modest in size compared with the torques associated with equilibrium exchange interactions.
For idealized systems in which spin-angular momentum is conserved, the microscopic theory of current-induced torques in F metals with stiff magnetic order is fully consistent with the STT idea. In systems, like F semiconductors, in which transport electrons  are strongly spin–orbit coupled, it predicts important quantitative corrections. The most interesting idea that comes out of considering STTs from a microscopic point of view, however, is that order in antiferromagnets can also be influenced by transport currents. A transport current alters the relationship between quasi-particle Hamiltonians and spin densities, and therefore the magnetization orientations at which they are self-consistent. (Easy magnetization directions can be altered by transport currents even in a uniform bulk ferromagnet  or an antiferromagnet if the crystal symmetry is sufficiently low.) In the microscopic theory, self-consistency signals the possibility of a steady-state magnetization when the quasi-particles are not in electrical equilibrium. At a general bias voltage, there may be no moment orientation configurations for which the densities and the Hamiltonian are consistent. This circumstance signals that a time-independent bias voltage has driven the order into a time-dependent precessing magnetic state.
The influence of a transport current on an AFM nanoparticle is most easily visualized by considering the common case in which it has two sublattices with magnetizations that are equal in magnitude and opposite in direction, and order that is sufficiently stiff to guarantee that local moment orientations on each sublattice are constant across the nanoparticle. In equilibrium, the exchange field direction matches the local moment orientation on each sublattice. When spin–orbit interactions are included in the microscopic theory, self-consistency occurs only when the moments are along the discrete high-symmetry (easy) directions that minimize the total energy. Any transport-induced spin density that is perpendicular to the local moment orientation will produce a small transverse exchange-correlation effective magnetic field and tend to induce local moment precession. For stiff magnetic order, the relative orientation of nearby spins cannot be significantly altered. The current-induced driving term for the total AFM order parameter is therefore the difference between the local torques that act on one magnetic sublattice and those that act on the sublattice with the opposite spin, not the sum.
The order parameter of an antiferromagnet is not the total spin density (which vanishes in the ideal case), but the difference between sublattice spin densities, and is not conserved, even in the absence of spin–orbit coupling. For example, a simple electron-hopping process, which transfers an electron from one sublattice to the other, will alter the difference between the spin densities of the two sublattices. The analogous process in a ferromagnet has no influence on the total spin. The dynamics of the order parameter of an antiferromagnet is therefore not at all related to the conservation of total spin angular momentum, and the way in which it is influenced by transport electrons cannot be described simply in terms of a global reaction torque. Since spin–orbit coupling is typically weak and exchange interactions short-ranged, local torques can however be related approximately to approximate local conservation of spin angular momentum on each atomic site. The net spin current into a particular atomic site tends to produce a torque that acts on the spin centred on that site, ‘attempting’ to restore spin conservation locally. With that motivation, we retain the term STT in discussing the influence of currents on antiferromagnets. Because they are induced by transport currents carried by quasi-particles with energies near the Fermi energy, these torques will inevitably be small compared with the torques that would be produced by near-neighbour exchange interactions if the relative orientations of moments within the nanoparticle were substantially changed. When summed over all atomic sites in the cluster, however, they can be comparable to the weaker anisotropy torques that fix the easy directions of the staggered moment.
Explicit calculations, originally by Núñez et al.  for toy model antiferromagnets and later by Haney et al. , Haney & MacDonald  and Xu et al.  using ab initio methods for chemically realistic but structurally idealized models, have established a number of properties of STTs in circuits containing AFM nanomagnets: (i) torques do in general act on the collective moment orientations of antiferromagnets and can be even larger than the torques that act in ferromagnets, (ii) changes in AFM nanoparticle orientation do alter the resistance of a circuit in which they are embedded, and (iii) the presence of an antiferromagnet in a circuit can contribute to torques that act on F elements of the same circuit. These results have been established by considering specific types of magnetic order and nanoparticle interfaces that have prefect structural order. One surprising discovery in these calculations is that current flowing perpendicular to the F layers of a G-type antiferromagnet can produce a torque that is proportional to the antiferromagnet volume, in contrast to the familiar result from ferromagnets in which the torque is proportional to current and therefore to nanoparticle cross-sectional area. This is one example which demonstrates, albeit for an idealized geometry and for an idealized antiferromagnet, that torques in antiferromagnets can be even stronger than the torques in ferromagnets. Another interesting theoretical observation is that, because of differences between the symmetry properties of ferromagnets and antiferromagnets, the STTs exerted by an antiferromagnet on a ferromagnet vanish not only when their moment orientations are parallel (P) or antiparallel (AP) as in ferromagnets, but also when they are perpendicular . What has been missing in all this theory is a solid qualitative understanding of how the structural disorder present at realistic antiferromagnet/normal or antiferromagnet/ferromagnet interfaces influences resistances and torques.
An important example is the influence of transport currents on exchange bias, which can be viewed as an example of an STT exerted by an antiferromagnet on a ferromagnet. Exchange bias in the absence of a transport current is known to be dependent on the structure of the interface in a complex (see below) manner. In fact, it has been extremely difficult to develop reliably predictive theories for exchange bias because of the interplay between interface disorder and exchange coupling within the F and AFM materials. In §3, we discuss experimental results for the influence of currents on exchange bias. The details of the experimental finding are at present still confusing, although it seems to be clear that there is an effect. It may be that the difficulties that confront attempts to develop predictive theories of STTs in circuits containing AFM elements have much in common with those that stand in the way of a quantitative understanding of exchange bias. One interesting recently proposed strategy [15,16] develops a phenomenology that starts from the assumption that sublattice spins are approximately conserved. Sublattice spin is, as we have mentioned, not conserved, even in the absence of spin–orbit interactions. Still, this approach might be successful when only smoothly varying order parameter configurations play a role.
3. Experimental overview
Because antiferromagnets are already widely used in spintronics as exchange-bias materials, and because much hard-won knowledge has accumulated on the material properties of several AFM metals that are compatible with other materials used in spintronics, the suggestion that antiferromagnets exhibit their own suite of spin-torque and magnetoresistive phenomena is very intriguing and raises a large number of questions. Can we alter the moment direction in AFM nanoparticles at moderate current levels? Could we detect such a change resistively? Can we use transport currents to usefully control exchange-bias effects? Can we invent interesting magnetoresistive circuits that involve antiferromagnets alone? Why have these effects not been obvious experimentally? Progress in answering these questions has been slow, but there is reason to hope that it could accelerate. In this brief review, we discuss the limited progress achieved to date and then offer some speculations on future research directions that could prove to be fruitful.
4. Spintronics in ferromagnetic and antiferromagnetic systems
In the original proposal for AFM metal spintronics , it was suggested that some phenomena observed in conventional spintronics, which uses F components, ought to occur in systems where all F components are replaced by AFM materials. The main focus was on two well-known spintronic phenomena—giant magnetoresistance (GMR) and STT. In F systems, GMR refers to the fact that the system's resistance depends on the relative orientation of magnetic moments in a circuit's F constituents [17,18], while the STT implies that a large electrical current density j can perturb the system's magnetic state [1–3], e.g. by inducing precessions and/or reversals of the magnetic moments in the F systems. The corresponding effects in AFM systems were termed AFM giant magnetoresistance (AGMR) and antiferromagnetic spin-transfer torque (ASTT). AGMR refers to the dependence of the AFM system's resistance on the relative orientation of magnetic order parameters in constituent AFM parts, while ASTT emphasizes a possibility to perturb the order parameter of the AFM by injecting into it a large j.
In what follows, we discuss the feasibility of probing these new AFM effects in experiments and provide a brief review of experimental efforts to date. It is convenient to start with a discussion of how GMR and STT effects are detected in F systems, and then expand the discussion to AFM systems.
5. Experiments in ferromagnetic systems: giant magnetoresistance and spin valves
In F systems, both GMR and STT can be observed in a simple multi-layered structure known as a spin valve . The simplest type of spin valve consists of two F layers separated by a thin non-magnetic (N) spacer. The resistance of such a spin valve changes in a magnetic field. In fact, its exceptional responsiveness to magnetic fields is already exploited in a number of technological applications, including magnetic field sensors, galvanic isolators and non-volatile random access memory devices.
The spin-valve resistance is smallest when the magnetizations of the two F layers are P and largest when the magnetizations are AP. The source of this resistance difference is GMR, which originates from the spin-dependent scattering of electrons traversing the multi-layer film [17,18]. The electron scattering and, therefore, the film resistivity depend on the magnetic configuration of the spin valve, P or AP. The role of the external magnetic field is to change this configuration. The intimate connection between the resistivity and the magnetic configuration can be understood most simply by invoking the usual two-current model for an F metal , in which spin-up and spin-down electrons carry current independently in P, and assuming each F to be a perfect spin filter—if the F's magnetization is ‘up’ (‘down’), it transmits only spin-up (down) electrons while all spin-down (up) electrons are completely reflected. Then, the P state where both Fs are ‘up’ (or ‘down’) would have a non-zero current across both layers carried by spin-up (spin-down) electrons and result in a finite spin-valve resistance RP. By contrast, in the AP state, where two Fs are AP (one ‘up’, one ‘down’), no current can flow across the spin valve since both spin-up and spin-down electron channels are blocked by one of the Fs, thus resulting in an infinite spin-valve resistance RAP. The size of the GMR is defined as (RAP−RP)/RP and is infinite for perfect spin-filtering. In a real spin valve, the filtering, i.e. the spin-selectivity of transmission and reflection, is not perfect and the GMR ratio is therefore finite.
Experimental detection of the GMR is reduced to simply measuring the DC resistance of a spin valve as a function of an externally applied magnetic field. When the magnetizations of the two F layers switch from P to AP, the spin-valve resistance increases. Unless the GMR ratio is very small, the magnetization switching event is easily measured with a simple multimeter. The AP alignment of the two F layers is achieved by making them respond differently to an external magnetic field; an AFM layer in contact with one of the F layers can be used to effectively ‘pin’ the magnetization in this layer through an effect called ‘exchange bias’ .
Note that in collinear bulk antiferromagnets, current is also carried independently by up and down spins. In the AFM case, however, the two spin constituents have equal partial conductivities. This simple spin-valve picture would therefore predict that there is no resistive signal of reversal in antiferromagnets. The effects that occur in theoretical model calculations are not bulk effects, but are more subtle interface effects. Evidently, control of these effects in antiferromagnets will require control over the interfaces between AFM nanoparticles and other elements in the circuits.
6. Experiments in ferromagnetic systems: spin-transfer torque
The STT effect refers to a novel method to manipulate magnetic moments on nanometre length and picosecond time scales using an electrical current [1–3]. The same F/N/F spin valve discussed above can be used for experimental observation of the STT. Here, the current transfers vector spin between the two F layers and may induce precession and/or reversal of the F-layer magnetizations. Since the spin-valve resistance depends on the relative angle θ between the magnetizations owing to GMR (a phenomenological description  gives , the GMR can be used to monitor the relative orientation of the two Fs.
In a typical experiment, we fix the orientation of one F by pinning its orientation with an adjacent AFM layer via the phenomenon of exchange bias. This pinned Fp will act as a polarizer for conduction electrons flowing across such an exchange-bias spin valve (EBSV). If initially the magnetization of the other free layer Ff is almost P with Fp and electrical current flows from Ff to Fp (forward bias), the electrons going from Fp to Ff will stabilize this P alignment and no STT excitation is present. When current is reversed and flows from Fp to Ff, the electrons reflected off Fp (polarized opposite to Fp) will exert a torque on Ff, so rotate it away from Fp, thus resulting in an STT excitation of the system. If, on the other hand, the initial orientation of Ff is almost AP with Fp, a forward bias current will try to change its orientation while reverse bias will stabilize the AP alignment of the two Fs.
Thanks to GMR, the experimental detection of STT is reduced to simply measuring DC resistance of a spin valve as a function of the applied current. At the high current densities needed to produce sufficiently strong STTs, the relative orientation of the F magnetizations changes and so does the spin-valve resistance. The current-induced changes in the latter can be easily detected using standard DC resistance measurements.
7. Possible experiments in antiferromagnetic systems: antiferromagnetic spin valves, antiferromagnetic giant magnetoresistance and antiferromagnetic spin-transfer torque
Now let us turn to the new AFM effects, AGMR and ASTT, and discuss how they can be probed in experiments.
According to the original predictions , the resistance of an AFM spin valve, in which two G-type AFM layers are separated by an N spacer, should depend on the relative orientation of magnetic order parameters in the two AFM layers. The simplest approach to probe AGMR is then similar to what is used to detect GMR in F/N/F spin valves: the resistance of an AFM/N/AFM structure should be measured for different relative orientations of the two AFMs. However, unlike F/N/F systems, in which a small magnetic field can be used to control the relative orientation of Fs, AFM layers respond weakly to the field. One has to find other means to control the orientation of AFMs. One possible solution could be to sandwich the AFM/N/AFM spin valve between two F layers and use the phenomenon of exchange bias  at F/AFM interfaces to control the orientation of AFMs. For a sufficiently thin AFM layer, the field-mediated magnetic reversal of an adjacent F should result in a reversal of the AFM owing to the exchange coupling of magnetic moments across the F/AFM interface. One can thus use a magnetic field to control AFMs by controlling the adjacent Fs. In §8, we discuss the first attempt to measure such AGMR in AFM/N/AFM spin valves.
Another prediction of the original theory  is that a spin-polarized current injected into an AFM can alter its micromagnetic state, e.g. by reversing the AFM order parameter. In metallic systems, this injection can be realized in F/N/AFM or F/AFM systems where F is used to polarize the current. The polarized current is then injected into an AFM across an F/AFM interface or through a N spacer, which should be thin enough for the polarized current to get through and alter the magnetic state of the AFM. In an F/N/F system, such a scheme allows for STT detection because when current alters the magnetic state of one of the Fs, the F/N/F resistance changes owing to the GMR effect and can be easily detected. A similar detection scheme may be used for ASTT detection in F/N/AFM/N/AFM or F/AFM/N/AFM structures. Here, variations in the resistance of AFM/F/AFM owing to AGMR would monitor the magnetic state of an AMF subject to polarized current.
An alternative method for ASTT detection exploits the exchange-bias phenomenon at an AFM/F interface, as was originally proposed  for studying effects of high currents on AFMs. Exchange bias refers to unidirectional exchange anisotropy at an interface between an F and an AFM [21–23]. It originates from the exchange coupling of the moments in F to uncompensated pinned moments in the AFM and can be produced either by cooling the sample to below the blocking temperature of the AFM in the presence of a magnetic field, or by applying the field during sample growth. The direction of the applied field usually (but not always) defines the direction of the bias and results in a shift of the F's magnetization loop opposite to the bias by an amount known as the exchange-bias field. The bias strength depends on the details of the AFM micromagnetic state and can be affected by many factors including surface roughness, domain structure, etc. . However, since the magnitude of the exchange-bias field is known to be associated with the interfacial an AFM moments , one can use it to monitor the state of these moments. For instance, if an electric current, injected from an F into an AFM, changes the exchange-bias field, it can be taken as indirect evidence of the effects of the current on the AFM.
Before we turn to actual experiments, we emphasize again that the prediction of new AFM effects was based in part  on explicit calculations for idealized metallic AFM systems. First, (i) the calculations considered only ballistic transport across both AFM and N layers, and assumed perfect epitaxial interfaces. The AGMR and ASTT are consequences of quantum interference effects in clean AFM heterostructures. Disorder that produces diffusive scattering, both in bulk and at interfaces, is expected to weaken the basis for AFM spintronic effects. Second, (ii) the magnetic order of AFMs used in calculations is of a very special type, often oversimplified compared with that of real metallic AFM materials. Núñez et al. , for instance, considered a simple one-dimensional AFM with magnetic order that is staggered along the direction of current flow.
Theory based on these idealizations does not provide much guidance on searches for AFM effects that are strong enough to be useful. The systems that Núñez et al.  had in mind are AFM transition metals similar to Cr and its alloys, or the rock salt structure intermetallics used as exchange-bias materials . Indeed, Cr would be an excellent choice for the AFM because the magnetic structure of Cr with its spin-density-wave antiferromagnetism would be the closest analogue of one-dimensional AFMs used in Núñez et al. , while epitaxial growth could provide Cr heterostructures with long-enough electron mean free paths. However, neither Cr nor Cr alloys have been tested to date, partially because of the substantial challenges that would have to be overcome to grow clean AFM heterostructures with Cr. As for the metallic AFMs used for exchange bias, virtually all of them are Mn alloys, e.g. Fe–Mn, Ir–Mn, Ni–Mn, Pt–Mn, etc. . These materials have relatively high AFM-blocking temperatures and can be readily grown in thin-film form, e.g. by sputtering deposition, convenient for the AFM heterostructures of interest. This type of AFM has provided most of the initial experimental results described in §8, some of which are quite promising. Nevertheless, idealizations (i) and (ii) are not at all satisfied by these materials. Their magnetic structure is usually complex, with a multi-spin sublattice based on magnetic moments pointing in different directions in different parts of a large unit cell. Their crystal structure usually deviates from cubic symmetry, a property that makes it difficult to grow suitable epitaxial films. The sputtered films used in experiments reviewed in §8 are likely to have significant disorder that weakens quantum interference effects.
8. Initial experiments: antiferromagnetic giant magnetoresistance—current-perpendicular-to-plane and current-in-plane experiments
To our knowledge, the only two experimental searches for AGMR have been performed to date by Wei et al.  and Wang et al. . Both studies were performed in multi-layer thin-film systems containing different combinations of AFM, N and F layers.
Wei et al.  measured both current-perpendicular-to-plane (CPP) and current-in-plane (CIP) magnetoresistances (MRs) of the following thin-film structures: AFM/N/AFM, F/AFM/N/AFM, F/AFM/N/AFM/F, AFM/F/N/ AFM, F/AFM and single F and AFM layers. All films were sputtered onto Si substrates and used N=10 nm of Cu, AFM=3 or 8 nm of FeMn and F=2, 3, 4, 6 or 10 nm of CoFe. After deposition, the samples were cooled from approximately 463 K through the Néel temperature of FeMn in a magnetic field of approximately 18 mT to induce the exchange bias at CoFe/FeMn interfaces. A mechanical point contact (approx. 10–100 nm in diameter) applied to the surface of a film was used to inject an electrical current approximately perpendicular to the film layers. The technique enables measurements of film CPP-MRs at applied current densities j up to 1013 A m−2. CIP-MR measurements (at low currents) were also made on the same samples using a standard four-probe technique.
For small currents (j<1011 A m−2), neither standard CIP-MR measurements, nor CPP-MR measurements with point contacts, showed MRs in any type of samples. This result indicates that no AGMR is present in these samples. For larger applied currents (j>1011 A m−2), a small (approx. 0.1%) positive CPP-MR (resistance is highest at saturation) was seen in some samples with at least one F layer. Such a CPP-MR observed in F/AFM/N/AFM/F samples might be tentatively attributed to AGMR and effects of the high j upon magnetic order in the two AFM layers . However, the presence of similar CPP-MRs in samples with only a single F layer (no AFMs) suggests that it is more probably associated with Fs in these samples .
In another experiment, Wang et al.  used a standard DC four-probe technique to measure the CIP-MR of AFM/N/AFM, AFM/N/AFM/F, F/AFM/N/AFM/F, AFM/F and AFM/N/F multi-layers with N=2.8 nm of Cu, AFM=0.5, 1, 2, 4, 8 or 10 nm of IrMn and F=4 or 8 nm of CoFe. The multi-layers were sputtered in the presence of an approximately 170 Oe magnetic field to induce the exchange bias at CoFe/IrMn interfaces. Only thicker IrMn layers (4 or 8 nm) have been found to actually induce the bias. CoFe layers adjacent to a thinner IrMn (less than 4 nm) showed no exchange bias in standard magnetization measurements.
No MR was observed in samples containing only AFMs (i.e. AFM/N/AFM). A small (0.06–0.5%) negative MR (resistance is lowest at saturation) was found in some samples with at least one F layer. Interestingly, MR was observed only in those samples where one F was unbiased either because it was adjacent to a thin AFM (e.g. in an AFM/N/AFM/F or an F/AFM/N/AFM/F structure) or to an N (e.g. in an AFM/N/F structure) layer. If one assumes that only a thin enough AFM can be switched together with an adjacent F, the MR observed in AFM/N/AFM/F and F/AFM/N/AFM/F multi-layers may be attributed to the GMR between uncompensated magnetic moments at AFM surfaces directly across the N spacer. However, the anisotropic MR (AMR) of the unbiased F layer may provide an alternative interpretation of the data. In this experiment , the current is driven perpendicular to the magnetic field applied along the bias direction. In high magnetic fields, the magnetic moments in an unbiased F layer should be P to the field and perpendicular to the current. When the unbiased F switches at small fields, some of these moments may point along the current direction and result in a resistance increase owing to the AMR effect.
With the above experiments at hands, we can only conclude that more convincing evidence of AGMR is needed. In addition, the absence of AGMR also casts doubt on the detection of any effects of current on the magnetic order of an AFM.
9. Initial experiments: antiferromagnetic spin-transfer torque—current-perpendicular-to-plane experiments
In the absence of AGMR, other means will have to be used in experiments to detect effects of high currents on the magnetic order in AFMs. The exchange bias at an F/AFM interface [21–23] is known to be associated with interfacial AFM moments and can be used to probe their behaviour. By driving an electrical current from F into AFM and by measuring the reversal (bias) field of the F, it should be possible to detect the effects of current on the interfacial AFM moments. A measurement of CPP-MR in EBSV of the form AFM/F/N/F can be used for such detection. Here, the CPP geometry provides the currents across the F/AFM interface, while the GMR in the F/N/F trilayer turns on an MR signal of reversal in the pinned F layer. In what follows, we describe experiments with point contacts  and lithographically patterned pillars  performed in the CPP geometry.
The first experimental report of an ASTT was given by Wei et al.  who used mechanical point contacts to measure the CPP-MR of a sputtered EBSV of the form AFM/F/N/F with AFM=3 or 8 nm of FeMn, F=3 or 8 nm of CoFe and N=10 nm of Cu. The sample geometry is shown in figure 1a. A mechanical point contact (approx. 10–100 nm in diameter) was applied to the surface of the EBSV film and used to inject an electrical current approximately perpendicular to the film layers. To protect the top layer from atmospheric contamination, it was covered by a thin (5 nm) layer of Au. All EBSV samples had a thick buffer layer (50 or 100 nm) of Cu (grey in figure 1a) which was used to secure an approximately CPP current flow across the EBSV sample. This geometry enables measurements of CPP-MR at applied current densities j up to 1013 A m−2.
Figure 1b  shows the MR of a point contact with resistance R=0.92 Ω recorded at different bias currents I ranging from −35 to 35 mA. For this contact, |I|=30 mA corresponds to j approximately 2×1012 A m−2, with positive currents flowing from the contact tip into the sample (figure 1a). The dark curves show the down MR sweeps where the field is ramped from positive to negative values; the lighter curves show the up MR sweeps where the field is ramped back (from negative to positive). For a given bias current I (given trace in figure 1b), the form of MR is typical for spin valves: sweeping down, R is constant at a minimum value at positive fields where magnetizations of the two F layers are P (P state), rises to a maximum when the magnetization of the free (top) F layer switches at a small (approx. −50 Oe) negative field, thus leading to AP alignment of the two Fs (AP state), and then decreases to its minimum value beyond the exchange-bias field at which the magnetization of the pinned F is finally reversed (P state). The up sweeps (lighter curves) show similar transitions in reverse order. It should be noted that the reversals of both free and pinned Fs and the corresponding variations in R proceed via a discrete series of irreversible steps that correspond to reversals of individual F domains in the F layers probed by the point contact; the domains closest to the contact contribute most to its resistance.
The reversal of the free layer seems to be little affected by the applied current. By contrast, the current clearly changes the exchange-bias field at which the pinned layer is reversed and also broadens its reversal transition. The exchange bias increases for −I and decreases for +I. Opposite changes for +I and −I indicate that these shifts cannot be due to Joule heating in the point contact, which should be independent of the current direction; Joule heating might however contribute to the broadening of the switching transitions. The current-mediated shift of the exchange bias is reversible when the current is stepped up or down, and can be quantified by assuming a linear variation of the exchange-bias field with the applied current. The white dashed, white solid and black dashed lines in figure 1c show the least-squares linear fits to the MR data points at the 30, 50 and 70 per cent levels, respectively, of the maximum change in R. Figure 1c also shows the down-sweep MR data of figure 1b in a two-dimensional grey-scale representation, with lighter colours indicating higher resistance.
Although MR curves (figure 1b) vary from run to run at a given current owing to the stochastic nature of switching transitions, the trend indicated by the linear fits in figure 1c is always present and suggests that, on average, the exchange bias increases with applied negative current and decreases with a positive one. The resulting slopes of 0.23, 0.26 and 0.2 T A−1 for three different resistance levels in figure 1c indicate that the current-mediated effect is significant and the exchange bias can be changed by as much as ±20 per cent of its value at zero current.
Different scenarios have been considered to explain the observed effect. The asymmetry with respect to current is usually a signature of STT effects [1–3] at high current densities. However, the conventional (ferromagnetic) STT between the free and pinned F layers is inconsistent with the observed polarity of the effect where positive currents promote the P configuration of the two Fs. By contrast, ASTT can provide a natural explanation of these observations. In support of the ASTT scenario,  found similar effects of current on exchange bias in EBSVs with AFM=IrMn. The data for IrMn are generally similar to those for FeMn (figure 1), with the current having clear and similar effects on the exchange bias, but little or no effect on the coercive field of the free F=CoFe layer. The latter was tentatively attributed to a relatively high coercivity and magnetocrystalline anisotropy of CoFe, which can affect the reversal process. The experiments with a softer F=Py  partially supported this hypothesis by revealing effects of current on the reversals of both free and pinned Py layers. However, the behaviours observed with Py were more complex than those for CoFe, which might be associated with intermixing of Mn (from IrMn or FeMn) and Ni (from Py=NiFe) across the AFM/F interface, thus adding still another AFM (i.e. NiMn) and further complicating the problem.
Another experiment in CPP geometry was performed by Urazhdin & Anthony  with lithographically patterned EBSVs. A sputtered Py(30 nm)/ Cu(10 nm)/Py(5 nm)/FeMn(1.5 nm) spin valve was partially ion milled to form a 120×60 nm2 EBSV nanopillar with the free Py(30) layer left unpatterened. This geometry made it possible to drive current densities as high as j approximately 3×1012 A m−2 through the patterned Py(5)/FeMn bilayer with a positive current flowing from the extended free Py(30) to the pinned F=Py(5). The AFM=FeMn layer was too thin to induce any exchange bias of adjacent Fs at room temperature, but cooling the sample down to 4.5 K in the presence of a 3 kOe magnetic field biased the F by approximately 0.2 kOe immediately after the cool down. However, repeated MR measurements yielded fluctuating values of the bias in the range from −0.16 to +0.13 Oe and were attributed to AFM reorientation between several metastable states.
These authors looked for effects of current pulses on the exchange bias at 4.5 K and in magnetic fields applied along the nanopillar easy axis. First, a pulse of current I0 was applied at a large enough field H0=±3 kOe to suppress current-induced reversal of the Py(5) layer. Then, the exchange-bias field was determined from a single MR measurement at a small I to minimize its effect on the magnetic state of the nanopillar. For positive H0=+3 kOe, they found no effect of current pulses on exchange bias, remaining at approximately −0.08 kOe level for all I0. For negative H0=−3 kOe, the bias field shows initial increases with the magnitude of I0 (from −0.08 to +0.12 kOe for −I0 and from −0.08 to +0.06 kOe for +I0), then slight drops and finally approximate saturation (at +0.1 kOe for −I0 and +0.02 kOe for +I0). Unlike the reversible (versus current) changes in the bias strength from the point-contact experiments [24,26,28], these changes are irreversible, but presumably cannot be sustained in repeated MR measurements where the bias fluctuations are larger than these current-induced variations. The authors attributed these non-monotonic variations in exchange bias to STT effects of current pulses on the Fe moments at the Py/FeMn interface. In support of this statement, they found the variations to correlate with the current-induced magnetodynamics of the two Py layers observed in differential-resistance measurements of the same nanopillar. However, like the exchange bias, this magnetodynamics is related to AFM magnetic moments in a complex and subtle way that makes it difficult to make any definitive conclusions about the AFM magnetic order.
10. Initial experiments: antiferromagnetic spin-transfer torque—current-in-plane experiments
The original ASTT prediction  examined only CPP electron transport across an F/AFM interface. Recently, there were several reports of the effects of current on exchange bias in CIP experiments where current is directed in the plane of the layers of F/N/F/AFM EBSVs. In CIP geometry, there is no net current across the F/AFM interface. However, because of scattering processes, conduction electron flow is not confined to specific layers of the EBSV structure, but can move from layer to layer, thereby transferring spin angular momentum between the layers.
In a series of papers, Tang et al. [30–35] reported on the effects of high CIP currents on the exchange bias in room-temperature experiments. They used a standard four-probe method to measure CIP-MR of sputtered EBSV of the form F/N/F/AFM with free F=10–12 nm of NiFe, N=4 nm of Cu, pinned F=3–12 nm of NiFe and AFM=15 nm of FeMn. All 5×5 mm2 EBSV films had a thin (10 nm) buffer layer of Ta, some had a thin (5 nm) Ta overlayer, and were sputtered in a DC magnetic field of approximately 300 Oe to induce the exchange bias in the sample plane. The exchange bias induced during deposition could be suppressed and/or reversed by the application of a current pulse of appropriate magnitude Ip (up to 600 mA) and duration td (0.1–1 s) in the presence of a magnetic field (Hp=±1.5 kOe). The exchange bias before and after application of the current pulse was checked by measuring CIP-MR at low currents (0.1–1 mA) and by locating the reversal fields of the pinned F layer.
Extensive measurements are summarized by the following observations: (i) the exchange bias induced during deposition can be changed by Ip only if Hp is applied opposite to the original bias direction, if Hp is applied along the original bias, no changes occur at any levels of Ip, (ii) both positive and negative directions of Ip produce similar changes in exchange bias, (iii) the exchange-bias change depends on the magnitude of Ip; for originally positive bias, increasing Ip gradually decreased the bias to zero, reversed its direction and increased the negative bias to an approximate saturation, and (iv) the current density needed for the exchange-bias reversal was estimated to be approximately 1.2×109–1010 A cm−2 and was found to depend on the bias strength (weaker bias requires less current to change it). The authors attributed their observations to STT effects of current on uncompensated moments in the AFM layer. A possible alternative explanation is that the current pulse heated the sample sufficiently (e.g. via Joule heating) to induce a new bias with strength and direction controlled by the direction of Hp and the temperature increase owing to Ip.
The results of Dai et al.  indicate that Joule heating plays a major role in the CIP-MR measurements of their EBSV. They used a standard four-probe method to measure the CIP-MR of sputtered AFM/F/N/F spin valves with AFM=40 nm of NiCoO, pinned F=3 nm of CoFe, N=3.8 nm of Cu and free F=4.5 nm of CoFe. The 1×5 mm2 samples had a 2.4 nm thick CoFe seed layer, and AC (7.5 Hz) currents up to 90 mA were used to measure CIP-MR at room temperature. The nominally insulating AFM layer prevented current from flowing through it. Nonetheless, they also observed a suppression of the exchange bias by sufficiently high current densities (j∼5×109 A m−2). Although the high currents were turned on only for a short time td=0.3 s, the sample resistance clearly increased at high j, indicating a possibility of Joule heating effects. In order to check the effects of heating on the exchange bias, they measured CIP-MR at different temperatures from 300 to 400 K. They found the changes in exchange bias induced by elevated temperatures to be qualitatively similar to those induced by high current densities. However, a detailed analysis of temperature- and current-induced effects on the exchange bias suggests that the heating is not the only mechanism for their high current effect. The authors attributed these extra effects of current to STT between the free and pinned F layers.
It is, by now, clearly established theoretically that STTs and GMR effects can, in principle, occur in circuits containing only normal and AFM materials, and that antiferromagnets can play an active role in STT phenomena in systems circuits containing both F and AFM elements. Because the strength of these effects is not protected by approximate spin conservation, it depends on details of both atomic and magnetic structures. Quantitative theoretical estimates are relatively easy to obtain for a perfect atomic structure and relatively simple magnetic structures, and it is established that the AFM spintronic effects can be as strong as the well-established F effects. On the other hand, it seems clear from the large body of experimental data on spintronic circuits containing AFM elements in which these effects do not play an obvious role, that large values are dependent on having well-ordered interfaces. The experimental literature reviewed here does, however, provide convincing evidence that AFM STT effects are present in current spintronic devices.
Where do we go from here? The most obvious direction towards improved understanding is to grow high-quality epitaxial layered structures, perhaps Cr/Au systems like those discussed theoretically by Haney et al. . This route would however require a large investment in epitaxial growth research. There is a fallback strategy that is promising, even in the absence of the experimental realization of structures that approach the idealized theoretical ones. It appears probable that AFM spintronic effects can be made stronger by achieving greater control over atomic and magnetic microstructures near the interfaces between antiferromagnets and normal metals and between antiferromagnets and ferromagnets. Progress in this direction would benefit from any improved understanding of, and ultimately informed control over, the magnetic microstructure of the AFMs being studied. One challenge here is the need for reliable techniques that provide information on local magnetic order in AFMs. X-ray magnetic circular dichroism spectroscopy  was previously shown to provide valuable information about interfacial order in AFMs. This technique however requires access to specialized major facilities like the Advanced Light Source . Another potential source of experimental information about the magnetic structure of AFMs is AFM magnetic resonance. Since AFM materials usually have much higher resonance frequencies than ferromagnets, this goal may pose separate experimental challenges. In spite of these obstacles, the immense potential benefit of learning how to enhance and control AFM effects in spintronic circuits warrants continued effort.
A.H.M. was supported in part by the Welch Foundation and by an ARO-MURI on many-body physics in bio-assembled nanoparticles. M.T. was supported in part by NSF grant DMR-06-45377.
One contribution of 12 to a Theme Issue ‘New directions in spintronics’.
- This journal is © 2011 The Royal Society