UNCLASSIFIED//FOR OFFICIAL USE ONLY 23 March 201 0 ICO D : 1 D ecem ber 2009 D IA- 08-1 003- 01 1 Defense Intelligence Reference Document A cquisition Threat S upport Metallic Spintronics UNCLASSIFIED//FOR OFFICIAL USE ONLY UNCLASSiFiED//ron orneiAh use e»ih¥ Metallic Spintronics Prepared by: Acquisition Support Division (DW O-3) Defense W arning Office Directorate for Analysis Defense Intelligence Agency A uthor: A A P Person 71 Administrative Note COPYRIGHT WARNING: Further dissemination of the photographs in this publication is not authorized. This product is one in a series of advanced technology reports produced in FY 2009 under the Defense Intelligence Agency, Defense Warning Office's Advanced Aerospace Weapon System Applications (AAWSA) Program. Comments or questions pertaining to this document should be addressed tolAAP Person i | AAWSA Program Manager, Defense Intelligence Agency, ATTN: CLAR/DWO-3, Bldg 6000, Washington, DC 20340-5100. UNCLASSIFIED//rQn OmCIAh UGH OML¥ UNCLASSIFIED//FOR OFFICIAL USE ONLY Contents 1. Introduction...........................................................................................................iv 2. Giant Magnetoresistance........................................................ 1 2.1 GMR Basics.......................................................................................................1 2.2 GMR Applications............................................................................................3 3. Spin-Transfer-Torque.............................................................................................4 3.1 STT Basics..................................................................... 4 3.2 STT Experiments........................................................... 7 3.3 STT Applications............................................................................................10 3.4 STT-Driven Motion of Magnetic Domain W alls............... 12 4. Antiferromagnetic Metal Spintronics...................................................................12 4.1 Antiferromagnetic GMR.................................................................................13 4.2 Antiferromagnetic STT...................................................................................14 5. Summary and Conclusions................................................. 16 6. References............................................................................................................18 Figures Figure 1. In a Magnetic Multilayer, Several Atomic Layers of Magnetic Material Alternate W ith Layers of Nonmagnetic Material.......... 1 Figure 2. Resistance of a Magnetic Multilayer R Versus Magnetic Field...................2 Figure 3. Differential Resistance dV/dl of a Mechanical Point Contact as a Function of Current for a Series of Magnetic Fields........... 4 Figure 4. Device Schematics for STT Experiments....................................................5 Figure 5. Qualitative Picture of STT................................................................ 6 Figure 6. Torques on a Magnetic Moment in a Magnetic Field and Subject to an Electrical Current.............................................................. 7 Figure 7. Spin-Torque-Driven Magnetic Switching....................................... 8 Figure 8. Oscillatory Voltage................................... 9 Figure 9. Scanning Transmission X-ray Microscopy Images.................................10 Figure 10. Conventional MRAM Cell............................................. 11 Figure 11. Racetrack Memory Concept....................................................................12 Figure 12. Schematic of Point Contact to Sample Geometry...................................15 iii UNCLASSIFIED //FOR OFFICIAL UDE OMh¥ UNCLASSIFIED //FOR OFFICIAL USE ONLY Metallic Spintronics 1. Introduction The rapid pace of progress in the computer industry over the past 40 years has been based on the miniaturization of chips and other computer components. Further miniaturization, however, faces serious challenges—for example, increasingly high power dissipation. To continue on pace, the industry must go beyond incremental improvements and embrace radically new technologies. A promising nanoscale technology known as spintronics (a neologism for "spin­ based electronics") has emerged. Spintronics refers to the role an electron spin plays in solid-state physics. Spintronics researchers aim to develop a revolutionary new class of electronic devices based on the spin of electrons in addition to the charge. In spintronic devices, information is carried not by the electron's charge, as in conventional microchips, but by the electron's intrinsic spin. Changing the spin of an electron is faster and requires less power than moving it. Therefore, if a reliable way could be found to control and manipulate spins, spintronic devices could offer higher data processing speeds, lower electricity consumption, and many other advantages over conventional chips, perhaps including the ability to carry out radically new quantum computations. Spintronics in ferromagnetic systems is built on a complementary set of phenomena in which the magnetic configuration of the system influences its transport properties and vice versa. Giant magnetoresistance (GMR) (Reference 1, 2) and spin-transfer-torque (STT) (Reference 3-5) phenomena exemplify such interconnections in multilayers composed of ferromagnetic (F) and nonmagnetic (N) layers. The physics and applications of metallic spintronics are discussed in this report from the perspective of these two phenomena. GMR, research on which was awarded the Nobel Prize in Physics in 2007, refers to a large change in resistance of magnetic multilayers when the relative orientation of magnetic moments in their constituent ferromagnetic layers is altered by an applied magnetic field. The inverse effect, STT, in which a large electrical current density j can perturb the magnetic state of a multilayer, has also been predicted (Reference 3, 4) and observed in experiments on current-induced reversal and precession of magnetization (Reference 5-9) and magnetic domain wall motion (Reference 10, 11)- iv UNCLASSIFIED//FOR OFFICIAL USE ONLY UNCLASSIFIED//6QR OFFICIAL USE OMCT Spintronics is a broad research field with (currently) three major subfields: (1) materials research that is attempting to create new materials that are both magnetic and semiconductors, (2) research on novel magnetotransport effects in ferromagnetic metals, and (3) research on techniques that can be used to manipulate individual electron spins. The first subfield is targeting magnetic semiconductors because devices based on such materials would be the easiest to integrate with the present semiconductor device technology and processing capabilities. However, despite extensive research, most semiconductor spintronic devices are still theoretical concepts awaiting experimental demonstrations. This report focuses on spintronics research in metallic systems within the scope of the second and third subfields. The second subfield has experienced an unprecedented period of new discovery over the past 20 years, including the discovery of GMR, and has already spawned major technological change in the information storage industry with the use of GMR sensors and read heads. The third subfield is vital for spintronic devices, as virtually any processing of information in such devices is associated with transport and manipulation of spins. New and efficient methods for manipulating spins that stimulate active research programs in spintronics at a large number of academic institutions and a half-dozen industrial research labs around the world are highly desirable. The prize to be gained is active control and manipulation of spin distributions (magnetic moments) for new and improved functionality in electronic/spintronic devices. The confluence of intense basic science and industry interest in ferromagnetic metal spintronics has not occurred on this scale in physics in a long time. The report is arranged as follows: Section 2 is dedicated to magnetotransport effects in magnetic systems where magnetic configuration can influence the system's transport properties. It discusses GMR in magnetic multilayers and related phenomena, highlights basic physical principles responsible for GMR, and describes technological applications of the effect. Section 3 focuses on the reverse connection between the system's magnetic configuration and its transport properties—the so-called STT effect. The physical origin and potential applications are discussed. Section 4 discusses other new directions in metallic spintronics, with a particular focus on spintronics with antiferromagnetic materials. Section 5 summarizes, with an eye to the future, the development of spintronic technologies and their aerospace applications. v UNCLASSIFIED//FOR OFFICIAL USE ONh¥ UNCLASSIFIED //FOR OFFICIAL UOE ONLY 2. Giant Magnetoresistance 2.1 GMR BASICS This section discusses the phenom enon of giant m agnetoresistance (GMR) . Excellent reviews of GMR are available elsewhere (Reference 1 2- 22) . The focus on physical concepts im portant for the sections to follow are discussed. GMR in m agnetic m ultilayers refers to a dram atic reduction in the resistance of the m ultilayers when subjected to an external m agnetic field. GMR's size is usually defined as the resistance change in m agnetic field relative to its peak value. The effect can be distinguished from the ordinary m agnetoresistance (MR) com ing from the direct action of the m agnetic field on the electron trajectories via the Lorentz force (Reference 23) , and from the anisotropic MR, which com es from dependence of the resistivity on the relative orientation of m agnetic m om ent to the current (Reference 24) . To prepare the m agnetic m ultilayers, where several atom ic layers of one (ferrom agnetic) m aterial alternate by layers of another (nonm agnetic) m aterial (see Figure 1 ) , a wide variety of deposition m ethods have been used, such as electrochem ical deposition techniques (Reference 25, 26) and various vacuum deposition techniques (Reference 27, 28) . The latter shares m ainly between two m ethods using either sputter deposition or m olecular beam epitaxy (MB E) system s. S putter deposition involves knocking off the atom s of the m aterial of interest from a target by particle bom bardm ent, followed by the deposition of high- energetic atom s (~2- 30 electronvolts [eV]) onto the substrate. A principal advantage of sputter deposition is the ease with which m any different m aterials can be deposited at relatively high deposition rates. In contrast, deposition rates in MB E system s are usually m uch lower than for sputtering system s, but m uch lower energies (~0.1 eV) of the evaporated m aterial Figure 1. In a Magnetic Multilayer, Several Atomic Layers of Magnetic Material (shown in grey) Alternate W ith Layers of Nonmagnetic Material (shown in white). GMR occurs in one of two different geom etries: (1 ) when the current flows in the plane (CIP geom etry) of the layers or (2) when the current flows perpendicular (CPP geom etry) to the layers. m ake this technique favorable for growth of highly oriented single- crystalline film s. The original observation of GMR (Reference 1 ) was m ade on MB E grown iron- chrom ium (Fe/Cr) m ultilayers with nearly perfect crystallinity. S ubsequently, by using sputtered sam ples that are grown m uch m ore rapidly than MB E sam ples, it was possible not only to reproduce these results but also to observe oscillations in the m agnetoresistance as the thickness of the nonm agnetic spacer layers was varied (Reference 29) . S ubsequent 1 UNCLASSIFIED//FOR OFFICIAL USE ONLY UNCLASSIFIED //FOR OFFICIAL USE ONLY studies (Reference 30) on sputtered cobalt- copper (Co/Cu) m ultilayers revealed m agnetoresistances at room tem peratures 3 to 4 tim es larger than those for iron­ chrom ium and 1 3 tim es greater than those for the perm alloy film s that were used as m agnetoresistive sensors in m agnetic reading heads at that tim e. The m uch higher num bers observed in m agnetic m ultilayers predeterm ined the fate of GMR in m agnetic recording technology. The current understanding is that GMR observed in m agnetic m ultilayers arises from the dependence of the resistivity on their internal m agnetic configuration and the role of the external m agnetic field to change this configuration. Figure 2b illustrates GMR in the sim ple lim it where the electron m ean- free- path is m uch longer than the layer thicknesses. The electrical transport properties of the system are described in term s of the so- called two- current m odel (Reference 31 ) , based on the suggestion by Mott (Reference 32) that, at tem peratures lower than the Curie tem perature, the spin- up and spin- down electrons will be alm ost independent and carry current in parallel. Electrons are m uch m ore strongly scattered by a m agnetic layer if they and the local m agnetization spin in opposite rather than the sam e direction (R > r) . For sim plicity, the figure is drawn with scattering only at interfaces; however, there is also scattering within the layers. At zero m agnetic field, where the m agnetizations of adjacent m agnetic layers are aligned antiparallel— for exam ple, because of exchange coupling between the layers (Reference 29)— the spin-down electrons are weakly scattered in layer Figure 2. (a) Resistance of a magnetic multilayer R versus magnetic field, (b) Origin of GMR in terms of spin-dependent electron scattering: Fl and F2 are ferromagnetic layers with a nonmagnetic layer in between. At zero magnetic field, the magnetizations in Fl and F2 are aligned antiparallel (center panel) and can be switched to parallel orientation by an applied field, (c) The equivalent resistance circuits corresponding to the three magnetic configurations shown in (b). See text for details. Fl but strongly scattered in F2. In contrast, the spin- up electrons are weakly scattered in layer F2 but strongly scattered in Fl. As a result, two channels are equivalent, leading to a total resistance in this "antiferrom agnetic" configuration Raf = (R+ r) /2 (see the corresponding resistance circuit in Figure 2c) . W hen the m agnetizations of the two F layers are set into parallel configuration by an applied m agnetic field, the spin- up electrons are weakly scattered in both layers and form a low- resistivity channel, whereas the spin- down electrons are strongly scattered in all the layers and form a high- resistivity channel. The reversal of m agnetic field just interchanges the roles of spin- up and spin- down channels. The current's shunting by the low- resistivity channel produces a low total resistance Rf = 2Rr/(R+ r) in this "ferrom agnetic" configuration. The size of the GMR is defined as (Raf- Rf) /Raf = (R- 2 UNCLASSIFIED//EQB QEETriAI USE AMI Y UNCLASSIFIED/ /FOR OFFICIAL UGE ONLY r) 2/(R+ r) 2 s 1 - The other definition AR/R = (Raf- Rf) /Rf = (R- r) 2/4Rr (unbounded from above) is also in use. Figure 2a shows a m agnetoresistance curve typical for m agnetic m ultilayers. The resistance is constant at a m inim um value Rf above a saturation field B s (parallel Fs) and rises to a m axim um value Raf as the applied m agnetic field B approaches zero (antiparallel Fs) . GMR occurs in two different geom etries (see Figure 1 ) : nam ely when the current flows in the plane of the layers, or CIP geom etry, or when current flows perpendicular to the layers, or CPP geom etry. Most of experim ents on GMR are carried out in the CIP geom etry because m easuring the fairly large resistance of a thin film is quite easy (film length is typically orders of m agnitude larger than its thickness) . Experim ents in the CPP geom etry are m ore difficult (Reference 33) and require special techniques for precision m easurem ents of very sm all resistances ~1 0 7- 1 0 8 Q resulting from the "short and wide" geom etry of a 1 - m m 2 "wide" and 1 - pm "long" sam ple. In order to increase the resistances to easily observable values, m icrofabrication techniques can be used to reduce the sam ple's cross- sectional area (Reference 34- 36) . Finally, a sim ple and inexpensive point- contact technique (Reference 37) m ay also be suitable for this purpose. The sam ples with a reduced cross- sectional area will be of interest for spin­ transfer- torque experim ents presented in S ection 3. 2.2 GMR APPLICATIONS GMR is currently used in m agnetic field sensors, including those in read heads for com puter hard drives, in galvanic isolators, and in nonvolatile random access m em ory devices. Reading inform ation stored on m agnetic hard disk drives in com puters was the first large- scale com m ercial application of GMR. The inform ation is stored by m agnetizing sm all regions (m agnetic dom ains) of a m agnetic recording disk in different directions. The stray m agnetic fields from these dom ains are detected by a GMR sensing elem ent called spin valve. The sim plest type of spin valve consists of two ferrom agnetic layers separated by a thin, nonm agnetic spacer. The spin- valve resistance is sm allest when the m agnetizations of the two ferrom agnetic layers are parallel and largest when the m agnetizations are antiparallel. The antiparallel alignm ent is achieved by m aking the two layers respond differently to an external m agnetic field; an antiferrom agnet in contact with one of the layers is used to effectively "pin" the m agnetization in this layer through an effect called "exchange bias." The exceptional responsiveness of spin valves to m agnetic fields has enabled very high areal packing densities in hard drives. O ther sensor applications using GMR elem ents include m onitoring of a ferrous gear rotation in m achinery operation (Reference 38) via detection of a changing m agnetic flux when a gear tooth passes near the sensor, m onitoring of electrical current via detection of the current- induced O ersted m agnetic field, and transferring high- frequency signals between isolated circuits (Reference 39) via m agnetic fields generated by a high- frequency inductor in one circuit and replicated in another circuit by a GMR sensor. U N C LASSI FI E D/ /EQB^SHGUX-U6MNb¥ 3 UNCLASSIFIED //TOR OFFICIAL USE ONLY 3. Spin-Transfer-Torque This section focuses on the spin- transfer- torque (S TT) phenom enon, which refers to a novel m ethod to control and m anipulate m agnetic m om ents in nanostructures by spin currents— one of the forefront and m ost exciting areas in m agnetism research today. 3.1 STT BASICS The previous section showed that the m agnetic state of a ferrom agnet can affect its electrical transport properties; for instance, the relative orientation of the m agnetic m om ents in m agnetic m ultilayers underlies the phenom enon of GMR (Reference 1 , 2) . The inverse effect, in which a large electrical current density can perturb the m agnetic state of a m ultilayer, has also been predicted (Reference 3, 4) . Here the current transfers vector spin between the m agnetic layers and induces precession and/or reversal of the layer m agnetizations. Altering the m agnetic state with spin currents is based on quantum m echanical exchange interaction and represents a novel m ethod of m agnetization control on the nanom eter length scale and the picosecond tim e scale. The first observation of such a spin­ transfer phenom enon in m agnetic m ultilayers was recorded by Tsoi et al. (Reference 5) (see Figure 3) . In this experim ent, the spin- transfer- induced excitations were produced by injecting high- density electrical currents into a Co/Cu m agnetic m ultilayer through a m echanical point contact. Point contacts sm aller than 1 0 nanom eters (nm ) are form ed when a sharpened Cu m etal wire (tip) is carefully brought into contact with the m ultilayer. The extrem ely sm all cross- sectional area of such a contact m akes it possible to achieve current densities in excess of 1 01 2 A/m 2. B ecause of its extrem ely sm all size (<1 0 nm ) , point contact is a very efficient probe of electrical transport properties in extrem ely sm all sam ple volum es inaccessible with other techniques (for exam ple, electron­ beam - lithography patterning) . The Figure 3. Differential Resistance dV/dl of a Mechanical Point Contact as a Function of Current for a Series of Magnetic Fields. The peak in dV/dl indicates the onset of S S T excitations. The inset shows that the threshold current at the peak in dV/dl increases linearly with the applied field. (Reference 5) latter qualifies point contact as the sm allest probe of S TT today. The S TT phenom enon currently attracts considerable attention because it com bines poorly understood fundam ental science questions with the prom ise of applications in a broad range of technologies. In high- speed, high- density m agnetic recording technology, for instance, S S T could replace the O ersted field currently used for writing m agnetic bits in storage m edia (for exam ple, in m agnetic random access m em ory [MRAM]) . This m ay lead to a sm aller and faster m agnetic m em ory. Another possible application is based on the spin- transfer- induced precession of m agnetization, which 4 UNCLASSIFIED//FOR OFFICIAL USE ONLY UNCLASSIFIED //FOR OFFICIAL USE ONLY converts a direct current (de) voltage input into an alternating current (ac) voltage output. The frequency of such a precession can be tuned from a few gigahertz (GHz) to > 1 00 GHz by changing the applied m agnetic field and/or de current, effectively resulting in a current- controlled oscillator for use in practical m icrowave circuits. S ince its prediction in 1 996, the S TT effect has been observed in a num ber of experim ents, including those with m echanical (Reference 5, 9, 40) and lithographic point contacts (Reference 6, 41 ), m anganite junctions (Reference 7) , electrochem ically grown nanowires (Reference 8) , lithographically defined nanopillars (Reference 42, 43) , tunnel junctions (Reference 44), and sem iconductor structures (Reference 45) . These different m ethods all share one characteristic feature: they m ake it possible to attain extrem ely high current densities (> 1 01 2 A/m 2 for m etallic structures) needed to produce sufficiently large spin- transfer torques (Reference 3, 4) . This is achieved by forcing the electrical current to flow through a very sm all constriction. The latter can be a m echanical point contact, a lithographically defined point contact or nanopillar, or a nanowire, as illustrated in Figure 4. In all cases, the m axim um current density jm ax = I/A is defined by the current I flowing through the device and the m inim um cross­ sectional area A of the current path. For typical m echanical point contacts, I ~1 m A and A ~1 00 nm 2 gives jm ax ~1 01 3 A/m 2. In lithographically defined structures, both I and A are typically larger, I ~1 0 m A and A ~1 0000 nm 2, that gives jm ax ~1 01 2 A/m 2. m echanical point contacts lithographycal point contacts m anganite electrodeposited trilayer junctions nanow ires Figure 4. Device Schematics for STT Experiments. All experim ents share one com m on feature: a sm all constriction for electrical current— that is, point contact, junction, nanowire, or nanopillar. B lack (grey) indicates insulator; dark blue indicates m agnetic m aterial. lithographycal pillar devices The basic physical m echanism underlying S TT relies on conservation of angular m om entum . Consider a pedagogically sim ple case where a conduction electron crosses an interface between a nonm agnetic m etal (N) and a ferrom agnet (F) . W e assum e the initial state of the electron's spin S in N is noncoIlinear to the F's m agnetization M. O nce into F, S is subject to an exchange torque caused by M that tends to reorient S . Therewith, according to Newton's Third Law, there should also exist a reaction torque that acts on M - S TT torque. D eep into F, S is aligned with M, and the change in angular m om entum that occurs from its reorientation has been transferred to M. Hence the phenom enon's nam e: spin- transfer torque. O f course, the torque applied to M by a single- spin S is negligibly sm all owing to S being negligibly sm all com pared with M. For high current density crossing the N/F interface, however, the num ber of such spins can be very large and the resulting effective S m ight becom e com parable to M. This highlights the need for high current densities to observe the S TT phenom enon. A typical experim ent on current- driven excitation of a ferrom agnet usually involves two single- dom ain thin- film m agnets separated by a nonm agnetic spacer. Here one m agnet 5 UNCLASSIFIED//FOn OFFICIAL USE ONLY UNCLASSIFIED //FOR OFFICIAL UOE ONLY (Fa) is "hard" and used to polarize the current, while the spacer (N) is thin enough for the polarized current to get through and excite the second "free" m agnet (Fb) . This Fa/N/Fb trilayer structure is sim ilar to a GMR spin valve. The GMR effect can thus be used to m onitor the orientation of Fb relative to Fa - GMR varies linearly with cos0, where 0 is angle between m agnetic m om ents of Fa and Fb, and a phenom enological description (Reference 46) gives the trilayer resistance R(0) = Rf+ (Raf- Rf) (1 - cos0) /2. W hen current flows across Fa/N/Fb, the current- induced torques act on both Fa and Fb layers (Reference 3, 47) . This is schem atically illustrated in Figure 5. This qualitative picture of S TT assum es both Fa and Fb layers are perfect spin filters, so that electron spins aligned with the m agnetic m om ent of, for exam ple, Fa layer are com pletely transm itted through the layer, while spins aligned antiparallel to the layer m om ent are com pletely reflected. W hen electron current crosses the Fb/N/Fa trilayer from right to left (Figure 5a), electrons transm itted through Fa will be polarized along Fa. If spin- diffusion length in N is long enough, this spin- polarized current will reach Fb and exert a torque on Fb in a direction so as to align Fb with Fa. Repeating the argum ent for Fb, we find that electrons reflected from Fb will be polarized antiparallel to Fb and, hence, in turn exert a torque on Fa trying to align Fa antiparallel with Fb. The net result is a pinwheel- type m otion with both Fa and Fb rotating in the sam e direction (clockwise in Figure 5a) , as described previously by S lonczewski (Reference 3) . W hen the current crosses the trilayer from left to right, the directions of the torques are reversed (Figure 5b) — the torque on Fa is trying to align Fa parallel with Fb, while the torque on Fb is trying to align Fb antiparallel with Fa. Figure 5. Qualitative Picture of STT. (a) For left- going electrons m agnetic m om ents (thin arrows) of both Fa and Ft> are rotated clockwise, (b) For right- going electrons the directions of the torques (thick arrows) on F^ and Ft are reversed. (Reference 47) The above discussion im plies the asym m etry of S TT with respect to current direction as follows. Let's fix the orientation of the polarizer Fa; in experim ents this is usually accom plished by m aking Fa very thick (com pared with Fb) or by pinning its orientation with an adjacent antiferrom agnetic layer via the phenom enon of exchange bias. If initially Fb is alm ost parallel with Fa, the left- going electrons will stabilize this parallel alignm ent, and no S TT excitation is present. W hen current bias is reversed, the torque on Fb will try to rotate Fb away from Fa and will result in S TT excitation of the system . This asym m etry with respect to current polarity is one of the m ain features of S TT in experim ents; see, for instance, Figure 5, where S TT excitations are present only at negative bias. 6 UNCLASSIFIED//FnP nttK IAL USE ONLY UNCLASSIFIED //FOR OFFICIAL UDE ONLY 3.2 STT EXPERIMENTS S ection 3.1 introduced the physics of S TT in m agnetic nanosystem s. W e have seen that a high- density electrical current can result in torques on m agnetic elem ents of the system . These torques m ay be used to control and m anipulate the system 's m agnetic state. However, the resulting behavior of the system can differ significantly from case to case, depending on particular conditions of observation. For instance, in m odest external m agnetic fields, m agnetization of a sm all elem ent can be repeatedly reversed between two stable configurations, while at higher fields, where the reversal is energetically unfavorable, the m om ent can be set into precession at a very high frequency. To understand details of what happens with a m agnetic m om ent 5 in a particular situation, one can use Newton's S econd Law. Fors this would be the Landau- Lifshitz- Gilbert equation, where the rate of change of5 is set equal to the net torque acting on - S’: Equation 1 : ^= /ixg «fx^+,,^px(F-xp) d t |S | 2T) . The onset of S TT- driven m agnetic precession is revealed by a peak in differential resistance of the contact. The peak in dV/dl indicates the transition into precession is a reversible process, and in a sm all range of currents, one can continuously increase or decrease the angle of precession. However, other scenarios are also possible— for exam ple, fast transitions between static and steady- state precession states with current- dependent dwell tim e. If the applied m agnetic field is sm all, the m agnetic system can have m ore than one low- energy state. In the sim ple case of a m agnetic elem ent with uniaxial anisotropy, S TT can trigger a transition between two static states that are energetically favorable. An exam ple of such behavior is shown in Figure 7. Here the current is driven across a trilayer Py20nm /Cul2nm /Py4.5nm spin- valve structure patterned by electron beam lithography into a nanopillar with a 40 x 1 20 nm 2 cross- sectional area (Reference 48) . The differential resistance of the nanopillar exhibits a hysteresis as a function of an applied bias current as the m agnetization of the thin (free) perm alloy (Py) layer is aligned parallel and g82 15 8.1 ^ 8.0 -0.4 -0.2 0.0 0.2 0.4 I (nA) Figure 7. Spin-torque-driven magnetic switching for a Py20nm/Cul2nm/Py4.5nm spin valve with a 40xl20-nmz cross-sectional area, as the magnetization of the thin (free) magnetic layer is aligned parallel and antiparallel to the thicker magnetic layer by an applied current. (Reference 48) antiparallel to the thick (hard) Py layer by the current. The two exam ples presented above (Figures 3 and 7) dem onstrate how sim ple de resistance m easurem ents can be used for S TT observation. Here the m easured resistance of a device provides indirect inform ation about the relative orientation of m agnetic elem ents in the device. However, m easured critical currents highlighted by sharp variations in the resistance rem ain the only experim ental inform ation that can be used to quantitatively com pare theory and experim ent. Moreover, the de m easurem ents in Figure 3 provide no inform ation about fast evolution of m agnetization in the device 8 UNCLASSIFIED//FnP nFFTCTAI USB ONI Y UNCLASSIFIED //FOR OFFICIAL USE OW Lf associated with high- frequency precession of m agnetic m om ents. High- frequency techniques m ust be em ployed to provide such capabilities, as discussed next. The first experim ent providing unequivocal evidence that a de electrical current can result in high- frequency (tens of GHz) precession of m agnetic m om ents was reported by Tsoi et al. in (Reference 9) . Here an S TT device— point contact— was placed in a m icrowave cavity of a high- frequency, high- field electron spin resonance (ES R) spectrom eter. This arrangem ent allowed perform ing de transport experim ents, such as those described above, while the contact was irradiated with high- frequency m icrowaves. W hen the frequency of external m icrowaves m atched the precession frequency excited by the de current, an additional (rectified) voltage was detected across the contact. B y detecting this voltage while varying the external frequency, field, and applied current, Tsoi et al. (Reference 9) were able to m ap the frequency excited by de current as a function of applied field and current. In a m ore recent experim ent, Rippard et al. (Reference 49) fed m icrowaves to a point contact via electrical leads and reported observation of a sim ilar de response. Finally, the high- frequency dynam ics of the free- layer m agnetization can be m easured directly by detecting high- frequency oscillations in voltage across a spin valve under de current. Here the hard m agnetic layer is fixed, while the free layer exhibits a high- frequency precession relative to the hard layer. GMR results in a high- frequency m odulation of the spin- valve resistance, which in turn leads to a high- frequency com ponent of the voltage across the spin valve traversed by a de current. This voltage can be directly probed with a high- frequency spectrum analyzer, as was recently done by Kiselev et al. (Reference 50) and by Rippard et al. (Reference 41 ) . Moreover the voltage oscillations owing to spin- torque- driven m agnetic precession can be directly m easured in tim e dom ain using a sam pling oscilloscope (Reference 51 ) , as illustrated in Figure 8. Time (ns) Figure 8. Oscillatory Voltage Generated by Processional Motion of the Free Magnet in IrMn8nm/NiFe4nm/Cu8nm/NiFe4nm Nanopillar, in Response to a 335-mV de Voltage Step Applied to the Device at B = 630 Oersteds (Reference 51 ) The above exam ples illustrate how broadband instrum entation for m easuring voltage in GMR devices m ay provide im portant and often unique inform ation about high- frequency m agnetic dynam ics driven by spin- transfer torques. However, the detailed understanding of S TT is still the subject of debate and requires new experim ental techniques capable of probing m agnetization dynam ics on nanom eter length scales and sub-nanosecond tim e scales. In principle, this can be accom plished by the use of synchrotron x- rays that were recently shown (Reference 52) to probe interfacial phenom ena and directly im age the tim e- resolved response of m agnetic nanostructures to sub- nanosecond m agnetic field pulses (O ersted switching) and spin- polarized current 9 UNCLASSIFIED//rOR OmCIAL USE ONL¥ UNCLASSIFIED //FOR OFFICIAL UOE ONLY pulses (S TT switching) . Figure 9 shows scanning transm ission x- ray m icroscopy (Reference 53) im ages of in- plane com ponents of m agnetization M — Mx in panel (a) and Mv in panel (b)— in the free CoFe layer (indicated by blue ellipse) of a 1 00xl50- nm 2 m agnetic nanopillar. The im ages were obtained by scanning a focused (diam eter ~ 30 nm ) circularly polarized x- ray beam across the CoFe layer, with the photon energy tuned to the characteristic Co L3 resonance to provide m agnetic contrast through the x- ray m agnetic circular dichroism effect (Reference 54), and by m onitoring transm ission of the x- rays as a function of the position x, y with a fast avalanche detector. The M- vector field of the free layer can be reconstructed from the m easured Mx and My com ponents as illustrated in Figure 9c, and the ultrafast x- ray m icroscopy technique provided a m eans to m onitor this field as a function of tim e with ~1 00- picosond resolution. The spatial resolution of the technique is set by the spot size of the x- ray beam (~30 nm ) and currently lim its its application to spintronic devices > 1 00 nm in size (Reference 52) . Potentially, however, technical developm ent of the ultrafast x- ray m icroscopy m ay lead to an ultim ate technique for S TT studies that can probe the M- vector field on the nanom eter length scale with picosecond tim e resolution. (a) (b) (c) Figure 9. Scanning transmission x-ray microscopy images of Mx (a) and My (b) components of magnetization M combine into the vector field (c), which represents the direction of M in the plane of the CoFe free layer. (Reference 52) 3.3 STT APPLICATIONS The S TT m ethod to m anipulate m agnetic m om ents by an electrical current offers unprecedented spatial and tem poral control of spin distributions and attracts considerable attention because of its potential application in a broad range of technologies. The perspective of S TT for GHz com m unication applications and in m agnetic recording technology is discussed. The S TT application in high- frequency technologies is based on the spin- transfer- induced precession of spins. The previous section discussed how precession of m agnetization in GMR devices can convert a de current input into an ac voltage output. The frequency of this output can be tuned from a few GHz to > 1 00 GHz by changing the applied m agnetic field and/or the de current, effectively resulting in a current- controlled oscillator for use in practical m icrowave circuits. Hence, the S TT effect in 1 0 UNCLASSI FI ED//*6n OFFICIAE USE ONLY UNCLASSIFIED //EQB QEEIW Ak USE ONLY GMR structures provides a m eans to engineer a nanoscale high- frequency oscillator powered and tuned by de current. S uch an oscillator could have frequency characteristics spanning m ore than 1 00 GHz and perhaps into terahertz range. Linewidths as narrow as 2 m egahertz were dem onstrated (Reference 55) , leading to quality factors over 1 8,000. Potential applications for such high- frequency sources include integrated transceivers for wireless and wired applications, as well as wireless and wired chip- to- chip and on- chip com m unications. For the latter, logic circuits with a spin wave bus were proposed (Reference 56, 57) as an interface between electronic circuits and integrated spintronics circuits. Here spin waves are used for inform ation transm ission and processing, and the S TT effect can provide a m eans for efficient spin­ wave generation on the nanoscale. In high- speed, high- density m agnetic recording technology, S TT could replace the O ersted field currently used for writing m agnetic bits in storage m edia (for exam ple, m agnetic random access m em ory [MRAM]) , thus leading to sm aller and faster m agnetic m em ory. Figure 1 0a schem atically shows a bit cell of a conventional MRAM. The bit state is program m ed to a "1 " or "0" by switching between the parallel and antiparallel states of a GMR- like storage elem ent. The first- generation MRAM utilizes m agnetic tunnel junctions (MTJ) as storage elem ents because of their higher m agnetoresistance ratios and im pedance- m atching constraints. However in scaling MRAM to sm all dim ensions, the sam e constraints are expected to drive a transition from MTJs to fully m etallic spin- valve storage elem ents. The switching between "1 " and "0" states (writing) relies on m agnetic reversal of the MTJ's free layer, achieved by passing electrical currents down the "bit" and "write" lines, that generates a sufficiently strong m agnetic field at their intersection (that is, for a given MTJ) . However, as the spatial decay of this O ersted field is rather slow (~l/r2) , it m ay affect neighboring cells. This m akes scaling of MRAM to sm all dim ensions questionable. Figure 10. Conventional MRAM Cell (a) Versus STT MRAM Cell (b) (Reference 58) S TT MRAM (see Figure 1 0b) rem oves the constraints on scalability. Here the switching of the free layer m agnetic m om ent is achieved by S TT switching when a high- density electrical current is driven directly through the storage elem ent (MTJ or spin valve) . This writing is thus perform ed at high current levels, while the reading (m easuring the resistance of the elem ent) is done at low currents. S ince S TT MRAM elim inates the need for "write" and "bypass" lines, a m ore com pact m em ory can be realized. 1 1 UNCLASSIFIED/ /EQR QEEICIAh USE OML4 UNCLASSIFIED //FOR OFFICIAL UCE ONLY 3.4 STT-DRIVEN MOTION OF MAGNETIC DOMAIN W ALLS Yet another m anifestation of S TT in m etallic ferrom agnets is a m otion of m agnetic dom ain walls traversed by an electrical current. The original prediction of the effect dates back to 1 978, when Luc B erger predicted that a spin- polarized current should apply a torque to a m agnetic dom ain wall (Reference 59) . In a series of rem arkable but only recently appreciated works, B erger set the theoretical (Reference 59- 62) and experim ental (Reference 63- 65) groundwork for current- induced dom ain wall m otion (CID W M) , which is now docum ented in m aterials ranging from m agnetic sem iconductors (Reference 66) to perpendicular- anisotropy superlattices (Reference 67) . B ut the m ost widely studied m aterials by far have been m etallic ferrom agnets (Reference 1 0, 1 1 , 68- 82) , including Py (NisiFeis) , CoFe, and Co, because of their decades- long ubiquity in m agnetic storage technology. The CID W M effect can be qualitatively understood on the basis of the following argum ents. Consider an electrical current flowing between two m agnetic dom ains (A and B ) with opposite m agnetizations and, thus, traversing a 1 80- degree m agnetic dom ain wall. The situation is sim ilar to that of a single N/F interface discussed in S ection 3.1 . W hile in dom ain A, spins of conduction electrons are preferentially aligned with the m agnetic m om ent of A. O nce into dom ain B , the spins reverse to align with the m om ent of B . In reversing the electron spins, m agnetic m om ents in the dom ain wall experience a torque associated with the change in angular m om entum that occurs from the rotation of electrons spins. This spin- transfer torque can m ove the dom ain wall in the direction of the electron flow. Moving m agnetic dom ain walls with current was proposed as the basis for a new type of m agnetic m em ory called "racetrack” (Reference 83). In contrast to today's hard disk drives (HD D ) , which rely on spinning m otion of a disk to m ove their m agnetic regions where the data is stored past a read head, the racetrack m em ory exploits the idea of m oving m agnetically stored data electronically. Figure 1 1 illustrates the concept of the racetrack. The racetrack is a ferrom agnetic nanowire, with data encoded as a pattern of m agnetic dom ains along the wire. Current pulses can m ove the entire pattern along the wire. The two cartoons of Figure Ila Figure 11. Racetrack Memory Concept. (Reference 83) show the dom ain wall patterns in the racetrack before and after they have m oved past read and write elem ents. Reading is achieved by m easuring the resistance of a tunnel junction elem ent connected to the racetrack (Figure 1 1 b) ; writing (Figure 1 1 c), by applying local m agnetic fields— for exam ple, the fringing fields of a dom ain wall m oved in another nanowire. If U - shaped nanowires are placed norm al to the plane of a chip and arranged into high- density arrays of racetracks (Figure lid) , the resulting storage density can be higher than that in solid- state m em ory devices like flash RAM and 1 2 UNCLASSIFIED//FOR OFFICIAL UCE OHh¥ UNCLASSIFIED //EQR QEEtClAE USE ONLY sim ilar to conventional HD D s; but the racetrack m em ory would have m uch higher read/write perform ance than HD D . 4. Antiferromagnetic Metal Spintronics Recently, MacD onald and coworkers (Reference 84) predicted that effects sim ilar to GMR and S TT observed with ferrom agnets ought to occur in m ultilayer system s where the ferrom agnetic (F) com ponents are replaced by antiferrom agnets (AFM) . First, they predicted that the resistance of an AFM spin valve— where two AFM layers are separated by an N spacer— could depend on the relative orientations of the m agnetic m om ents in the two AFM layers (antiferrom agnetic GMR = AGMR) . S econd, they predicted that injecting a sufficiently strong current density into an AFM should affect its m agnetic state via current- induced spin torque. These new AFM effects could lead to new all- AFM spintronics where AFMs are used in place of Fs. Replacing F m etals with AFM m etals in a spintronic device has several advantages. W hile AGMR of an AFM spin valve was predicted (Reference 84) to be sim ilar in m agnitude to GMR in standard F spin valves, the critical current needed to alter the m agnetic order in AFMs can be sm aller than for Fs, partly because spin torques can act through the entire AFM volum e. The estim ate of the necessary current density j ~ 1 09 A/m 2 (Reference 84) was less than the typical j ~ 1 01 1 A/m 2 needed to reverse the m agnetic order in F/N/F m ultilayers (Reference 5- 9, 40- 43, 48- 51 ) . Finally, using AFM m etals in spintronic devices in place of F m etals would elim inate unwanted effects of shape anisotropy on the m agnetic stability of sm all elem ents, thus potentially offering better control of the m agnetic state in nanoscale system s and easing fabrication requirem ents. Following the original predictions of MacD onald and coworkers (Reference 84) , Xu et al. (Reference 85) calculated the AGMR for a sim ple AFM/N/AFM/N = FeMn/Cu/FeMn/Cu m ultilayer, and Gom onay and Loktev (Reference 86) provided additional theoretical evidence that polarized current can destabilize the equilibrium state of an AFM. Note, however, that all calculations to date are for perfect sam ples and depend on quantum coherence. It is known that disorder can reduce the predicted effects. Experim ents are thus crucial to see if any such effects are visible in real sam ples. 4.1 ANTIFERROMAGNETIC GMR To the author's knowledge, the only experim ental study searching for AGMR was perform ed by W ei et al. (Reference 87) . They have m easured current- in- plane (CIP) and current- perpendicular- to- plane (CPP) m agnetoresistances (MR) of m agnetic m ultilayers containing two antiferrom agnetic layers separated by a nonm agnetic layer. S uch an antiferrom agnetic spin valve, AFM/N/AFM, was predicted (Reference 84) to exhibit AGMR sim ilar to GMR seen in ferrom agnetic spin valves, F/N/F, containing two ferrom agnetic layers separated by a nonm agnetic layer. Note, however, that calculations (Reference 84) assum ed ballistic transport in sam ples with perfect layers and interfaces where AGMR is a consequence of quantum interference effects. Thus, any disorder that produces diffusive scattering and weakens quantum interference will weaken any such AGMR. 1 3 UNCLASSIFIED//EQR QFHCiAh USE ONLY UNCLASSIFIED //FOR OFFICIAL UOE ONLY In F/N/F trilayers, the relative orientation of the m agnetizations of the two Fs is controlled by an externally applied m agnetic field B . To achieve well- defined antiparallel and parallel states of the two F layers, the m om ent of one F is often "pinned," via exchange coupling (exchange bias) to an adjacent AFM layer (Reference 88, 89) , leaving the m om ent of the other free to reverse in m uch sm aller B . In a sim ple AFM/N/AFM sam ple, just applying a field B is not expected to be efficient because of the weak effect of external fields on m agnetic m om ents in AFMs. To achieve better control of the AFMs, AFM/N/AFM spin valves can be sandwiched between two F layers to give F/AFM/N/AFM/F, with the two AFM layers differently exchange- coupled to their respective F neighbors. Applying a m agnetic field to change the m agnetic order of the F layers should then also affect the order of the AFM layers. In addition to AFM/N/AFM and F/AFM/N/AFM/F m ultilayers, W ei et al. (Reference 87) have studied a variety of structures— F/AFM/N/AFM, AFM/F/N/AFM, F/AFM, and single F and AFM layers— to isolate the MR observations of interest from potential spurious effects. For sm all applied currents, neither standard current- in-plane (CIP) MR m easurem ents on extended m ultilayer film s nor CPP MR m easurem ents with point contacts showed any MRs for sam ples of all types. For larger applied currents, sm all positive CPP MRs (resistance is highest at saturation) were som etim es observed in sam ples with at least one F layer, while no MR was seen in sam ples with no Fs. These observations suggest sputtered AFM/N/AFM m ultilayers do not show AGMR, possibly owing to m ostly diffusive transport in such im perfect film s. S m all MRs observed at higher currents in film s with F layers m ay be associated with the suppression at high currents of spin accum ulation induced within and around Fs (Reference 90) . Further studies on high- quality film s are still needed to verify any possible existence of AGMR in such structures. 4.2 ANTIFERROMAGNETIC STT S tim ulated by the theoretical studies in (Reference 84-86), four experim ental searches for effects of S TT on AFMs have been published so far (Reference 91 - 94) , all working with exchange- biased spin- valves (EB S Vs) of the form AFM/F1 /N/F2. Here the AFM lies outside the "active" GMR region of the two F layers and serves m ainly to "pin" the m agnetization of the adjacent F2 layer to a higher reversing (switching) field than that of the "free" Fl layer, leaving the Fl layer free to rotate at a lower field. The pinning is produced either by heating the sam ple to above the blocking tem perature of the AFM, applying a m agnetic field, and then cooling to room tem perature with the field on or by applying a m agnetic field during sam ple growth. W ei et al. (Reference 91 ) used a point contact to inject a high de current density j ~ 1 01 2 A/m 2 approxim ately CPP into an EB S V film . U razhdin and Anthony (Reference 92) sent a de CPP current density j ~ 5 x 1 01 1 A/m 2 into electron- beam -lithography-fabricated nanopillar EB S Vs. Tang et al. (Reference 93) sent a de current- in-plane (CIP) current density ~ 1 01 0 A/m 2 into an EB S V film with a m etallic AFM. D ai et al. (Reference 94) sent an ac CIP current density j ~ 1 09 A/m 2 into an EB S V film with an insulating AFM. In all of the studies (Reference 91 - 94) , it was found that a sufficiently high bias current can influence the m agnetic reversal of the "pinned" Fl layer. As the pinning (exchange bias) is known to be associated with interfacial AFM m om ents (Reference 88, 89) , this observation can be taken as evidence of effects of the current on the AFM predicted in (Reference 84- 86) , given that other spurious effects (for exam ple, Joule heating) can be ruled out. 1 4 UNCLASSIFIED//EQB QEHCIAL USE ONLY UNCLASSIFIED //FOR OFFICIAL UOE ONLY The original observation of the effect was reported by W ei et al. (Reference 91 ) , who m easured m agnetoresistance of a point contact to EB S V film at room tem perature (~ 295K) with negative current flowing from the contact tip into the film . The sam ple geom etry is shown in Figure 1 2a. A point contact is used to inject a de current into a sputtered N/F1 /N/F2/AFM/N = Cu(50 or 1 00nm ) /CoFe(3 or 1 0nm ) /Cu(1 0nm ) /CoFe(3 or 1 0nm ) /FeMn (3 or 8nm ) /Au(5nm ) m ultilayer (or inverted versions thereof— that is, Cu/FeMn/CoFe/Cu/CoFe/Au) . The sam ple is heated to ~ 450K (above the blocking tem perature of FeMn) and then cooled in a m agnetic field of 1 80 O ersteds to exchange­ bias the "pinned" layer F2 to a higher m agnetic field than needed to reverse the "free" layer F2. The top layer is covered by a 5- nm - thick layer of Au to protect it from atm ospheric contam ination. The m agnetic field H is applied in the plane of the layers and along the direction of exchange bias. Magnetic coupling between the two F layers should be negligible, because the N layer is thick enough (1 0 nm ) to elim inate exchange coupling, and the two F layers are wide enough (~ m m ) to m inim ize dipolar coupling. The bottom N layer is Cu, m ade thick enough (50 or 1 00 nm ) to approxim ate an equipotential, thereby generating an approxim ately CPP current flow through the F1 /N/F2/AFM EB S V. Figure 12. (a) Schematic of point contact to sample geometry. Omitted is a 5-nm-thick protective Au capping layer between the point contact and the multilayer. The bottom N layer is also much thicker than shown to help produce a nearly CPP current, (b) R (vertical scale) versus applied magnetic field B for a series of currents I. The solid black curves are hysteresis curves starting from large positive B and finishing at large negative B. The grey curves start at large negative B and finish at positive B. In the dark curves, the "free" layer, Fl, switches at ~ -5 ml and the pinned layer, F2, switches at fields ranging from below -40 mT (large positive I) to about - 60 mT (large negative I), (c-e) Grey-scale plots of R versus B for different values of I. W hite is maximum R (antiparallel state) and black is minimum R (parallel state). Unes are linear fits to the data at 30 percent (dashed white), 50 percent (solid white) and 70 percent (dashed black) of maximum R, Sample (c) is the sample of (b), with the AFM layer on the bottom (furthest from the point contact). Sample (d) is similar to (c), except inverted, so that the AFM layer is on top (closest to the point contact). Sample (e) differs from (b) only in that it has two equally thick (8-nm) F layers. (Reference 91 ) Figure 1 2b shows m agnetoresistance curves for a series of currents I applied to a point contact with resistance R = 0.92 Q. The dark curves show sweeps from positive to negative field, and the lighter curves show sweeps back down from negative to positive field. For this contact, I = 30 m A corresponds to j ~ 2 x 1 08 A/cm 2. For dark curves, a large positive field + B along the pinning direction causes the m om ents of both the "free" and "pinned" F layers to point along + B , producing the m inim um resistance Rp in this parallel orientation of the two Fs. Reducing the m agnitude of B , the "free" F2 layer reverses at a sm all- m agnitude negative B , giving the m axim um resistance Rap in antiparallel configuration. Finally, a larger m agnitude negative B breaks the exchange­ bias pinning of the "pinned" Fl layer, and its m om ent rotates to along - B , returning the U N C LASSI FI E D/ /W MW CIAUUS&AMLX 1 5 UNCLASSIFIED //FOR OFFICIAL UCE ONLY sam ple to Rp. For lighter curves of back- sweeps (Figure 1 2b) , the m easured resistance follows these changes in reverse order. Focusing on the dark curves, we see that the switching field of the free layer Fl is essentially independent of the m agnitude of I and shows little broadening. In contrast, the switching field of the pinned layer F2 broadens significantly as the m agnitude of I increases, and the m idpoint of the switching also shifts with I, increasing for - I and decreasing for + I. S im ilar behaviors are seen also in the lighter curves. O pposite shifts for + I and - I indicate these shifts cannot be due to Joule heating, which should cause shifts in the sam e direction for both directions of I. B ut Joule heating m ight contribute to the broadening of the switching transitions. The shifts of the dark curves are specified m ore clearly in Figure 1 2c, which shows grey- scale plots of the heights of the curves in Figure 1 2b, with white representing the antiparallel state of m axim um resistance, and black the parallel state of m inim um resistance. D ata for three representative contacts (out of 29) on three different sam ples show that the behavior of interest is not lim ited to a single sam ple or contact, and that sim ilar results are generally obtained for straight line fits to 30 percent (white dashed lines), 50 percent (solid white lines), and 70 percent (solid black lines) of the m axim um change in R. The sam ple in Figure 1 2d is a contact with R = 1 .6 Q to an inverted version of the sam ple in Figures 1 2b and c, so that the "directions" of currents are reversed. The sam ple in Figure 1 2e has equal thickness Fl = F2 = 3- nm layers. All three sam ples show the sam e features— that is, electrons passing through F2 into the AFM layer enhance pinning, and electrons passing through the AFM layer into F2 reduce it. S im ilar to the case of AFM= CoFe, a negative current density ~1 01 2 A/m 2 injected through the F= CoFe into an AFM= IrMn/CoFe interface (Reference 95, 96) was found to increase the exchange bias, while a positive current decreased it. W ei et al. (Reference 91 ) proposed the following qualitative explanation for these asym m etric changes in switching (exchange bias) field with current. Near the switching field, the m etastability of F2's opposite to field orientation is due alm ost entirely to exchange interactions with uncom pensated m om ents in the surface layer of the antiferrom agnet. S om e of these spins are pinned, thereby inducing an energy barrier for ferrom agnetic layer spin reversal (Reference 88, 89) . Electrons flowing from F2 into AFM induce torques on m om ents in the AFM m atrix, altering its m agnetic configuration (Reference 84) . These S TT torques tend to favor parallel alignm ent of m om ents at the F2/AFM interface and will therefore tend to increase the exchange bias field. Electrons flowing in the opposite direction will tend to have the opposite effect. The observed variations in exchange bias m ediated by an electrical current thus can be taken as good evidence of the S TT effect in AFM. However, such transport m easurem ents do not distinguish between effects of the current on the bulk AFM and those on interfacial AFM m om ents, and m ore elaborate techniques are needed to obtain a detailed understanding of the phenom enon. 5. Summary and Conclusions The sem iconductor industry has distinguished itself by a long- term trend known as Moore's law (Reference 97) that foresees an exponential increase in transistor density on a chip, doubling approxim ately every 2 years. Continuing at this pace, the transistor density will reach ~ 1 01 3 cm '2 by 2035, which at clock speeds of ~1 0 GHz would result in ~40 MW /cm 2 of power dissipated on a chip. U nless the energy dissipation from 1 6 UNCLASSIFIED/ /FOR OFFICIAL USE ONLY UNCLASSIFIED //FOR OFFICIAL USE ONLY transistor switching can be reduced dram atically, the therm al load associated with 40 MW /cm 2 will exceed that in a rocket nozzle. The failure of therm al m anagem ent on a chip m ight end the continued progress of the sem iconductor industry well before 2035. The International Technology Roadm ap for S em iconductors (http://www.itrs.net) has term ed this im m inent collapse the "Red B rick W all," where "Red" indicates no "known m anufacturable solutions" (of reasonable confidence) exist for continued scaling in som e aspect of the sem iconductor technology. The scenario above m otivates the search for signal- processing devices that dissipate very little energy when they switch. The em erging spintronic technology m ight offer such devices where inform ation is carried by spin— in contrast to CMO S transistors, where it is carried by charge— since spin has an inherent advantage over charge when it com es to energy dissipation. Therefore, if a reliable way can be found to control and m anipulate spins, spintronic devices could offer higher data processing speeds, lower electricity consum ption, and m any other advantages over conventional chips, perhaps including the ability to carry out radically new quantum com putations. A spintronic device calls for efficient m ethods to generate, conduct, process, and detect spin- encoded signals. W e have reviewed the physics and em erging applications of two principal spintronic phenom ena— giant m agnetoresistance and spin- transfer- torque— that provide m eans to detect (GMR) and m anipulate (S TT) the spin signals. GMR has already spawned m ajor technological change in the inform ation storage industry with the usage of GMR sensors and read heads and, along with tunneling m agnetoresistance (TMR) , is expected to continue to dom inate the detection of spin- encoded signals. S TT is a m ore recent developm ent in spintronics that provides an efficient m eans of controlling and m anipulating spin distributions on the nanom eter length scale and the picosecond- tim e scale, thus positioning S TT as the m ethod of choice for fast processing of spin signals in nanodevices. W hat is the future of spintronic applications? A num ber of new spintronic devices based on GMR and S TT have been proposed. These include high- frequency (GHz) oscillators, sources, and detectors, as well as m agnetic field sensors— for exam ple, in nonvolatile m em ories such as racetrack and S TT m agnetic random access m em ory (S TT- MRAM) . However, m uch fundam ental work rem ains to be done before we see com m ercial applications of these devices. For the m em ory industry, developm ent of these spintronic applications m ay lead to a universal m em ory that would com bine cost benefits of D RAM, speed of S RAM, and nonvolatility of flash RAM. Potentially all logic operations on a chip could be carried out by m anipulating spins in m etallic system s instead of m anipulating charges in sem iconductor transistors, as in conventional m icrochips. Moreover, such operations could be com bined on a chip with a universal m em ory. This would result in a new scalable and radiation- resistant electronics, com puters, and so forth. The radiation resistance would be of particular interest for aerospace applications because the radiation in space is known to severely dam age conventional electronics by building up a destructive charge in transistors. Long space trips that would expose onboard electronics to years of radiation would benefit from the radiation resistance and reduced power consum ption (for exam ple, like a nonvolatile m em ory that can retain the stored inform ation even when not powered) of m etallic spintronic devices. More generally, the im pact of reduced power consum ption in electronic devices is hard to overestim ate, as we rely on such devices in alm ost every aspect of our everyday lives. 1 7 UNCLASSIFIED//FOR OFFICIAL UOE ONL¥ UNCLASSIFIED //FOR OFFICIAL UOE ONLY Finally, m etallic spintronics and its applications discussed in this report are all based on already well- established physical phenom ena such as GMR and S TT. As the spintronics field is still in a relative state of infancy, new and m ore exciting phenom ena are likely to be uncovered in the future. 6. References (1 ) B aibich, M. N., et al., Phys. Rev. Lett. 61 , 2472 (1 988) . (2) B inasch, G., et al., Phys. Rev. B 39, 4828 (1 989) . (3) S lonczewski, J. C., J. Magn. Magn. Mater. 1 59, LI (1 996) . (4) B erger, L., J. Appl. Phys. 81 , 4880 (1 997) . (5) Tsoi, M., et al., Phys. Rev. Lett. 80, 4281 (1 998) . (6) Myers, E. B ., et al., S cience 285, 867 (1 999) . (7) S un, J. Z., J. Magn. Magn. Mater. 202, 1 57 (1 999) . (8) W egrowe, J.- E., et al., Europhys. Lett. 45, 626 (1 999) . (9) Tsoi, M., et al., Nature 406, 46 (2000). (1 0) B each, G. S . D ., et al., Phys. Rev. Lett. 97, 057203 (2006) . (1 1 ) B each, G. S . D ., et al., Phys. Rev. Lett. 1 02, 067201 (2009) . (1 2) Fert, A., B runo, P., in U ltrathin Magnetic structures II: An Introduction to the Electronic, Magnetic and S tructural Properties, ed. by J.A.C. B land and B . Heinrich (B erlin: S pringer, 1 994) , p.82. (1 3) Levy, P. M., in S olid S tate Physics, ed. by H. Ehrenreich and D . Turnbull, vol. 47 (B oston, London: Acad. Press, 1 994) , p.367. (1 4) D ieny, B ., J. Magn. Mag. Mat. 1 36, 335 (1 994) . (1 5) Gijs, M. A.M., B auer, G. E. W ., Adv. Phys. 46, 285 (1 997). (1 6) Anserm et, J.- Ph., J. Phys.: Cond. Mat. 1 0, 6027 (1 998). (1 7) B ass, J., Pratt, W . P., Jr., J. Magn. Magn. Mat. 200, 274 (1 999). (1 8) Fert, A., Piraux, L., J. Magn. Magn. Mat. 200, 338 (1 999) . (1 9) Gijs, M. A.M., in Magnetic Multilayers and Giant Magnetoresistance: Fundam entals and Industrial Applications, ed. by U . Hartm ann (B erlin: S pringer, 2000) , p.1 30. 1 8 UNCLASSIFIED//FOR OFFICIAL USE ONLY unclassified/^ren erneiAL uoc only (20) Coehoorn, P., in Magnetic Multilayers and Giant Magnetoresistance: Fundam entals and Industrial Applications, ed. by U . Hartm ann (B erlin: S pringer, 2000) , p.65. (21 ) B arthelem y, A., Fert, A., and Petroff, F., in Handbook of Magnetic Materials, edited by K. H. J. B uschow (Am sterdam : Elsevier, 1 999) , p.l. (22) Tsym bal, E.Y., Pettifor, D .G., in S olid S tate Physics, ed. by H. Ehrenreich and F. S paepen, Vol. 56 (Academ ic Press, 2001 ) pp.1 1 3- 237. (23) Ashcroft, W ., Merm in, D ., S olid S tate Physics (S aunders College Publishing, Forth W orth, 1 976) . (24) Ciureanu, P., in Thin Film Resistance S ensors, ed. by P. Ciureanu, S . Middelhoek (Institute of Physics Publishing, B ristol, 1 992) , p. 253. (25) Electrocrystallization, ed. by R. W eil, R. G. B arradas (The Electrochem ical S ociety, Pennington, 1 981 ) . (26) Lashm ore, D . S ., D ariel, M. P., J. Electrochem . S oc. 1 35, 1 21 8 (1 988). (27) O hring, M., The Materials S cience of Thin Film s (Academ ic, B oston, 1 992) . (28) S laughter, J. M., Pratt, W . P., Jr., and S chroeder, P. A., Rev. S ci. Instrum . 60, 1 27 (1 989). (29) Parkin, S . S . P., More, N., and Roche, K. P., Phys. Rev. Lett. 64, 2304 (1 990) . (30) Parkin, S . S . P., B hadra, R., and Roche, K. P., Phys. Rev. Lett. 66, 21 52 (1 991 ) . (31 ) Fert, A., Cam pbell, I. A., J. de Physique 32, Cl- 46 (1 971 ). (32) Mott, N. F., Proc. Roy. S oc. London, S er. A 1 53, 699 (1 936) , and 1 56, 368 (1 936), and Adv. Phys. 1 3, 325 (1 964) . (33) Pratt, W . P., Jr., et al., Phys. Rev. Lett. 66, 3060 (1 991 ). (34) D ieny, B ., et al. J. Appl. Phys. 69, 4774 (1 991 ) . (35) Piraux, L, et al., Appl. Phys. Lett. 65, 2484 (1 994) . (36) B londel, A., Meier, J. P., D oudin, B ., Anserm et, J.- Ph., Appl. Phys. Lett. 65, 301 9 (1 994). (37) Tsoi, M., Jansen, A. G. M., B ass, J., J. Appl. Phys. 81 , 5530 (1 997) . (38) S m ith, C., Proc. U K Conf, for S ensors, Test and Measurem ent Technology, B irm ingham , England, Jan. 22- 23, 1 997. (39) D aughton, J. M., et al., IEEE Trans. Magn. 30, 4608 (1 994) . 1 9 UNCLASSIFIED//FOn OFFICIAL USE ONLY UNCLASSIFIED//FOR OFFICIAL USE ONLY (40) Ji, Y., Chien, C. L., S tiles, M. D ., Phys. Rev. Lett. 90, 1 06601 (2003). (41 ) Rippard, W . H., Pufall, M. R., Kaka, S ., Russek, S . E., S ilva, T. J., Phys. Rev. Lett. 92, 27201 (2004). (42) Katine, J. A., Albert, F. J., B uhrm an, R. A., Myers, E. B ., Ralph, D . C., Phys. Rev. Lett. 84, 31 49 (2000) . (43) Grollier, J., et al., Appl. Phys. Lett. 78, 3663 (2001 ) . (44) Liu, Y., Zhang, Z., W ang, J., Freitas, P. P., Martins, J. L., J. Appl. Phys. 93, 8385 (2003) (45) Moriya, R., Ham aya, K,, O iwa, A., Munekata, H., Jpn. J. Appl. Phys. Part 2 - Lett. Express Lett. 43, L825 (2004). (46) D ieny, B ., et al., Phys.Rev.B 43, 1 297 (1 991 ). (47) W aintai, X., Myers, E. B ., B rouwer, P. W ., Ralph, D . C., Phys. Rev. B 62, 1 231 7­ 1 2327(2000) (48) B raganca, P.M., et al., Appl. Phys. Lett. 87, 1 1 2507 (2005). (49) Rippard, W . H., et al., Phys. Rev. Lett. 95, 067203 (2005) . (50) Kiselev, S . I., et al., Nature 425, 380 (2003). (51 ) Krivorotov, I. N., et al., S cience 307, 228- 231 (2005) . (52) Acrem ann, Y., et al., Phys. Rev. Lett. 96, 21 7202 (2006) . (53) Kilcoyne, A. L. D ., et al., J. S ynchrotron Radiat. 1 0, 1 25 (2003). (54) S tbhr, J., et al., S cience 259, 658 (1 993). (55) Rippard, W . H., Pufall, M. R., Kaka, S ., S ilva, T. J., Russek, S . E., Phys. Rev. B 70, 1 00406 (2004). (56) Khitun, A., Nikonov, D . E., B ao, M., Galatsis, K., W ang, K. L., Nanotechnology 1 8, 465202 (2007). (57) S chneider, T., et al., Appl. Phys. Lett. 92, 022505 (2008) . (58) D riskill- S m ith, A. A. G., Huai, Y., Future Fab Inti. 23, 28 (2007). (59) B erger, L., J. Appl. Phys. 49, 21 56 (1 978). (60) B erger, L., J. Appl. Phys. 55, 1 954 (1 984). (61 ) B erger, L., Phys. Rev. B 33, 1 572 (1 986). 20 UNCLASSIFIED//EQB QEMCIAL USE ONLY UNCLASSIFIED//rOR OFFICIAL USE ONLY (62) B erger, L, J. Appl. Phys. 63, 1 663 (1 988) . (63) Freitas, P. P., B erger, L., J. Appl. Phys. 57, 1 266 (1 985) . (64) Hung, C.- Y., B erger, L, J. Appl. Phys. 63, 4276 (1 988). (65) Hung, C.- Y., B erger, L., S hih, C.Y., J. Appl. Phys. 67, 5941 (1 990) . (66) Yam anouchi, M., Chiba, D ., Matsukura, F., O hno, H., Nature 428, 539 (2004) . (67) Ravelosona, D ., et al., Phys. Rev. Lett. 95, 1 1 7203 (2005) . (68) Tsoi, M., Fontana, R. E., Parkin, S . S . P., Appl. Phys. Lett. 83, 261 7 (2003) . (69) Grollier, J., et al., Appl. Phys. Lett. 83, 509 (2003) . (70) B each, G. S . D ., Nistor, C., Knutson, C., Tsoi, M., Erskine, J. L., Nat. Mater. 4, 741 (2005). (71 ) B each, G. S . D ., Knutson, C., Tsoi, M., Erskine, J. L., J. Magn. Magn. Mater. 31 0, 2038 (2007) . (72) Hayashi, M., et al., Phys. Rev. Lett. 96, 207205 (2006). (73) Hayashi, M., et al., Nat. Phys. 3, 21 (2007). (74) Klaui, M., et al., Appl. Phys. Lett. 83, 1 05 (2003) . (75) Kim ura, T., O tani, Y., Yagi, I., Tsukagoshi, K., Aoyagi, Y., J. Appl. Phys. 94, 7226 (2003). (76) Vernier, N., Allwood, D . A., Atkinson, D ., Cooke, M. D ., Cowburn, R. P., Europhys. Lett. 65, 526 (2004). (77) Lautenberg, M., et al., Phys. Rev. Lett. 97, 046602 (2006). (78) Thom as, L., et al., Nature 443, 1 97 (2006) . (79) Yam aguchi, A., et al., Phys. Rev. Lett. 92, 077205 (2004) . (80) Yam aguchi, A., et al., Phys. Rev. Lett. 96, 1 79904 (2006) . (81 ) Klaui, M., et al., Phys. Rev. Lett. 95, 026601 (2005) . (82) Meier, G., et al., Phys. Rev. Lett. 98, 1 87202 (2007) . (83) Parkin, S . S . P., Hayashi, M., Thom as, L., S cience 320, 1 90 (2008). (84) Nunez, A. S ., et al., Phys. Rev. B 73, 21 4426 (2006) . 21 UNCLASSI FI ED//FQn OFFICIAL UPC ONLY UNCLASSIFIED/ /FOR OFFICIAL USE ONL¥ (85) Xu, Y., W ang, S ., Xia, K., Phys. Rev. Lett. 1 00, 226602 (2008) . (86) Gom onay, H., Loktev, V., Low Tem p. Phys. 34, 1 98 (2008). (87) W ei, Z., S harm a, A., B ass, 1 , Tsoi, M., J. Appl. Phys. 1 05, 07D 1 1 3 (2009) . (88) Meiklejohn, W . H., B ean, C. P., Phys. Rev. 1 02, 1 41 3 (1 956). (89) S tohr, J., S iegm ann, H. C., Magnetism : from fundam entals to nanoscale dynam ics (S pringer S eries in S olid- S tate S ciences, Vol. 1 52, S pringer, 2006) . (90) U razhdin, S ., Chien, C. L., Guslienko, K. Y., Novozhilova, L., Phys. Rev. B 73, 05441 6 (2006). (91 ) W ei, Z., et al., Phys. Rev. Lett. 98, 1 1 6603 (2007) . (92) U razhdin, S ., Anthony, N., Phys. Rev. Lett. 99, 046602 (2007) . (93) Tang, X.-L., Zhang, H.- W ., S u, H., Zhong, Z.-Y., Jing, Y.- L, Appl. Phys. Lett. 91 , 1 22504 (2007). (94) D ai, N. V., et al., Phys. Rev. B 77, 1 32406 (2008) . (95) B asset, J., S harm a, A., W ei, Z., B ass, J., Tsoi, M., Proc. S PIE 7036, 703605 (2008). (96) W ei, Z., B asset, J., S harm a, A., B ass, J., Tsoi, M., J. Appl. Phys. 1 05, 07D 1 08 (2009). (97) Moore, G. E., Electronics Magazine (McGraw Hill, New York, 1 965) . 22 UNCLASSIFIED//FnP AEHCIAL MGS OML¥