Applied spintronics: From spin transport to spin-charge interconversion and emerging devices Authors: Manuel Bibes, Olena Gomonay, Timo Kuschel, Jairo Sinova, Mathias Weiler, Shunsuke Fukami, Mathias Kläui Published: 2026 ABSTRACT Spintronics exploits the electron’s spin degree of freedom to go beyond the capabilities of conventional charge-based electronics. Since the discovery of giant magnetoresistance, the field has evolved from the study of spin-polarized transport in ferromagnets to a broader framework encompassing spin-orbit coupling, collective spin dynamics, and electric-field control of magnetic states. In this Review, we discuss the physical mechanisms that underpin modern applied spintronics, with particular emphasis on spin-charge interconversion, magnetoresistive effects, and the electrical detection and manipulation of collective spin phenomena. We examine how spin currents, spin-orbit torques, and voltage-controlled magnetic properties enable efficient control of magnetization and magnetic excitations in nanoscale devices. These mechanisms form the basis of key technologies such as magnetic sensors and magnetic random-access memory (MRAM) while also opening pathways toward emerging architectures including racetrack memories, spin-based logic, and unconventional computing. By highlighting common principles across materials platforms and device concepts, we provide a perspective on the current state of spintronics and outline promising directions for future spin-based electronics. INTRODUCTION Spintronics explores the active role of the electron spin in transport, dynamics, and information processing, with the long-term objective of complementing or extending conventional charge-based electronics (1, 2). Early developments, most notably giant magnetoresistance (GMR) (3, 4), established the technological relevance of spin-dependent transport and triggered sustained efforts to integrate magnetic functionalities into microelectronic architectures (5). Over the past two decades, the field has undergone a substantial conceptual broadening, driven by advances in materials growth, nanoscale characterization, and the recognition of spin-orbit coupling (SOC) as a unifying framework for spin-charge interconversion. A defining feature of modern spintronics is the shift from purely spin-polarized transport in ferromagnets toward mechanisms that exploit relativistic effects and interfacial symmetry breaking. SOC enables the generation and detection of spin accumulation and spin currents in nonmagnetic systems and efficient electrical control of magnetic order. These effects are intrinsically compatible with planar device geometries and do not rely on external magnetic fields, making them particularly attractive for scalable technologies. At the same time, collective spin phenomena have moved to the forefront of both fundamental and applied research. Magnetization textures, spin excitations, and topologically nontrivial spin states introduce additional degrees of freedom that can be manipulated electrically and detected through transport signatures. Their behavior reflects an interplay between exchange interactions, SOC, and electronic structure, often amplified at interfaces or in reduced dimensions. Understanding this interplay is essential for establishing reliable control schemes and for assessing the limits of energy efficiency and speed. A second major trend concerns the increasing emphasis on voltage-based control. While current-driven torques remain central to many device concepts, electric-field effects offer qualitatively different advantages by enabling reversible and local modulation of magnetic properties with minimal dissipation (6). Voltage-controlled magnetic anisotropy and magnetoelectric coupling exemplify this approach, providing routes to nonvolatile control that are not limited by Joule heating. These mechanisms naturally connect spintronics with ferroelectricity and multifunctional oxides as well as two-dimensional materials, broadening the materials landscape and the range of achievable functionalities. This Review focuses on applied aspects of spintronics, with particular attention to spin-charge interconversion, electrical detection and control of collective spin states, and their implications for device concepts beyond conventional memories. Rather than aiming for exhaustive coverage, we emphasize physical mechanisms that have reached a sufficient level of maturity to impact realistic device strategies while highlighting emerging directions that challenge existing paradigms. By articulating common physical principles across materials classes and device architectures, we aim to provide a coherent perspective on the present status and future directions of spin-based electronics. SPIN-CHARGE INTERCONVERSION s-d exchange coupling The s-d exchange interaction (Fig. 1) is the microscopic engine that creates spin-polarized currents, mediates spin torques, and determines spin-dependent scattering in virtually all solid-state spintronic technologies. It is the microscopic interaction that ties conduction-electron spins to localized magnetic moments. The introduction of the s-d coupling was initially considered for describing the interaction between conduction electrons and impurities. Later, this concept was applied in spintronics to describe the interaction between current and magnetic textures. In ferromagnets, the s-d exchange interaction is the basic model interaction responsible for the spin polarization of the charge current and also mediates a spin-transfer torque (STT) that enables all-electrical switching in spin valves, nanopillars, STT-MRAMs (7), and racetrack memory concepts (8). Exchange, spin-orbit coupling, and charge and spin currents. (Left) Elementary interactions between electron spin and the lattice. (Top) s-d exchange, describing the local coupling between the spin of a conduction s-electron and d-electron magnetic moments. (Bottom) Spin-orbit coupling, where the motion of an electron in the electric field of lattice ions produces an effective magnetic field that interacts with the electron spin. (Right) Different regimes of electrical transport. An unpolarized current carries equal numbers of spin-up and spin-down electrons. A spin-polarized current has unequal spin populations. A fully spin-polarized current contains only one spin orientation. A pure spin current corresponds to opposite flows of spin-up and spin-down electrons, producing spin transport without net charge current. The s-d exchange also plays an important role in the current-induced dynamics of antiferromagnets (9) and the recently found altermagnets (10, 11). In particular, for symmetry protected compensated magnetic systems, STT mediated by s-d exchange can induce stable precession of magnetic moments, corresponding to the spin-torque oscillator (STO) regime (9). SOC and spin-charge interconversion SOC is the relativistic interaction connecting the spin of an electron and its motion (Fig. 1), arising from the expansion of the Dirac equation at low particle velocities and serving as one of the fundamental interactions from which many topological effects arise. While SOC has long been associated with the dephasing of spin polarization through well-known mechanisms as the Dyakonov-Perell mechanism (12), in the past several decades, with the advent of the intrinsic spin Hall effect (SHE) in particular (13–15), SOC has become a key source of topological physics and spin current (Fig. 1) and spin accumulation generation in spintronics. This weak relativistic effect, many orders of magnitude too weak in free electrons, is strongly renormalized by the connected band structure in solids, often increasing the connected Compton wavelength scale by many orders of magnitude relative to its free electron parameter, although albeit always weaker than atomistic electron volt scales. Two fundamental effects originating from SOC are the SHE (Fig. 2A), where a spin-dependent deflection arising from SOC gives rise to a spin current, and the Edelstein effect (16)—sometimes termed Rashba-Edelstein effect (Fig. 2B)—where a charge current generates a nonequilibrium spin accumulation in systems with broken inversion symmetry and strong SOC. Spin Hall and Edelstein effects. (A) Sketch of the SHE. Spin-dependent deflections due to the spin-orbit coupling generate a pure spin current transverse to a charge current. Its polarization is perpendicular to the plane defined by the charge and spin current (z-direction in the figure). (B) Sketch of the Edelstein effect. In this effect a nonequilibrium homogeneous in-plane spin polarization is generated by a charge current. It is a companion effect of the SHE. The SHE, and its inverse effect (where a spin current is injected in a material with strong SOC, such as Pt, to convert this spin current into a charge current), is principally connected to the Berry’s curvature in momentum space, which can be viewed as generating emergent magnetic monopoles in momentum space giving rise to a spin-dependent deflection (17). While some of this can originate from scattering from impurities, it is the intrinsic contribution arising from the entanglement of the electronic band structure through the SOC that gives this Berry’s curvature-dominated effect. As we discuss below, it has strong parallels to the anomalous Hall effect (AHE) measured in magnetic system with certain magnetic point group symmetries and is associated with a transverse Hall-like response; in real devices, however, this does not eliminate the longitudinal dissipation associated with charge transport. The Edelstein effect on the other hand is a dissipative effect, proportional to the diagonal conductivity of the system. It gives rise to a spin accumulation, rather than a spin current, and is often exploited at an interface to generate diffusive spin currents. In contrast to the SHE, its existence requires broken inversion symmetry, which is of course naturally occurring at interfaces. As we discuss below in the “Spin-orbit torque” section, these two effects—combined with STT—are at the heart of spin-orbit torques (SOTs), now at the forefront of the new generation MRAM on-chip technology. This has created the subfield of spin orbitronics (18, 19). As a future perspective, these effects have been generalized to magnetic systems, both ferromagnetic, antiferromagnetic, and recently altermagnetic. The Edelstein effect has now in fact been recently predicted in the so-called p-wave magnets arising from nonrelativistic effects with unique feature but a direct analogy of current-induced spin accumulation (20). ELECTRICAL DETECTION OF COLLECTIVE SPIN Anomalous Hall effect The AHE is a fundamental magnetotransport effect that has been known even longer than the spin itself, discovered by Hall himself in 1881 (21). It was understood from the outset that its origins were beyond the normal Hall effect, giving already effective magnetic fields of several tens to hundreds of Tesla inside a magnetic material. The key role of spin, SOC, and the electronic band structure of solids became clear in the early theories trying to explain the microscopic origin of the AHE in the middle of the 20th century. However, the confusing debate between different contributions connected to impurity scattering and intrinsic band structure effects, muddled the field considerably for many decades—generating often more confusion than understanding (22). Two decades ago, with a better understanding of the topological nature of Hall transport, the understanding of the microscopic theory developed enough to bring all known linear response approaches and materials issues at a consistent basis (22, 23). This led to a deeper understanding of the dominance of the intrinsic AHE microscopic contribution to the effect—now virtually the only contribution being calculated in SOC magnetic materials—which in turn paved the way to the discovery of the intrinsic SHE and the field of topological insulators. At the same time, the emergence of the topological AHE arising from noncollinear spin textures, giving rise to effective magnetic fields connected to the textures and their effective real space Berry’s curvature, has played a key role in the development of the skyrmionics field (24). In the past decade, AHE has continued to pave the way to new surprising discoveries. AHE has also played a key role in compensated magnetic systems. The AHE was predicted first to exists in certain noncollinear compensated magnetic systems, where it was clear that the broken symmetries led to the effect from magnetic symmetry considerations (25–27). What was perhaps more unexpected was the prediction in collinear symmetry protected compensated systems—its first prediction being on d-wave symmetric systems (28)—which led to the discovery of a third distinct class of compensated collinear magnetic systems, namely, altermagnets (10). Magnetoresistive effects Magnetoresistive effects are among the most direct and widely exploited manifestations of spin-dependent transport, providing an electrical readout of magnetic order. They arise from the dependence of charge transport on the relative orientation between carrier spins, magnetization, and current flow, thereby converting collective magnetic states into resistance variations. Their well-established physical basis, combined with robustness and scalability, explains their central role in both fundamental spintronics and device applications. Anisotropic magnetoresistance (AMR) represents the simplest form of this coupling (Fig. 3A) (29, 30). In ferromagnetic conductors, resistivity depends on the angle between the magnetization and the electrical current, reflecting spin orbit–induced anisotropies in scattering. Microscopically, this effect originates from the interplay between SOC and the electronic band structure, which renders momentum relaxation sensitive to magnetization orientation. Although AMR amplitudes are typically limited to a few percent, the effect has been extensively used to study magnetization reversal, domain evolution, and current-driven dynamics in metallic ferromagnets. Magnetoresistance effects in ferromagnetic devices. (A) AMR occurs in a single layer of a ferromagnetic conductor. The resistance depends on the angle between the magnetization and the current. Typical amplitude are in the range of 1%. (B) GMR is observed in multilayers where ferromagnetic layers are separated by metallic spacers. Co/Cu and Fe/Cr systems are the most common. Depending on the magnetization configuration—parallel or antiparallel—the resistance is low or high. At room temperature, GMR reaches a few percent. (C) Tunnel magnetoresistance (TMR) occurs in trilayers where two magnetic layers are separated by an ultrathin insulator so that electrons must cross by quantum-mechanical tunneling. Early experiments with amorphous aluminum oxide barriers showed TMR values in the 20–30% range. With epitaxial MgO barriers the TMR can reach 200% at room temperature. M is the magnetization, and J is the electrical current. GMR is observed in multilayer heterostructures composed of alternating ferromagnetic and non-magnetic layers (Fig. 3B) (3, 4). In these systems, resistance depends strongly on whether adjacent magnetizations are aligned parallel or antiparallel, reflecting spin-dependent scattering and transmission combined with finite spin diffusion lengths. GMR enabled a step change in magnetic sensor sensitivity and laid the foundations of modern spin-based information technologies. Beyond applications, it established a quantitative framework for describing spin injection, accumulation, diffusion (31), and relaxation in nanoscale conductors, concepts that remain central to spin transport theory. Tunnel magnetoresistance (TMR) further enhances magnetoresistive signals by exploiting quantum tunneling through an insulating barrier (Fig. 3C) (32). In magnetic tunnel junctions (MTJs), conductance depends on the spin polarization (33) and symmetry of electronic states at the ferromagnet-insulator interfaces (34). The strong sensitivity of tunneling to interfacial electronic structure makes TMR both a powerful readout mechanism and a sensitive probe of interface quality. Advances in barrier materials and interface control have led to resistance contrasts of several hundred percent at low (35) and room temperature (36, 37), establishing TMR as the dominant readout scheme in magnetic random-access memory (MRAM). At the same time, TMR provides insight into spin filtering, symmetry selection, and interfacial hybridization (38). Magnetoresistive effects also arise in nonuniform magnetic configurations. Domain walls introduce spatially varying magnetization, leading to additional resistance contributions associated with spin mistracking and local spin accumulation (39). These effects enable electrical detection of domain-wall position and motion. In materials with strong SOC, topologically nontrivial textures such as skyrmions (40) generate characteristic transport signatures linked to their real-space spin structure, providing purely electrical means to detect and manipulate magnetic textures. In systems with strong spin-orbit interaction, magnetoresistance can also originate from interfacial coupling between spin currents and magnetic order. Spin Hall magnetoresistance (SMR) (41, 42) reflects the modulation of interfacial spin absorption by magnetization orientation and probes the interaction between spin orbit–generated spin currents and collective magnetization, rather than spin-polarized charge transport within the magnet. Magnetoresistive effects play a key role in the time domain. Magnetization precession under resonant excitation modulates resistance, enabling electrical detection of ferromagnetic resonance and spin torque–driven oscillations. These techniques provide compact, on-chip access to magnetic dynamics and quantitative information on damping, anisotropy, and torque efficiencies. Together, magnetoresistive effects form a versatile electrical toolbox for probing static and dynamic spin phenomena, combining fundamental insight with exceptional compatibility with scalable device architectures. COLLECTIVE SPIN TRANSPORT Spin currents and nonlocal detection of spin transport Incoherent spin currents, e.g., angular momentum transport of multiple frequencies and uncorrelated phase relations, have been studied in magnetic metals and insulators with the main carriers being conduction electrons and magnons, respectively. The experiments are based on two main devices that generate the spin current either by diffusive spins from a magnetic material into a nonmagnetic spin conductor or by the SHE that injects the angular momentum into a magnetic material via spin-flip scattering and STT. The spatial separation of generation, transport, and detection gives this spin transport the suffix “nonlocal.” Based on the idea and first work of Johnson and Silsbee (43), pioneering experiments were done by Jedema et al. (44) who injected spins electrically into a magnetic metal at room temperature. Two spatially separated magnetic strips are connected via a nonmagnetic channel. A charge current in one magnetic strip causes a spin accumulation that diffuses into the nonmagnetic material and can be detected in the second magnetic strip as nonlocal voltage arising from spin-dependent electrochemical potentials. The obtained lateral spin-valve signal can be manipulated by the magnetic state of the injector and detector electrode as well as by the choice of the materials (45, 46). For example, clean in situ prepared interfaces (47) and highly spin-polarized Heusler compounds (48) can support large spin-valve signals. The lateral spin-valve response allows to extract several physical parameters such as the spin polarization of the magnetic material and the spin diffusion lengths of the nonmagnetic material (47), as well as the ratio between the spin and charge current (45). While ferromagnetic nanostructures have been the focus of the initial lateral spin-valve studies, later, two-dimensional (2D) materials were investigated (49) and in recent times even altermagnetic materials (50). In 2015, Cornelissen et al. (51) reported on the lateral long-distance transport of magnon spins in a ferrimagnetic insulator at room temperature. They realized the spin injection via the SHE from a Pt strip, thereby polarizing thermal magnons, and then the detection via the inverse SHE (ISHE) in a second Pt strip (Fig. 4). Slowly alternating charge currents and lock-in technique, as well as the dependence of the SHE injection and ISHE detection on the magnetic field direction, have been used to disentangle electrically and thermally induced spin currents. Shortly after, other groups followed the same approach (52) and investigated vertical magnon spin transport through insulators as well (53, 54). Spin injection, transport and detection scheme. Schematic of nonlocal transport with two nonmagnetic wires (blue) where spins are injected (left wire using, e.g., the SHE) and detected (right wire using, e.g., the inverse SHE). The study of the injector-detector distance dependence revealed a magnon spin diffusion length of up to 10 μm in Y3Fe5O12 (YIG) and hematite and slightly lower values in other materials, e.g., in ferrites (55, 56). In addition, sign changes in the injector-detector distance dependence have been identified (57–59) and can be understood by the magnon chemical potential landscape of thermally induced magnons (60). Magnon transport effects such as the magnon planar Hall effect and magnon AMR have been demonstrated in four-strips nanostructures (61). The nonlocal magnon spin transport has been identified in antiferromagnets (62), altermagnets (63), multiferroics (64, 65), 2D materials (66, 67), and ultrathin YIG films reaching the 2D transport regime (68). Multiterminal devices enable logic operations (69) and realize the magnon spin transistor concept (70–72). Furthermore, these devices have been used to discuss spin superfluidity (62, 72, 73) and the magnon Hanle effect (74, 75), indicating that this type of spin transport can be in some cases not fully incoherent. We note that recently the generation of orbital angular momentum has been discussed (76), and surprising, nonreciprocities have been observed in the nonlocal transport when orbital angular momentum is involved (77). Spin waves Coherent spin waves are excitations of the magnetic order parameter with a single frequency and a well-defined phase correlation. The quanta of these excitations are called magnons, and the research field of coherent spin-wave transport is consequently referred to as magnonics. Bloch introduced the concept of spin waves to explain the temperature-dependence of magnetization (78). Subsequent foundational contributions to spin wave theory are the mapping of spin operators to bosons (79) and the description of nonlinear spin-spin interactions (80). While the interaction of spins is caused by the exchange interaction at a nearest-neighbor level responsible for magnetic order (81), dipolar spin-spin interactions (82) become dominant in the long-wavelength limit. Generally, both exchange and dipolar coupling contributes to the spin-wave energy and dipolar-exchange spin waves have a generally anisotropic dispersion (83). The transport of coherent spin waves leads to intriguing wave phenomena, such as diffraction (84) and refraction (85). The prototypical device in coherent magnonics consists of a magnetic thin film or stripe onto which a pair of microwave antennas is lithographically defined. Spin waves are launched by the application of microwave power to one antenna and propagate to the second antenna, where they are detected electrically. This process is fully coherent, such that a microwave delay line with well-defined amplitude and phase transmission characteristics is formed. The transport of coherent spin waves can thus be used for analog signal processing (86). For these applications, one takes advantage of the short wavelength of spin waves compared to electromagnetic waves at the same frequency. At 10-GHz frequency, magnon wavelengths are in the nanometer to micrometer range compared to centimeter range for photons. Moreover, the spin-wave transmission can be tuned by external magnetic fields and is generally nonreciprocal. The short wavelength of magnons also allows to design their dispersion by micro- and nanopatterning. To this end, periodic structures are frequently used, and a magnonic crystal with tailored properties is formed (87). Recently, it has been shown that the magnon dispersion can be controlled not only by periodic spatial modulation but also by periodic modulation in time, leading to Floquet engineering of the magnon dispersion (88). While coherent magnons can be generated and observed in many different electrically conductive and insulating magnetic materials, many seminal experiments have used the ferrimagnetic insulator YIG (89), which offers the longest spin-wave lifetimes of any known material. Using thin-film YIG, researchers have demonstrated that the spontaneous formation of room-temperature magnon Bose Einstein condensates under parametric pumping (90). The nonlinear interaction of spin waves can be exploited for computing (91). Neuromorphic computing that takes advantage of nonlinear spin-wave phenomena is of particular interest because magnon-based analog computing devices can be reprogrammed on the fly by manipulating the magnetic environment (92). ELECTRICAL CONTROL OF COLLECTIVE SPIN PHENOMENA Conventionally, magnetization has been manipulated by magnetic fields. While this approach has been used very successfully and is still being used in write heads of hard drives, it suffers from poor scaling: As one reduces the size of devices, the switching fields cannot be decreased because thermal stability actually calls for larger switching fields for reduced magnetic volume. To keep the generated field large, the current has to be large, leading to increasing current densities for reduced dimensions, which eventually becomes unsustainable. Therefore, using electrical currents to control spin has recently become the method of choice and is being used in current state-of-the-art magnetic random access memory. The effects scale with the current density, which is thus kept constant during miniaturization. This favorable scaling then reduces the power consumption with decreased device sizes. By electrically generating spin and orbital angular momentum, currents can be used to manipulate magnetization. In this section, the pertinent mechanisms are discussed: Generating spin currents by using magnetic materials that lead to spin-polarized charge currents and using these to switch the magnetization is called STT (the “Spin-transfer torque” section), using nonmagnetic materials that lead to a charge to spin and orbital conversion and thereby to pure angular momentum currents (the “Spin-orbit torque” section). Beyond using currents that entail Ohmic losses, a potentially energetically more favorable approach is to use electric voltages. These can change the magnetic anisotropy leading to switching (the “Voltage-controlled magnetic anisotropy” section) and magnetoelectric coupling effects manipulate the magnetization (the “Magnetoelectric coupling” section). Spin-transfer torque The discovery of electrical control of magnetization orientation by STT represented a paradigm shift in spintronics—it moved the field of spintronics from passive detection of magnetic states to their active command. The basic idea of STT control can be summarized as follows (Fig. 5A): An electrical current flowing in a ferromagnetic metal is generally spin polarized—the number of up-spin and down-spin electrons that contribute to charge transport is different. The flow of charge of the electrical current is thus accompanied with a flow of net angular momentum. When this spin-polarized electronic current enters an adjacent second ferromagnetic metal in a heterostructure, and the local magnetization and nonequilibrium spin polarization are misaligned, this results in a torque on the magnetization due to absorption of spin angular momentum. Total angular momentum conservation dictates the magnitude of this torque. Spin-transfer and SOTs. (A) STT. (B) SOT via SHE. (C) SOT via Edelstein effect. (D) Magnetization switching via SOTs [see (146, 147)]. The theoretical inception of STT by Slonczewski (93) and Berger (94) provided the mathematical framework for this phenomenon. Early experimental validation relied on point-contact geometries (95) and, later, nanostructuring in the form of multilayered nanopillars (96). These experiments proved that high-density currents could induce deterministic switching of the magnetization (97), a finding that laid the groundwork for modern MRAM (see the “Magnetic random access memories” section). The STT efficiency is governed by the magnitude of the spin polarization and the magnetic damping of the material, as formalized in the modified Landau-Lifshitz-Gilbert equation (98–100). To enable the STT control, the two magnetic layers need to be decoupled. This is achieved by a metallic or insulating interlayer that suppresses the exchange interaction between the magnetic metals. In typical STT devices, one magnetic layer is “fixed” by means of large magnetic anisotropy, large thickness, or exchange bias by pinning its magnetization to an adjacent antiferromagnetic layer (101). The second layer is “free” by means of small anisotropy and reduced thickness, reducing the total magnetic moment that needs to be controlled. The electrical detection of the switching state and the microwave power generation from the magnetic oscillation of STOs rely on magnetoresistive effects, particularly GMR or TMR (see the “Magnetoresistive effects” section). The STT in such trilayers can be used to either electrically switch the magnetization of the free layer or induce oscillations in the magnetization of the free layer. While switching is the cornerstone for deterministic write processes in MRAM, current induced steady-state oscillations of the magnetization give rise to STOs. These nanoscale devices convert dc currents into microwave signals (102, 103). Several STOs can be synchronized by mutual locking (104), providing a hardware platform for neuromorphic computing where oscillators serve as physical neurons (105). The STT has so far been predominantly studied in ferromagnetic metals with natural spin polarization and is absent in conventional collinear antiferromagnets with vanishing spin polarization. The recently found altermagnets however have spin-split bands, resulting in the related spin splitter torque (106) that may open an avenue for efficient all-electrical control of unconventional magnets. Spin-orbit torque SOTs have emerged as a central mechanism in modern spintronics for the electrical manipulation of magnetic order and the architecture of modern MRAM technology. They arise from the coupling between the charge and spin degrees of freedom mediated by relativistic SOC, enabling electric currents to generate nonequilibrium spin polarizations and currents that exert torques on magnetic moments (18). SOTs provide a powerful alternative to conventional STT mechanisms by allowing efficient control of magnetization without requiring spin-polarized currents injected from magnetic layers. SOTs arise primarily from two microscopic mechanisms. In one, the SHE generates a spin current in a nonmagnetic material, which is then injected in the adjacent magnetic layer, exerting a torque through the usual STT (Fig. 5B). In another mechanism, at the interface where inversion symmetry is broken, the SOC generates a current-induced spin accumulation through the Edelstein effect (Fig. 5C), which in turn exerts a torque in the adjacent layer through either direct proximity exchange coupling or spin diffusion. In practice, both mechanisms contribute in current device geometries. There exists also a variant observed in diluted magnetic semiconductors, in which the internal Edelstein effect (connected to bulk broken inversion symmetry) of a magnetic system can give rise to a self-torque in the material (107, 108). These processes give rise to two principal types of torques acting on the magnetization: field-like and damping-like torques. The field-like torque acts similarly to an effective magnetic field generated by the current-induced spin polarization, tending to rotate the magnetization around the spin polarization axis. The damping-like torque, often associated with spin Hall currents entering the magnetic layer, modifies the magnetization dynamics in a way analogous to magnetic damping and is particularly efficient for switching magnetic states. The term “damping-like” should not be understood as an identification with intrinsic Gilbert damping. The relative magnitude of these torques depends strongly on the materials system, interface properties, and device geometry. A major advantage of SOTs is their ability to manipulate magnetization in ultrathin magnetic heterostructures using in-plane charge currents. Typical SOT devices consist of heavy metals with strong SOC (such as Pt, Ta, or W) adjacent to ferromagnets or antiferromagnets. In these structures, current-induced torques can drive domain wall motion, excite magnetic dynamics, or deterministically switch the magnetization direction. As described below, these capabilities are particularly promising for next-generation nonvolatile memory technologies, including SOT-MRAM, which offers advantages in switching speed, endurance, and energy efficiency compared with earlier STT devices (see the “Magnetic random access memories” section). Beyond ferromagnets, SOTs also enable electrical control of antiferromagnets, which have zero net magnetization but exhibit ordered spin sublattices. In certain crystal symmetries, current-induced spin polarizations can couple directly to the staggered magnetic order, allowing manipulation of the antiferromagnetic state through the so-called Néel SOT (109–111). This development has opened the rapidly growing field of antiferromagnetic spintronics, where ultrafast dynamics and robustness against magnetic perturbations offer potential technological advantages (112). The study of SOTs continues to expand into new material systems, including topological materials, two-dimensional crystals, and noncollinear magnets. These platforms offer opportunities to enhance torque efficiency and uncover new mechanisms linked to Berry curvature, topology, and symmetry. As a result, SOTs represent a unifying framework connecting relativistic electronic structure, spin transport phenomena, and practical device functionality, making them one of the key pillars of modern spintronics research. Voltage-controlled magnetic anisotropy Voltage-controlled magnetic anisotropy offers a direct means of manipulating magnetic order using electric fields, without relying on charge currents. Magnetic anisotropy determines the stability and orientation of magnetization, and even small voltage-induced modifications can substantially alter switching thresholds and dynamic behavior. The effect is most pronounced in ultrathin ferromagnets adjacent to insulating layers, where anisotropy is dominated by interfacial contributions. An applied electric field modifies the electronic structure at the interface, altering orbital occupancy and SOC, which in turn changes the anisotropy energy. This interfacial origin naturally favors scaling to nanometer dimensions and is compatible with standard device geometries. Experimentally, voltage-controlled magnetic anisotropy is commonly characterized by shifts in anisotropy fields or changes in coercivity under applied bias (113, 114) (Fig. 6, A and B). In MTJs with perpendicular anisotropy, voltage-induced modulation can significantly reduce the energy required for magnetization reversal or enable voltage-assisted switching schemes. These approaches are attractive for reducing write energy in nonvolatile memories, particularly when combined with thermal activation or weak magnetic fields. Electric-field control of magnetism. (A) Schematic of the sample used for a voltage-induced magnetic anisotropy change. (B) Magneto-optical Kerr ellipticity ηk for different applied voltages as a function of the applied field. The thickness of the Fe film was 0.48 nm. A significant change in the hysteresis curve indicated a large change in perpendicular anisotropy following application of the bias voltage. Right inset: voltage modulation response of the Kerr ellipticity dηk = dV. Left inset: magnetization direction at points A and B in the hysteresis curves. From (114). (C) Sketch of the heterostructure used to control magnetization amplitude with ferroelectricity. Applying a voltage between the conducting Nb-SrTiO3 substrate and the La0.88Sr0.12MnO3 (LSMO) film allows switching the polarization P of the ferroelectric PZT film. Depending on P direction, charge is accumulated or depleted in the LSMO, which modifies its magnetization amplitude as visible in (D). From (122). Beyond binary switching, voltage control enables continuous tuning of magnetic properties. Modulating anisotropy affects ferromagnetic resonance frequencies, domain-wall energies, and spin-wave spectra, providing a mechanism for electrically reconfigurable magnetic dynamics. This tunability is relevant for microwave devices, oscillators, and adaptive spintronic circuits (115). From a materials perspective, the magnitude and sign of the effect depend sensitively on interface chemistry, crystallographic orientation, and growth conditions. Oxide-based interfaces are of particular interest due to their strong electronic correlations and sensitivity to electric fields. However, achieving reproducible and stable voltage control remains challenging, especially in the presence of leakage currents and dielectric degradation. Voltage-controlled magnetic anisotropy exemplifies a broader strategy in spintronics: replacing dissipative current-driven control with electric-field-based mechanisms. While not universally applicable, it provides a compelling pathway toward low-power operation in nanoscale magnetic devices. Magnetoelectric coupling Magnetoelectric coupling provides a more general framework for electric-field control of magnetism by linking electric polarization and magnetic order (6, 116, 117). When resorting to ferroelectrics rather than dielectrics, magnetoelectric phenomena enable nonvolatile and multifunctional control schemes. Magnetoelectric coupling can arise in single phase materials such as multiferroics in which ferroelectric and magnetic orders are coupled. Regrettably, room-temperature multiferroics are scarce, with the notable exception of BiFeO3 that is ferroelectric and antiferromagnetic (118, 119). To circumvent this scarcity, hybrid or composite magnetoelectric systems can be engineered by combining a ferroelectric with a ferro- or ferrimagnet, for instance, in bilayers. These composite approaches offer considerable flexibility, as ferroelectric and magnetic materials can be optimized independently (120). Magnetoelectric effect can then be mediated by strain or electrostatic effects. In strain-mediated schemes, the piezoelectric deformation of the ferroelectric layer is transferred to the adjacent magnetic layer, inducing lattice distortions that modify magnetic anisotropy, exchange interactions, or coercivity. However, because the remanent strain state is generally identical for opposite polarization directions, strain-mediated magnetoelectric effects are typically volatile and mainly effective under applied bias. In contrast, electrostatic magnetoelectric coupling arises when ferroelectric polarization directly alters the interfacial electronic structure of the magnetic layer, modifying bonding, orbital hybridization, and carrier density and thereby producing nonvolatile changes in magnetic anisotropy (121), Curie temperature, or magnetization amplitude (122–125) (Fig. 6, C and D). Compared to voltage-controlled magnetic anisotropy, magnetoelectric approaches often provide enhanced nonvolatility and multifunctionality, as the ferroelectric polarization itself serves as a memory element. From a device perspective (126), magnetoelectric coupling is particularly attractive for architectures that combine memory and logic (127). By directly coupling electric and magnetic order parameters, it becomes possible to design devices in which information is stored, manipulated, and read out within a unified physical framework. This capability underpins several proposed beyond–complementary metal-oxide semiconductor (CMOS) logic concepts (128, 129). Substantial challenges remain before magnetoelectric effects can be widely exploited in practical devices. Achieving strong coupling at room temperature, ensuring endurance of ferroelectric layers, and integrating complex materials with semiconductor platforms are ongoing areas of research (130). Nonetheless, magnetoelectric coupling represents a key ingredient in strategies aimed at ultra-low-power and functionally rich spintronic systems. APPLICATIONS As discussed above, the development of spintronics that explores the interplay between charge and spin has clarified electrical transport phenomena depending on magnetic states, as well as the excitation of magnetic dynamics by electric currents and electric fields. The former has provided a means for the electrical detection of magnetic states, namely information sensing functionality, while the latter has enabled electrical control of magnetic states, namely information writing functionality. These advances have led to the commercialization of magnetic sensors and magnetic random access memory. Furthermore, in recent years, research has further expanded to encompass the interplay of charge and spin with electromagnetic waves, lattice vibrations, and heat currents, opening new avenues of various applications. The current status, key metrics, and challenges of the applications described in this section are summarized in Table 1. Table 1.Status, key metrics, and challenges of various applications using spintronic phenomena.Definition of abbreviations are as follows: AMR, anisotropic magnetoresistance; GMR, giant magnetoresistance; TMR, tunnel magnetoresistance; Field, magnetic field writing; STT (IP), STT writing for in-plane magnetized systems; STT (P), STT writing for perpendicularly magnetized systems; SOT, spin-orbit torque; DW, domain wall; MESO, magnetoelectric spin-orbit; FESO, Ferroelectric spin-orbit; p-comp, probabilistic computing; ANN, artificial neural network; SNN, spiking neural network; Reservoir comp., reservoir computing; CMOS, complementary metal-oxide semiconductor.ApplicationsStatusKey metrics, challengesSensorAMRProduct (early ‘90s)Sensitivity, resolution, linearity, response time, bandwidth, temperature drift, long-term stability, power consumption, sizeGMRProduct (late ‘90s)TMRProduct (mid ‘00s), expandingMRAMFieldProduct (mid ‘00s)Speed, power consumption, capacity, retention, endurance, temperature range, magnetic-field immunity, thermal budget, integration with advanced CMOS nodesSTT (IP)Product (early ‘10s)STT (P)Product (late ‘10s), expandingSOTPrototype (early ‘20s)Race-track memoryDWR&D (several bits to several tens of bits)Speed, power consumption, capacity, 3D architecture with vertical integration, reliabilitySkyrmionLogicMESOR&DPower consumption, speed, scalability, CMOS compatibilityFESOUnconventional computingp-comp.R&D (hundreds of bits)Power consumption, speed, scalability, variability, footprint, reproducibility, linearity/nonlinearity, CMOS compatibility, drift, magnetic-field immunityANN, SNNR&DReservoir comp.R&D Status, key metrics, and challenges of various applications using spintronic phenomena. Definition of abbreviations are as follows: AMR, anisotropic magnetoresistance; GMR, giant magnetoresistance; TMR, tunnel magnetoresistance; Field, magnetic field writing; STT (IP), STT writing for in-plane magnetized systems; STT (P), STT writing for perpendicularly magnetized systems; SOT, spin-orbit torque; DW, domain wall; MESO, magnetoelectric spin-orbit; FESO, Ferroelectric spin-orbit; p-comp, probabilistic computing; ANN, artificial neural network; SNN, spiking neural network; Reservoir comp., reservoir computing; CMOS, complementary metal-oxide semiconductor. In this section, we focus on magnetic sensors (the “Sensors and disk drive heads” section), MRAM (the “Magnetic random access memories”section), racetrack memory (the “Racetrack memories” section), logic (the “Spin-based logic” section), and unconventional computing (the “Unconventional computing” section) and provide an overview of recent progress and future prospects of these applications. Sensors and disk drive read heads Magnetic sensors based on magnetoresistive effects represent an early and successful application of spintronics. The discovery of the GMR effect in metallic multilayers (3, 4) and the subsequent development of the tunnel magnetoresistance (TMR) effect in MTJs (32, 131) enabled highly sensitive electrical detection of magnetic fields. Rapid progress in materials engineering and thin-film fabrication led to the commercialization of GMR- and later TMR-based sensors in the 1990s and 2000s. A landmark achievement was their implementation as read heads in hard disk drives (HDDs). The introduction of GMR read heads, followed by TMR-based heads, markedly enhanced field sensitivity and signal-to-noise ratio, enabling the exponential increase in HDD storage capacity over several decades. This technological impact established magnetoresistive sensing as a cornerstone of modern information infrastructure (132). In addition to HDDs, magnetoresistive sensors have found broad applications in automotive, industrial, and biomedical fields. Their high sensitivity, compactness, and room-temperature operation make them attractive for biomagnetic measurements, including magnetoencephalography and magnetocardiography (133). In addition, magnetostrictive materials have enabled mechanical sensors with high precision and robustness (134, 135). More recently, spin-caloritronic phenomena (136) such as the spin Seebeck effect and the anomalous Nernst effect have opened new pathways toward heat sensing. Together, these advances highlight the versatility of spintronic sensing technologies and their expanding role in both conventional and emerging applications. Magnetic random access memories MRAM is a kind of random-access memory that stores digital information as the magnetization direction, typically consisting of arrays of MTJs with peripheral circuitry. By appropriately designing the MTJ, the magnetization direction can remain stable against thermal fluctuations, enabling nonvolatile data retention. Moreover, as the magnetization reversal does not involve atomic displacement, intrinsically fast switching and high endurance are expected. These features provide MRAM with the capability of functioning as a nonvolatile working memory. The first generation of MRAM (Fig. 7A) used the TMR effect for readout and magnetic fields generated by currents, based on Ampère’s law, for writing (137). This field-writing MRAM was commercialized in the mid-2000s, but its adoption remained limited to niche markets, mainly due to constraints in capacity and cost, despite advantages such as strong radiation tolerance. MRAM using STT for writing (93, 96) has been developed since the early 2000s (138), leading to the commercialization of in-plane magnetized MTJ-based MRAMs in the early 2010s as the second generation (Fig. 7B). However, the required write current was still relatively high, restricting large-scale market penetration. A major breakthrough came with the discovery of perpendicular-easy-axis CoFeB/MgO-based MTJs reported in 2010 (139, 140), exhibiting superior read and write performance, robust retention, and excellent scalability. In the late 2010s, perpendicular STT-MRAM entered mass production at major foundries, marking the third generation of MRAM (Fig. 7C), with growing adoption in, e.g., IoT and automotive applications. MRAM roadmap. Magnetic tunnel junction (MTJ) cell structure for MRAM applications. That for (A) the first-generation MRAM with magnetic-field writing, (B) the second-generation MRAM with STT writing for in-plane magnetized MTJ, (C) the third-generation MRAM with STT writing for perpendicularly magnetized MTJ, and (D) SOT-MRAM. Key advantages of MRAM include nonvolatility, high speed, high endurance, scalability, low-voltage operation, compatibility with back-end-of-line integration, and a relatively wide operating temperature range. These properties are critical for replacing embedded SRAMs, DRAMs, and Flash memories in advanced nodes, and distinguish MRAM from other emerging nonvolatile memories such as resistive random-access memory and phase-change memory. Furthermore, its back-end-of-line compatibility opens opportunities for compute-near-memory and compute-in-memory architecture (141, 142). Looking ahead, continued scaling of MTJs toward advanced technology nodes is under active investigation (143–145). Also, for even higher-speed and higher-endurance applications, particularly high-level cache replacement, three-terminal MTJs are more suitable. In this context, leveraging spin–orbit torque (SOT) (Fig. 7D) is considered promising (146–151), and prototype SOT-MRAM chips have been demonstrated (152, 153). We here note that the typical random access frequency for the first--generation MRAM with magnetic-field writing (Fig. 7A), the second-and third-generation MRAM with STT writing (Fig. 7, B and C), and the SOT-MRAM (Fig. 7D) is about 30 MHz, 200 to 400 MHz, and 500 MHz to 1 GHz, respectively. Racetrack memories While hard drives and magnetic random access memory are established technologies in the market, they entail limitations for instance in terms of access speed (hard drives) and storage density (MRAM). Therefore, new technologies are being proposed, such the magnetic domain-wall racetrack memory that was first proposed by Parkin at IBM (see Fig. 8) (154). In the magnetic domain-wall racetrack memory device, the magnetization state of the magnetic domains is used as “0” and “1” bits. At the writing element, the bits are written as illustrated in Fig. 8D. The written bits can be pushed forward or backward by shifting all DWs in the same direction by the use of an external magnetic field (155) or electric currents (154). Thus, moving these domains by moving the domain walls (156) result in moving the bits as desired. Then, the encoded bits can be read by the stationary reading element, which can be a MTJ on the track (see Fig. 8C). The racetrack memories have many advantages compared to the conventional memory devices (157). Compared to the conventional HDD there are no mechanically moving parts, thus possibly the racetrack memory can be faster beyond the mechanical limitations, and the problems due to mechanical failure do not exist. Another possible advantage of this racetrack memory is that it may allow one to construct 3D devices orienting the racetrack upward and store bits in a larger 3D space (see Fig. 8E), which may result in a high-storage density while achieving high speed, robustness, and low-power consumption similar to the charge-based solid-state memory devices. Schematic of the magnetic domain-wall racetrack memory. (A) Vertical-configuration racetrack offers the highest storage density by storing the pattern in a U-shaped nanowire normal to the plane of the substrate. The two cartoons show the magnetic patterns in the racetrack before and after the DWs have moved down one branch of the U, past the read and write elements, and then up the other branch. (B) Horizontal configuration uses a nanowire parallel to the plane of the substrate. (C) Reading data from the stored pattern is done by measuring the tunnel magnetoresistance of a magnetic tunnel junction element connected to the racetrack. (D) Writing data are accomplished, for example, by the fringing fields of a DW moved in a second ferromagnetic nanowire oriented at right angles to the storage nanowire. (E) Arrays of racetracks are built on a chip to enable high-density storage. Adapted with permission from (154). While the writing using Oersted fields, STT switching or using fringe fields of domain walls and reading using GMR or TMR is relatively established, a key challenge is the motion of all domains and domain walls synchronously in the racetrack (158). This was initially achieved using STT in in-plane magnetized systems (154, 158–160). Due to the limited efficiency of STT, the threshold current densities are high and the velocities are limited (∼10 to 100 m/s for ∼1012 A/m2), and the widths of DWs in in-plane magnetized thin film wires are relatively large, which is on the order of the width of the nanowires, i.e., typically hundreds of nanometers (156). In these DWs, their large width limits the maximum information storage density. To overcome both problems, materials with perpendicular magnetic anisotropy resulting in narrower domain (few nanometers) walls have been proposed combined with switching by SOTs that can be orders of magnitude more efficient than STT (18). Domain wall motion in the out-of-plane magnetized thin films have been extensively studied, for example, in ferromagnetic multilayers, i.e., ferromagnet/heavy metal. These multilayers can exhibit a strong perpendicular magnetic anisotropy, the Dzyaloshinskii-Moriya interaction leads to domain walls with a fixed chirality (161, 162), and the heavy metal leads to large spin current generation resulting in fast domain walls (163, 164). Beyond single magnetic layers, synthetic antiferromagnets were used to increase the domain wall velocity even further (163), and more recent developments include the use of ferrimagnets, chiral structures, and other special geometries that allow for high precision positioning of domain walls (165). Using synthetic antiferromagnetic coupling can also be used for electrical read out with a tunnel junction (166). Since domain walls by virtue of their topology have to touch the edges of the wire, they are susceptible to edge roughness potentially limiting reproducibility of the motion. To overcome this problem, topologically nontrival skyrmion spin structures (162, 167) have been proposed for the racetrack memory where the information is encoded in the position of skyrmions (168). These spin structures entail a topologically enhanced stability and can be moved reproducibly (169) and in synthetic antiferromagnets also very rapidly (170). Given their ability to move around corners, this might enable also more complex device geometries and integration with logic (171). While many further developments have occurred that are beyond the scope of this short review [e.g., domain wall logic (172) and see also for instance (173)], it should be pointed out that a key challenge for the future is the realization of a fully 3D racetrack with first promising steps having been recently achieved (174). Spin-based logic The extension of spintronics from memory toward logic functionalities has long been motivated by the prospect of combining nonvolatility with reduced energy dissipation (175, 176). Early spin-based logic proposals relied primarily on current-driven manipulation of magnetic states, often encountering limitations associated with the absence of intrinsic gain, signal restoration, and efficient concatenation. As a result, spin-based logic has developed along two complementary directions: near-term hybrid CMOS/MTJ logic-in-memory schemes, which build on the maturity of STT- and SOT-MRAM technologies, and more disruptive concepts exploiting electric-field control or spin orbit–based transduction mechanisms (177), which aim to further reduce switching energy and integrate memory and logic more tightly. At present, the most technologically mature route toward spin-based logic relies on hybrid CMOS/MTJ circuits, or nonvolatile logic-in-memory. In these architectures, MTJs provide state retention, reconfigurability, and magnetoresistive readout, while CMOS supplies gain, signal restoration, and peripheral control (178). Although this approach gives up a purely spin-based signal path, it benefits from the maturity of STT- and SOT-MRAM technologies and offers a nearer-term implementation route. More ambitious all-spin logic concepts instead aim to keep information in the spin/magnetic domain throughout the logic chain but still face major challenges in gain, directional signal flow, and efficient spin transport (179, 180). Magnetoelectric spin-orbit (MESO) logic represents a more disruptive route toward spin-based logic (Fig. 9A) (128). In MESO architectures, information is encoded in the magnetic state of a nanomagnet, which serves as a nonvolatile state variable. Writing is achieved via electric field–driven magnetoelectric switching of the magnetization, while readout relies on spin orbit–mediated conversion of the magnetic state into an electrical signal. This decoupling of write and read operations enables voltage-driven magnetic control combined with efficient electrical transduction, without requiring large charge currents through the magnetic layer. As a result, MESO provides a physically appealing route toward low-power, scalable spin-based logic. Spin-based logic concepts. (A) MESO device concept. An input charge current polarizes the magnetoelectric, which switches the magnetization of an adjacent ferromagnet. The resulting magnetization state is read out through spin-charge conversion in the spin-orbit layer to generate an output voltage whose sign is set by the ferroelectric polarization direction. (B) FESO device concept. A (nonmagnetic) ferroelectric layer controls the Rashba state in the spin-orbit layer, changing the sign of the inverse Edelstein effect. Sending a spin current from the ferromagnet generates an output voltage whose sign is controlled by the ferroelectric state. Both MESO and FESO devices can be cascaded to build logic gates, enabling computing at low-power (175). FESO removes the need to switch magnetization and thus brings faster operation and immunity to magnetic cross-talk. (C) Unit circuit of a probabilistic bit (p-bit) with a superparamagnetic tunnel junction as well as time-domain and time-averaged output voltage VOUT as a function of time and input voltage VIN, respectively. Ferroelectric spin-orbit (FESO) logic shifts the logic state variable from magnetization to ferroelectric polarization (Fig. 9B) (129). In this approach, a magnetic layer provides a fixed symmetry-breaking field, while the ferroelectric polarization directly modulates interfacial SOC and the associated spin-charge conversion processes. This functionality has been demonstrated in several systems (129, 181, 182). Logic states are encoded in the direction of ferroelectric polarization and written purely by electric fields, eliminating the need for magnetization switching during logic operation. An important advantage of this scheme is that it does not rely on magnetoelectric coupling or multiferroic materials, whose coupling strength, reproducibility, and endurance often remain limiting for logic applications. As a result, FESO architectures enable a broader selection of ferroelectric materials optimized for switching speed, scalability, and reliability. Readout is achieved via polarization-controlled spin orbit–coupled transport, enabling direct electrical detection of the ferroelectric state and naturally supporting multilevel and analog functionality. Despite their promise, both MESO and FESO logic concepts face significant challenges related to device variability, signal robustness, and integration within CMOS back-end processes. Quantitative benchmarking against advanced CMOS technologies, including realistic system-level overheads, remains essential. Nonetheless, continued progress in materials growth (130), interface engineering, and spin-orbit physics sustains strong interest in these approaches as long-term complements to charge-based logic technologies. Unconventional computing In addition to the properties discussed above, spintronic systems can be deliberately designed to exhibit functionalities such as analog memory, nonlinearity, short-term memory, stochasticity, and wave-based dynamics. In conventional electronics, these characteristics are typically regarded as undesirable, as they deviate from the deterministic, sequential, and binary operation paradigm that underpins CMOS-based digital logic. However, it is increasingly recognized that general-purpose hardware struggles with certain classes of complex problems, including combinatorial optimization, probabilistic inference, and quantum simulation. This has stimulated growing interest in domain-specific unconventional computing architectures, in which the aforementioned, unexplored spintronic properties become advantageous rather than detrimental. Significant progress has been made toward unconventional computing platforms based on spintronic devices (105). For instance, analog memory elements (183) have been explored as artificial synapses for artificial neural networks (184). Thermally activated magnetization dynamics and plasticity have been used to realize artificial neurons and synapses for spiking neural networks (185). Nonlinear and transient magnetization dynamics, STOs, and spin waves have been harnessed for reservoir computing (186–191). Stochastic switching or superparamagnetic behavior in spintronic systems has enabled probabilistic/stochastic computing (184, 192–195). The wave nature of spin waves has been exploited in coherent Ising machines for solving optimization problems (196). As a representative example, probabilistic computing using stochastic MTJs (Fig. 9C) has demonstrated proof-of-concept implementations for combinatorial optimization (184, 195), machine learning (194), quantum-inspired simulation (197), and Bayesian inference (198). Projections suggest improvements of more than two orders of magnitude in energy efficiency, three orders in area efficiency, and four orders in computational speed compared with conventional hardware to execute probabilistic algorithms (199). Parallel advances in MTJ materials and device engineering are supporting these developments (200–202). Unconventional computing represents one of the most promising domains for the future applied spintronics. Nevertheless, substantial challenges remain before large-scale practical deployment becomes feasible (203), necessitating continued interdisciplinary research. For instance, in analog-based artificial or spiking neural networks, stably storing multiple states with scaled-down device dimensions is a primary challenge, as these multistates generally rely on domain wall positions or multidomain structures. Regarding reservoir computing, critical issues include ensuring reproducibility, tuning the timescale of internal dynamics, improving scalability, and mitigating the overhead of input/output circuits. Concerning probabilistic computing, enhancing stochastic fluctuations while suppressing device-to-device variability remains an ongoing task. Furthermore, for all these paradigms, monolithic integration into existing CMOS processes and a holistic device-circuit-algorithm codesign approach are essential requirements. Addressing these challenges will not only push the boundaries of energy-efficient AI hardware but also pave the way for a new era of brain-inspired intelligent computing systems (204). EMERGING DIRECTIONS Spintronics is constantly evolving with new research directions constantly developing. These directions include topological insulator/superconductor heterostructure systems that can entail time-reversal symmetry breaking or special magnetic order such as spin liquids and many more. A special mention goes to 2D van der Waals materials systems where twisting multilayers can lead to novel magnetic phenomena including unconventional magnetism. Large magnetoresistance effects have been observed in MTJs and spin-filter tunneling devices. Spin transport has been studied in graphene and transition metal dichalcogenides and materials with valley-spin coupling. Interfaces are particularly exciting as they can be atomically flat over large areas with arbitrary rotation angles leading to complex coupling. While scientifically very active and conceptually rich, the low copuling across the van der Waals gap leads mostly to comparatively low ordering temperature. Environmental stability, contact resistance, sample variability, scalable growth, and reproducible interfaces all pose challenges for integration with microelectronics. For this review, we have chosen to present three topics in more detail that we think warrant particular attention including antiferromagnetism and altermagnetism, magnonics, and orbitronics. Last, we note that spin also has an important role to play in emerging quantum sciences and technologies with conspicuous examples being single spin qubits, which are however beyond the scope of this review. Spintronics with antiferromagnets and altermagnets Antiferromagnets are compensated magnetically ordered materials, whose compensation is directed by symmetry (Fig. 10). In collinear ordered systems, this symmetry is the combined symmetry of time reversal and parity/translation, making these systems spin degenerate, having an effective time reversal symmetry in momentum space. Any spin-dependent transport in these systems must therefore be of relativistic SOC origin. Illustration of the research landscape of unconventional magnetism. Altermagnets and coplanar p-wave magnets are the primary examples exhibiting combined properties of ferromagnets, antiferromagnets, and magnetic counterparts of weak relativistic effects that operate at THz dynamics. Their connection to an incomplete list of multiple subfields of physics. On the other hand, their compensating symmetry renders them immune to external magnetic fields and makes their dynamics exchange enhanced, enabling ultrafast dynamics in the terahertz (THz) range, low stray fields, and high thermal stability—making them ideal for dense, fast, and energy-efficient spintronic devices. The challenge is on how to electrically manipulate their order parameter to make them active components of possible memory devices. This is the field of antiferromagnetic spintronics (205). However, one of the key challenges in the application of antiferromagnets is the reliable electrical readout and manipulation of the Néel vector. One possible solution is to use materials with special symmetry (CuMnAs and Mn2Au) that allow staggered spin polarization of the electrical current and efficient Néel SOTs (109, 110). Another approach is based on the direct and inverse SHEs. In this case, magnetic dynamics are excited due to interfacial spin torques and spin accumulation at the interface between the antiferromagnet and the heavy metal electrode. Reading of the transferred spin is then realized by measuring SMR or TMR. The magnetic ordering in noncollinear antiferromagnets can also be probed using the AHE. Effective manipulation and switching of antiferromagnetic order can also be achieved in hybrid structures consisting of antiferromagnetic and thin ferromagnetic layers. Because of the strong interfacial exchange coupling, the ferromagnetic layer maps the underlying antiferromagnetic structure and can produce strong torques. An alternative to electrical manipulation is optical excitation, which can also be used to induce antiferromagnetic dynamics and switching between different states. The mechanisms behind switching involve magneto-optical effects (206–208), laser-induced heating (209) and the associated spin Seebeck effect, or the thermo-magnetoelastic effect, which can also be induced electrically (210–212). Altermagnets are a recently found new symmetry class of collinear compensated magnetically ordered materials (10). Despite their zero net magnetization, they break time reversal symmetry in an unconventional way, exhibiting unconventional spin-polarized d/g/i-wave band structure in reciprocal space connected to their lattice symmetries. Hence, these materials have the possibility of combining THz speeds and strong polarized currents, whose polarization is independent of the current (unlike antiferromagnets), leading to the possibility of teramagnetic memory technologies (1 Tb). These new unconventional magnets beyond s-wave order allow to expand the known ferromagnets and antiferromagnets paradigms to directly allow for teramagnetic devices based on strong signals due to their unconventional spin polarization in momentum space, which is driven primarily by exchange interactions. Their novel properties, the very large variety of material candidates reaching currently more than 400 and the many connections to other areas of condensed matter physics (topological quantum matter, spintronics, optics, neuromorphic computing, multiferroics, strongly correlated systems, twistronics, etc.), make this area one of the most active in recent years within the broad context of spintronics (213, 214). Magnonics Following the natural desire for high-performance, miniaturized devices for signal processing and computing, the most immediate perspective for magnonics is the race toward improved speed and scale (Fig. 11). Magnon transport. (A) Nonlocal spin transport by incoherent magnons. A dc current is used to generate a spin accumulation beneath the left heavy metal electrode by the SHE. The spin accumulation diffuses to the right heavy metal electrode where it is detected by the dc voltage arising due to the inverse SHE. (B) Spin-wave transport by coherent magnons. The Oersted field from an alternating microwave-frequency current launches coherent spin waves at the left transducer. The spin wave propagates to the right transducer where a microwave voltage is generated due to Faraday’s law of induction. While excitations in ferromagnets that have been predominantly used in magnonics so far are in the GHz range, the natural magnon modes in magnetization-compensated antiferromagnets and altermagnets with Néel order are in the THz range. Because the frequency of the precessional dynamics sets the timescale for coherent spin wave control, it is desirable to explore magnonics in Néel-ordered materials. This endeavor is challenging due to the limited availability of coherent THz sources and detectors and the small net magnetic moment of magnons in fully compensated magnets. Despite these challenges, spin pumping from antiferromagnetic modes at frequencies exceeding 200 GHz has been experimentally observed using magnetic resonance spectroscopy (215, 216). Alternatively to direct microwave excitation, coherent THz magnons can also be generated by ultrafast optical excitation (212). As traditional magnonic devices use quasi-2D magnon transport in thin magnetic films, there are two potential pathways to improve the magnonic circuit density: first, going to 2D materials with emerging topological band structures (217), and second, going to true 3D magnonics (218), where for example spin chirality can be designed by geometry (219). A further transformative vision for magnonics is the exploration of quantum effects in the few excitation limit. A key requirement to observe quantum effects is the strong coupling of magnons to suitable detection circuits. Strong coherent magnon-photon (220) and magnon-phonon coupling (221) in planar resonator structures has been established in the low-damping ferrimagnet YIG (see the “Spin waves” section). By using the cavity-mediated strong coherent coupling of magnons and photons in a qubit, single-magnons can be detected (222). Magnonic systems are conducive to canonical quantum effects such as squeezing (223, 224). Squeezing in a YIG sphere coupled to a qubit has very recently been experimentally demonstrated (225). Quantum magnonics requires the precise determination of single magnon states (226). While classical many-magnon state tomography has been experimentally realized (227), quantum state tomography of individual magnons is an outstanding experimental challenge (228). Orbitronics Conventional spintronics has primarily focused on the detection and manipulation of the electron’s spin angular momentum. However, electrons also possess orbital angular momentum, which provides an additional internal degree of freedom. Recently, transport and interconversion phenomena associated with the orbital angular momentum have attracted significant attention, giving rise to an emerging research paradigm referred to as orbitronics (76, 229). Representative orbitronic phenomena include the orbital Hall effect (230–232) and the orbital Rashba-Edelstein effect (233–235), which are regarded as orbital counterparts to the SHE and the Rashba-Edelstein effect, respectively. In addition, experimental reports have described orbital pumping analogous to spin pumping (236), as well as an orbital Seebeck effect corresponding to the spin Seebeck effect (237). These observations suggest that orbital angular momentum can be generated, transported, and detected in solid-state systems in ways similar to spin currents, yet with distinct underlying mechanisms. Furthermore, beyond dipolar orbital angular momentum, also orbital multipoles can be generated by currents, which is a feature that sets orbitronic effects apart from spin. Concerning the material, orbitronics effects have been observed in a variety of systems, including 3d, 4d, and 5d transition metals and their oxide such as Ti, Cr, Mn, CuOx, Ru, and Ta (238–242). In addition, several studies pointed out that conversion from orbital angular momentum to spin angular momentum is important to obtain a large current-induced torque. Notably, the materials platforms in which orbitronic effects are prominent (238, 241, 243) and the characteristic length scales over which orbital currents propagate (244–246) have been reported to differ substantially from those of spin-based phenomena. These differences raise the possibility of functionalities that surpass spin-mediated effects or enable applications that are otherwise difficult to realize. At the same time, several theoretical aspects remain controversial, including the relationship between spin-orbit interaction and orbitronic responses (247) and the diffusion length of orbital currents (248, 249). Continued theoretical and experimental efforts are therefore essential to establish orbitronics on firm physical foundations toward device applications (76, 250). CONCLUSION Spintronics has progressed from a field centered on spin-dependent transport in magnetic multilayers to a broad research area that integrates spin, charge, orbital, and collective excitations within solid-state systems. This evolution has been driven by advances in materials engineering, the recognition of SOC as a central mechanism for spin-charge interconversion, and the development of electrical and voltage-based control schemes for magnetic order. As reviewed here, these advances have enabled a diverse range of functionalities, including electrical readout of magnetic states through magnetoresistive effects, current-driven manipulation of magnetization via spin-transfer and SOTs, and low-power control through electric-field effects and magnetoelectric coupling. These physical mechanisms underpin technologies that are already widely deployed, such as magnetoresistive sensors and MRAM, while also motivating new device paradigms including racetrack memories, spin-based logic architectures, and unconventional computing platforms. At the same time, emerging directions such as antiferromagnetic and altermagnetic spintronics, magnonics, and orbitronics are expanding the scope of the field by introducing new degrees of freedom and operating regimes. The continued convergence of materials science, condensed matter physics, and device engineering will be essential for translating these advances into scalable technologies. In this context, spintronics provides a versatile platform for exploring energy-efficient information processing and for extending the functionality of future electronic systems.