Spintronics is the branch of electronics that uses spin angular momentum and its associated magnetic properties, alongside or instead of electric charge, to control the behavior of materials and devices. Conventional electronics primarily represents information through voltages, currents, and stored charge. Spintronic devices additionally exploit spin polarization, the orientation of magnetic layers, or spatial patterns of magnetic order. The field connects condensed-matter physics, materials science, and electronic engineering.
Its principal established applications include magnetic read heads, magnetic-field sensors, and nonvolatile magnetic memory. These technologies depend on the ability to convert a magnetic configuration into an electrical signal and, increasingly, to change that configuration electrically. Research also investigates spin currents, magnetic textures, spin waves, and ultrafast magnetic dynamics as possible foundations for information processing. Although spin-based devices can retain information without continuous power, their operation is not intrinsically free of dissipation: generating, transmitting, and detecting spin signals all impose physical and engineering costs.
Spin, magnetic order, and electronic states
The electron has an intrinsic angular momentum with spin quantum number . When its spin is measured along a chosen axis, the allowed projections are and , where is the reduced Planck constant. These two outcomes are commonly called spin up and spin down. They are not evidence that the electron is a small rotating sphere; spin is an intrinsically quantum property. Its magnetic moment couples to a magnetic field, producing a difference in energy between suitable spin states.
In solids, the usefulness of spin depends strongly on the collective behavior of electrons. In a ferromagnet, the exchange interaction, arising from quantum statistics and electrostatic interactions, favors an ordered arrangement with a spontaneous magnetization. Iron, cobalt, nickel, and many of their alloys are important examples. Exchange can produce different electronic states for opposite spin projections, giving the two spin populations unequal densities of states, velocities, or scattering rates. Consequently, an electric current passing through a ferromagnet can become spin polarized.
A simple measure of current polarization is
where and are spin-resolved current components along a specified axis. Polarization is not necessarily a single intrinsic material constant. A value inferred from tunneling may differ from one inferred from diffusive conduction because the experiments weight electronic states and interfaces differently. Moreover, a spin-polarized current need not imply a large nonequilibrium bulk magnetization.
Spin–orbit coupling links an electron’s spin to its orbital motion and surrounding electric potential. It enables electrical manipulation of spin and conversion between charge and spin currents, but it also provides channels for spin relaxation. Magnetic anisotropy makes particular magnetization directions energetically favorable. It can originate from spin–orbit coupling and crystal symmetry, device shape, or interfaces. Anisotropy supplies the energy barriers that stabilize magnetic memory states.
Spintronic information need not be encoded in an individual electron. Most commercial devices use the collective magnetization of a nanoscale magnetic element, involving many coupled atomic moments. Such a magnetic bit is generally a classical two-state system, not a quantum bit. Coherent electron-spin control belongs to an overlapping area of quantum technology, but ordinary magnetic memory does not require the preservation of a quantum superposition.
Historical development and the emergence of devices
The foundations of spintronics predate the name. In 1857, William Thomson, later Lord Kelvin, reported that the electrical resistance of ferromagnetic metals depends on the direction of magnetization relative to current. This anisotropic magnetoresistance reflects spin–orbit-dependent electronic transport. It eventually became useful in magnetic recording and sensing, although its resistance changes are generally smaller than those obtainable in later multilayer devices.
In 1936, Nevill Mott developed a two-current picture of transport in ferromagnetic metals. In its simplest form, electrons with opposite spin projections conduct through approximately separate channels with different resistivities. This approach established a conceptual basis for understanding how magnetic order affects electrical conduction. It is most directly applicable when spin-flip processes do not rapidly mix the channels over the relevant device dimensions.
A second important line of development concerned quantum tunneling between magnetic electrodes. Michel Jullière reported spin-dependent tunneling in 1975 and introduced a model relating the resistance contrast to electrode spin polarizations. Early junctions did not immediately provide a practical room-temperature technology. Reliable fabrication of ultrathin insulating barriers, control of interfaces, and sufficiently large signals remained major obstacles.
The decisive development for metallic multilayers was the independent discovery of giant magnetoresistance in 1988 by groups led by Albert Fert and Peter Grünberg. They found that the resistance of layered magnetic structures changed strongly when the relative orientations of magnetic layers changed. Fert’s group investigated iron/chromium multilayers, while Grünberg’s group studied related layered structures. Their work showed that nanoscale engineering could produce magnetoresistance far exceeding the conventional anisotropic effect. Fert and Grünberg received the 2007 Nobel Prize in Physics for the discovery.
The subsequent spin valve provided a device architecture with one relatively fixed magnetic layer and another that could rotate in a weak applied field. IBM introduced giant-magnetoresistive hard-disk read heads in 1997. The electrically detected resistance change enabled sensitive reading of small magnetic recording regions and contributed to the development of higher-density storage.
During the 1990s and 2000s, room-temperature tunneling devices, current-driven magnetization switching, and improved oxide barriers broadened the field. John Slonczewski and Luc Berger independently developed influential theories of current-induced magnetic torque in 1996. In 2004, separate experimental teams reported large room-temperature tunneling magnetoresistance using crystalline magnesium oxide barriers. These advances shifted attention from reading magnetization electrically toward writing it electrically and integrating magnetic elements with semiconductor circuits.
Spin injection, diffusion, and relaxation
A spintronic device must usually create a nonequilibrium spin population, transport it, and detect its effect. Spin injection can occur when a charge current crosses from a ferromagnet into a nonmagnetic conductor. Because the ferromagnet transmits the two spin components differently, the receiving material develops spin accumulation: an imbalance represented by different electrochemical potentials for opposite spin projections. Optical excitation and spin–orbit-mediated conversion provide other ways to generate spin populations.
In a simple diffusive description, the spin accumulation obeys
where is a diffusion coefficient and is a spin-relaxation time. In a homogeneous one-dimensional channel at steady state, spin accumulation decays over the characteristic distance
This spin-diffusion length is a central design parameter. It is not universal for a named material: temperature, impurity concentration, structural disorder, carrier density, and interfaces can all alter it. For example, spin transport through copper over submicrometer distances is accessible in metallic nanostructures, while appropriate semiconductor and low-dimensional systems can support substantially longer transport distances under suitable conditions.
Spin relaxation has several microscopic mechanisms. In Elliott–Yafet relaxation, spin–orbit coupling mixes spin character in electronic states, allowing momentum-scattering events to change spin. In the D’yakonov–Perel’ mechanism, electrons in crystals lacking inversion symmetry precess around momentum-dependent effective fields; scattering changes those fields and modifies the accumulated spin rotation. Hyperfine coupling to nuclear spins can be important in localized semiconductor systems. Which process dominates depends on material symmetry, disorder, localization, and temperature.
Efficient injection across dissimilar materials is often difficult. A transparent contact between a metallic ferromagnet and a much less conductive semiconductor can produce very weak spin injection because of conductivity mismatch. A suitably resistive, spin-selective tunnel barrier can alleviate this problem by changing the balance of interfacial and bulk spin resistances. However, a barrier that is too resistive can reduce usable current and increase voltage requirements.
A pure spin current carries angular momentum without a net charge current. In a simplified two-channel picture, opposite spin populations move in opposite directions so that their charge flows cancel while their spin flows add. The concept requires care in materials with strong spin–orbit coupling, where electron spin alone is not conserved and angular momentum can pass to orbital motion and the lattice. Nevertheless, spin-current descriptions provide a powerful framework for interpreting transport, torque, and spin-to-charge conversion.
Magnetoresistance and magnetic tunnel junctions
Magnetoresistance allows a magnetic configuration to be read electrically. In a basic giant-magnetoresistive structure, two ferromagnetic layers are separated by a nonmagnetic metal. When the layer magnetizations are parallel, one spin channel can encounter relatively weak scattering through both layers. When they are antiparallel, each channel encounters stronger scattering in at least one layer. The resistance is therefore commonly lower in the parallel configuration, although the detailed sign and magnitude depend on material combinations and transport conditions.
Giant magnetoresistance can be measured with current flowing in the plane of the layers or perpendicular to them. The distinction changes the role of interfaces, current paths, and spin-diffusion lengths. In current-perpendicular-to-plane structures, electrons must cross the interfaces sequentially, making interfacial spin-dependent scattering especially important. Practical spin valves stabilize the reference layer through magnetic coupling or exchange bias, allowing the other layer to respond to an external field.
Tunnel magnetoresistance uses an insulating rather than metallic spacer. A magnetic tunnel junction normally consists of two ferromagnetic electrodes separated by an insulating barrier only a few atomic layers thick. Its resistance depends on the relative magnetization directions because tunneling probabilities depend on the available spin-dependent electronic states and the properties of the barrier.
The resistance contrast is often defined as
where and are the antiparallel and parallel resistances. In the simplified Jullière model,
with and effective tunneling polarizations. This expression is useful for intuition but neglects many features of real junctions, including coherent wave propagation, electronic symmetry, defects, and bias-dependent effects.
Crystalline magnesium oxide barriers illustrate why the barrier cannot be treated as a passive separator. In appropriately oriented iron-based electrodes and MgO, the decay rates of tunneling states depend strongly on their symmetry. The barrier preferentially transmits particular electronic states, producing a large difference between parallel and antiparallel conductance. Coherent tunneling therefore offers much larger room-temperature resistance contrasts than many earlier amorphous-barrier devices.
Device optimization requires balancing several quantities. A thin barrier reduces resistance and can increase current, but it becomes more vulnerable to defects and dielectric breakdown. A thicker barrier can improve isolation while raising operating voltage. The resistance–area product, magnetoresistance ratio, junction uniformity, noise, and endurance must all be considered together. A high magnetoresistance measured in a large laboratory junction does not by itself establish suitability for a dense, low-voltage memory array.
Electrical control of magnetization
An electric current can change magnetization because electrons carry angular momentum as well as charge. In spin-transfer torque, a spin-polarized current enters a magnetic region whose magnetization is not aligned with the incoming spin polarization. The transverse spin component is absorbed, transferring angular momentum to the magnet. Depending on current direction and amplitude, this torque can reverse a magnetic bit, move a magnetic domain wall, or sustain magnetic precession.
A common dynamical description is the Landau–Lifshitz–Gilbert equation with additional torque terms:
Here is the unit magnetization vector, is the positive magnitude of the gyromagnetic ratio in this convention, is the effective magnetic induction, and is the Gilbert damping parameter. The first term describes precession; damping drives the magnetization toward a lower-energy configuration. Current-induced torques can oppose damping or push the magnetization across an anisotropy barrier.
A different route uses spin–orbit torque. In a magnetic layer adjacent to a material with strong spin–orbit coupling, an in-plane charge current can generate a spin accumulation or a transverse spin current that exerts torque on the magnet. The spin Hall effect is an important source: carriers with different spin orientations are deflected differently, producing transverse spin transport. Platinum, tantalum, and tungsten are widely studied spin-current source materials. Interfacial spin–orbit effects can also contribute.
Spin-transfer and spin–orbit devices have different geometrical trade-offs. In a conventional two-terminal spin-transfer memory cell, the write current passes through the tunnel barrier used for reading. This arrangement is compact but exposes the barrier to write stress and couples read and write requirements. A common spin–orbit design uses a separate lateral write channel and a vertical read path. This separation permits independent optimization but generally requires more area and additional interconnects.
Other control methods modify the magnetic energy landscape rather than supplying torque in the same way. Voltage-controlled magnetic anisotropy changes interfacial anisotropy through an applied electric field. Strain coupled through a piezoelectric material can alter anisotropy through magnetoelastic effects. These approaches may assist switching or reduce required current, but their complete energy cost includes voltage generation, charging, timing, and readout.
Switching is inherently dynamical and often stochastic. Thermal fluctuations affect the initial magnetization and the probability of crossing an energy barrier during a pulse. Consequently, a quoted switching current is meaningful only alongside pulse duration, error probability, temperature, and device geometry. Subnanosecond demonstrations and low-current long-pulse switching describe different operating regimes rather than interchangeable performance measures.
Materials, interfaces, and fabrication
Spintronic performance is often determined as much by interfaces as by bulk materials. An interface can filter spins, generate anisotropy, absorb angular momentum, or introduce spin-flip scattering. Atomic intermixing, roughness, oxidation, and crystallographic disorder can therefore change both transport and magnetic dynamics. Thin-film growth and thermal processing are integral parts of the device physics rather than merely manufacturing details.
Cobalt–iron–boron alloys combined with magnesium oxide are important in magnetic tunnel junctions. Annealing can promote crystallization of the magnetic electrode near the barrier and improve tunneling properties. The redistribution of boron and the choice of neighboring layers affect this process. Suitable interfaces can also generate perpendicular magnetic anisotropy, allowing magnetization to point normal to the film. Perpendicular geometries are useful for small memory cells because they can provide favorable stability and switching characteristics.
Reference layers often contain several coupled magnetic films. In a synthetic antiferromagnet, ferromagnetic layers are coupled antiparallel through an appropriate nonmagnetic spacer. Their moments partly cancel, reducing stray fields acting on the storage layer while maintaining a stable reference direction. This is distinct from a material with atomic-scale antiferromagnetic order, in which neighboring magnetic sublattices compensate internally. Both forms of compensation are useful, but their structures and dynamics differ.
Semiconductors offer electrical gating, optical access, and compatibility with established electronic architectures. Silicon is attractive because of its relatively weak spin–orbit coupling and mature fabrication infrastructure, but spin injection and detection require carefully engineered contacts. III–V semiconductors provide strong optical selection rules and tunable spin–orbit interactions. Dilute magnetic semiconductors, such as manganese-doped gallium arsenide, have illuminated interactions between carriers and magnetic order, although their operating-temperature limitations restrict many device applications.
More specialized candidates include highly spin-polarized Heusler alloys, magnetic oxides, two-dimensional materials, and topological insulators. A half-metal is an idealized material with metallic states for one spin channel and a gap for the other at the Fermi level. Such a structure suggests complete spin polarization, but real interfaces, finite temperature, and disorder can substantially reduce the polarization observed in devices. Similarly, unusual spin textures in topological materials do not automatically translate into efficient, manufacturable switching elements.
Fabrication introduces constraints beyond obtaining desirable material parameters. Patterning can damage sidewalls, redeposit conducting material across barriers, and alter magnetic anisotropy. Annealing must remain compatible with surrounding transistors and interconnects. A device stack that performs well on a small research substrate may require substantial modification before it can tolerate wafer-scale deposition, etching, thermal cycling, and statistical process variation.
Magnetic memory, read heads, and sensors
Magnetic random-access memory, or MRAM, stores information in magnetic configurations rather than accumulated charge. A common cell contains a magnetic tunnel junction whose free layer can point parallel or antiparallel to a reference layer. The resistance distinguishes the two states, while a transistor selects the cell. Because the magnetic state persists without continuous power, MRAM is nonvolatile; this does not mean that reading, writing, or peripheral circuitry consumes no energy.
Early MRAM designs used magnetic fields generated by nearby current lines. Spin-transfer-torque MRAM instead writes through angular-momentum transfer from current passing through the junction. This approach improves the scalability of selecting and switching individual cells and has become an established commercial memory technology. Spin–orbit-torque MRAM is investigated for fast writing and separation of the read and write paths, with cell area and integration complexity among its major trade-offs.
Retention depends strongly on the magnetic energy barrier. A simplified thermal-activation model gives
where is a characteristic retention time, an attempt time, the reversal barrier, the Boltzmann constant, and the absolute temperature. For a simple uniformly magnetized element with uniaxial anisotropy, the barrier is approximately . Real patterned elements may reverse nonuniformly, so this expression is not universally adequate. Shrinking the magnetic volume tends to threaten retention unless anisotropy or another stabilizing mechanism compensates.
Memory design therefore balances retention, writing energy, speed, endurance, and read margin. Increasing the energy barrier can make a bit more stable but harder to switch. Reading must produce a distinguishable signal without causing read disturbance. Large arrays also require exceptionally low error probabilities, control of device-to-device variation, and suitable error-correction strategies. System-level benefits depend on access circuitry and data movement, not simply the switching energy of one isolated junction.
Magnetic read heads were the first major large-scale application of modern spintronics. They convert the stray fields from recorded magnetic regions into resistance changes. Spin valves and later tunnel-magnetoresistive sensors provided high sensitivity in progressively smaller sensing elements. Shielding, magnetic stability, linearity, and noise are as important as the maximum resistance contrast.
Related sensing elements measure position, rotational angle, electrical current, and weak magnetic fields. Current sensors infer current from its surrounding magnetic field, enabling galvanic separation between the measured conductor and sensing circuit. Automotive and industrial systems use magnetoresistive devices for motion and position sensing. Their engineering requirements differ from memory: a sensor may need a continuous, nearly linear response, whereas a memory cell requires two sharply distinguishable and stable states.
Collective excitations and emerging computing architectures
Spintronics extends beyond uniformly magnetized bits. A domain wall separates regions with different magnetization directions. In racetrack-memory proposals, sequences of domains encode data along a narrow track, and electrical pulses move the walls past fixed reading and writing locations. The potential advantage is dense storage with relatively few access devices. Precise motion remains difficult because defects can pin walls, neighboring walls interact, and pulse-driven displacement can vary from one event to another.
A magnetic skyrmion is a localized, spatially twisting magnetization texture. Skyrmions can be stabilized by combinations of exchange, anisotropy, magnetic fields, dipolar interactions, and chiral interactions associated with spin–orbit coupling. Their compact size and current-driven motion make them candidates for information carriers. However, topological classification does not guarantee indefinite stability in a finite device: skyrmions can annihilate at edges or through thermally activated and current-driven processes. Their transverse motion can also cause unwanted interaction with track boundaries.
Another branch, magnonics, uses magnons, the quantized excitations of spin waves. Spin waves can propagate through magnetic insulators without a conducting charge current traversing the magnetic medium. Yttrium iron garnet is widely studied because of its low magnetic damping. Interference, phase, and frequency provide possible ways to encode information. Practical challenges include efficient conversion between electrical signals and spin waves, miniaturization, attenuation, and interactions among modes.
Antiferromagnets offer compensated magnetic order, small external stray fields, and potentially very fast intrinsic dynamics. Their state can be described using an order parameter associated with the difference between sublattice magnetizations. Electrical manipulation has been demonstrated in selected materials, including copper manganese arsenide and manganese–gold compounds. Reading and controlling compensated order generally requires mechanisms different from those used for a simple ferromagnetic moment. Ferrimagnets, with oppositely oriented but unequal sublattice moments, provide another route to rapid dynamics and reduced net magnetization.
Spintronic devices are also studied for unconventional computing. Magnetic tunnel junctions can act as tunable resistive elements, oscillators, or stochastic switches. Networks of oscillators may perform synchronization-based tasks, while thermally fluctuating magnetic elements can generate probabilistic outputs. These functions are being explored in relation to artificial neural networks, optimization, and probabilistic computation. A device-level demonstration does not establish an advantage over conventional digital hardware; comparisons must include interconnects, conversion circuits, accuracy, and the computational task.
Thermal gradients add another dimension through spin caloritronics. The spin Seebeck effect concerns thermally generated spin transport in magnetic systems. In common magnetic-insulator/metal experiments, a spin current entering the metal is detected through the inverse spin Hall effect. Careful interpretation is necessary because ordinary thermoelectric voltages and unintended temperature gradients can coexist with the desired signal.
Measurement, reliability, and physical limits
Spin-dependent effects must be distinguished from ordinary electrical and thermal artifacts. A resistance change under a magnetic field can arise from several mechanisms, including anisotropic magnetoresistance, ordinary magnetoresistance, Hall voltages mixed into the measurement, and genuine spin-valve behavior. Reversing a magnetic electrode or observing a hysteretic voltage is not, by itself, sufficient evidence of spin transport. Device geometry, control samples, field orientation, and quantitative consistency are essential.
Nonlocal measurements help separate spin accumulation from the main charge-current path. In a lateral spin valve, current is injected through one magnetic contact, while a separate detector senses a voltage outside the principal charge-current circuit. Changing the relative injector and detector magnetizations changes the detected signal. Applying a transverse magnetic field causes injected spins to precess and dephase, producing a Hanle response from which transport and relaxation parameters can be inferred under an appropriate model.
Magnetic dynamics are studied using ferromagnetic resonance, time-resolved optical probes, electrical rectification, and microwave emission measurements. In resonance experiments, linewidths provide information about damping and inhomogeneity, but separating intrinsic damping from other broadening mechanisms requires care. Spin pumping—the transfer of angular momentum from a precessing magnet into a neighboring material—can modify damping and generate detectable electrical signals. Extracting conversion efficiencies requires accounting for interfacial transmission, current shunting, and geometry.
Reliability is statistical. A memory technology must operate across distributions of switching thresholds, retention barriers, and tunnel resistances. Rare failures matter when arrays contain millions or billions of cells. Repeated electrical stress can damage a tunnel barrier, while elevated temperature accelerates thermally activated processes. Accelerated tests and extrapolation models are therefore indispensable, but their assumptions must match the physical failure mechanism. A single successful switching trace cannot establish endurance or a low write-error rate.
Energy comparisons require similarly explicit boundaries. A magnetic bit may retain its state without static refresh power, yet writing commonly involves Joule heating in conductors and energy dissipation in access transistors. Readout, clocking, sensing, and interconnect charging can exceed the energy expended inside the magnetic element. Pure spin currents likewise do not imply zero dissipation: the circuitry that generates and detects them generally consumes energy.
Major research questions concern how efficiently angular momentum can be generated, transmitted across interfaces, and absorbed in a useful magnetic mode. They also include the limits of stable magnetic order at small dimensions, the role of nonuniform reversal during fast switching, and reproducibility across complex material stacks. Quantitative comparison increasingly requires reporting temperature, pulse duration, error probability, device dimensions, and full electrical loading together, rather than presenting an isolated headline value for speed or switching current.