Spin-Torque Diode Effect in Spintronics

• Spin-torque diode effect is the conversion of alternating (RF) signals into a direct (DC) output through spin-driven magnetic dynamics in nanostructures like MTJs, vortices, and skyrmions.
• The phenomenon utilizes spin-transfer and spin-orbit torques to induce resonant Lorentzian and anti-Lorentzian responses, enabling precise control over microwave detection and energy efficiency.
• Its applications span high-sensitivity microwave rectification, energy harvesting, and neuromorphic computing, with performance tuned by geometry, magnetic configuration, and bias conditions.

The spin-torque diode effect is the rectification of an alternating electrical drive into a dc output by magnetization dynamics in a spintronic structure. In its standard form, an rf current excites magnetic motion through spin-transfer torque, spin-orbit torque, or related spin-current mechanisms, while the device resistance oscillates through tunnel magnetoresistance or giant magnetoresistance; time-averaging the product of current and resistance yields a finite dc voltage or dc current. Although first formulated most transparently for magnetic tunnel junctions (MTJs) in the linear ferromagnetic-resonance regime, the effect now encompasses vortex oscillators, domain walls, skyrmions, dipolar-coupled bilayers, injection-locked spin-torque oscillators, and magnet$-$normal-metal bilayers driven by the spin-Hall effect (Taniguchi et al., 2013, Chopin et al., 2023, Fang et al., 2024, Gems et al., 10 Sep 2025).

1. Fundamental mechanism and rectification formalism

In MTJs, the canonical definition of the rectified output is the cycle average

$V_{\rm dc}=\frac{1}{T}\int_0^T I(t)R(t)\,dt,$

with the resistance set by the relative orientation of the free-layer and pinned-layer magnetizations. In the tunnel-junction formulation,

$$R = R_{\rm P}+\frac{\Delta R}{2}(1-\mathbf{m}\cdot\mathbf{p}),$$

so an rf current simultaneously drives precession and samples an oscillatory resistance. The resulting dc voltage is therefore a mixing signal rather than a static transport nonlinearity (Taniguchi et al., 2013).

A standard dynamical starting point is the Landau-Lifshitz-Gilbert equation augmented by Slonczewski and field-like torques,

$$\frac{d\mathbf{m}}{dt} = -\gamma \mathbf{m}\times\mathbf{H} -\gamma a_J\,\mathbf{m}\times(\mathbf{p}\times\mathbf{m}) +\gamma b_J\,\mathbf{p}\times\mathbf{m} +\alpha\,\mathbf{m}\times\frac{d\mathbf{m}}{dt},$$

with $a_J$ the damping-like spin torque and $b_J=\beta a_J$ the field-like torque. Linearization about the steady state leads to resonant expressions in which the diode signal contains Lorentzian and anti-Lorentzian components. At resonance, one representative result is

$$V_{\rm dc}(f_{\rm res}) = \frac{\Delta R I_{\rm ac}}{4} \frac{\tilde{a}_{J({\rm ac})}\sin^{2}(\varphi-\varphi_{\rm p})} {\alpha(H_X+H_Y)-2a_{J({\rm dc})}\cos(\varphi-\varphi_{\rm p})},$$

which makes explicit that the signal is controlled by both magnetoresistive modulation and effective linewidth (Taniguchi et al., 2013).

This decomposition also clarifies torque symmetry. In asymmetric-electrode MgO junctions, the in-plane torque mainly generates the Lorentzian contribution, whereas the out-of-plane torque generates the anti-Lorentzian contribution. Because the anti-Lorentzian amplitude scales with the demagnetization field while the Lorentzian amplitude scales with the linewidth, a large field-like torque can dominate the zero-bias spectrum and substantially enhance microwave-detection sensitivity (Matsumoto et al., 2011).

A closely related formulation appears in spin-torque vortex oscillators (STVOs), where the rf current drives vortex-core motion and the moving magnetic texture modulates the tunnel resistance. The rectified voltage is again the time average of a phase-correlated product of current and resistance, but the dynamical variable is the vortex-core orbit rather than a nearly uniform macrospin precession (Chopin et al., 2023).

2. Dynamical descriptions, linewidth, and spectral structure

The simplest theoretical regime is linear-response spin-torque ferromagnetic resonance, where the free layer is treated as a macrospin and the resonance frequency and linewidth are obtained from the linearized LLG equation. In this regime, the diode signal is often fit as the sum of Lorentzian and anti-Lorentzian terms, and the dc spin torque can either narrow or broaden the linewidth depending on the sign of current and the relative magnetization alignment. The central control variable is therefore not only the torque amplitude but also the spin-torque renormalization of dissipation (Taniguchi et al., 2013).

For perpendicular-free-layer MTJs with an in-plane pinned layer, the same logic survives but the geometry changes the equilibrium configuration and the tunable parameter becomes the applied-field direction. The resonance frequency remains Kittel-like, the anti-Lorentzian part vanishes at resonance, and the optimal operating point is encoded in the angle $$\psi=\cos^{-1}(\mathbf{m}^{(0)}\!\cdot\!\mathbf{p})$$. In that geometry the critical current

$$I_c=\frac{2\alpha eMV}{\hbar\eta}\left(\frac{H_X+H_Y}{2}\right)$$

sets the boundary of linear stability and the validity of the perturbative diode formulas (Taniguchi et al., 2013).

Beyond macrospin dynamics, reduced-coordinate descriptions become essential. In STVOs, the vortex core is treated as a quasiparticle at position $\mathbf{X}(t)$ governed by a Thiele equation containing gyrovector, damping, exchange, magnetostatic, Ampère-Oersted-field, and spin-transfer-force terms. Because a purely analytical Thiele-equation approach is only qualitative, the data-driven Thiele equation approach (DD-TEA) incorporates micromagnetic corrections that “absorb the difference” between the analytical model and full micromagnetics, retaining ultra-fast evaluation while recovering quantitative agreement with experiment (Chopin et al., 2023).

Other geometries require explicitly coupled dynamics. In dipolar-coupled Co/Cu/Co nanowire spin valves, two free Co layers are described by coupled LLG equations, and the rectified voltage depends on four angular variables, interlayer dipolar hybridization, and the modulation protocol. In that case the diode spectra reflect two coupled uniform modes rather than an isolated single-layer resonance, and field modulation and laser/temperature modulation generate qualitatively different line symmetries (Ogrodnik et al., 2017).

An even broader generalization appears in normal-metal/ferromagnet bilayers. There, the ac electric field in the normal metal generates a spin current by the spin-Hall effect, coherent magnetization dynamics in the ferromagnet produce a quadratic longitudinal spin-current source, and the inverse spin-Hall effect converts the result back into a dc in-plane charge current. The resonant enhancement is governed by the coherent-mode impedance

$V_{\rm dc}=\frac{1}{T}\int_0^T I(t)R(t)\,dt,$0

whose poles occur at coherent magnetization-mode frequencies (Gems et al., 10 Sep 2025).

3. Material systems, magnetic textures, and active resonant objects

The term “spin-torque diode effect” now covers multiple classes of devices and magnetic eigenmodes. The common ingredient is rectification by spin-driven magnetic motion, but the active resonator may be a quasi-uniform free layer, a vortex core, a domain wall, a skyrmion, or a hybrid bilayer mode (Chopin et al., 2023, Lequeux et al., 2015, Fang et al., 2024, Ogrodnik et al., 2017).

Platform	Active magnetic object	Distinctive observation
MgO MTJ	Free-layer FMR	Alignment and field direction maximize $V_{\rm dc}=\frac{1}{T}\int_0^T I(t)R(t)\,dt,$1
STVO	Vortex-core orbit	Chirality-dependent sign reversal; half-frequency feature
Nanostripe MTJ	Domain wall and edge/domain mode	Effective damping extracted from diode linewidth
Skyrmion-based MTJ	Skyrmion breathing mode	Resonance near 4 GHz; zero-field operation after stabilization
Co/Cu/Co nanowire	Two dipolar-coupled Co modes	Field and temperature modulation give different spectra
N$V_{\rm dc}=\frac{1}{T}\int_0^T I(t)R(t)\,dt,$2F bilayer	Coherent magnetization modes	SHE excitation and ISHE readout yield quadratic dc response

In domain-wall devices, the diode effect is not merely a detector of a resonance frequency. It resolves the motion of a single domain wall and the adjacent magnetic domains separately, and the fitted linewidths permit extraction of effective damping for each localized object. In the reported nanostripe MTJ, the domain-wall-present state exhibits two resonances, whereas after the wall is expelled only the higher-frequency mode remains, identifying the low-frequency mode as the domain-wall vibration and the high-frequency mode as a domain or edge mode (Lequeux et al., 2015).

In skyrmion-based MTJs, the active object is a topological texture rather than a nearly uniform magnetic layer. Quantitative magnetic force microscopy was used to verify a single skyrmion in the free layer, and the diode spectra reveal a skyrmion resonance that is spectrally distinct from the ordinary free-layer and reference-layer uniform modes. Micromagnetic simulations identify this response as a breathing mode of a Néel skyrmion (Fang et al., 2024).

4. Control parameters and optimization strategies

A central result of the MTJ literature is that the diode voltage is highly tunable by magnetic configuration. For in-plane MTJs, the conventional alignment with parallel easy axes is not optimal. The analytical optimum for the pinned-layer direction,

$V_{\rm dc}=\frac{1}{T}\int_0^T I(t)R(t)\,dt,$3

reduces to orthogonality when $V_{\rm dc}=\frac{1}{T}\int_0^T I(t)R(t)\,dt,$4 and shifts away from $V_{\rm dc}=\frac{1}{T}\int_0^T I(t)R(t)\,dt,$5 when finite dc current changes the linewidth through spin-torque antidamping or damping. Using representative parameters, the conventional alignment gives $V_{\rm dc}=\frac{1}{T}\int_0^T I(t)R(t)\,dt,$6 for $V_{\rm dc}=\frac{1}{T}\int_0^T I(t)R(t)\,dt,$7 mA and $V_{\rm dc}=\frac{1}{T}\int_0^T I(t)R(t)\,dt,$8 for $V_{\rm dc}=\frac{1}{T}\int_0^T I(t)R(t)\,dt,$9 mA, whereas the optimized angles yield $$R = R_{\rm P}+\frac{\Delta R}{2}(1-\mathbf{m}\cdot\mathbf{p}),$$0 analytically and about $$R = R_{\rm P}+\frac{\Delta R}{2}(1-\mathbf{m}\cdot\mathbf{p}),$$1 in full LLG simulations (Taniguchi et al., 2013).

In perpendicular-free-layer MTJs, the experimentally tunable variable is often the direction of the applied magnetic field rather than the pinned-layer direction itself. For relatively small applied field, the optimal in-plane projection of the field is parallel or antiparallel to the pinned-layer magnetization, i.e. $$R = R_{\rm P}+\frac{\Delta R}{2}(1-\mathbf{m}\cdot\mathbf{p}),$$2 or $$R = R_{\rm P}+\frac{\Delta R}{2}(1-\mathbf{m}\cdot\mathbf{p}),$$3 depending on the sign of $$R = R_{\rm P}+\frac{\Delta R}{2}(1-\mathbf{m}\cdot\mathbf{p}),$$4. At larger field, the optimal field azimuth shifts to finite values so that the free layer maintains the optimal relative angle to the pinned layer. For $$R = R_{\rm P}+\frac{\Delta R}{2}(1-\mathbf{m}\cdot\mathbf{p}),$$5 kOe, the reported optimal directions are $$R = R_{\rm P}+\frac{\Delta R}{2}(1-\mathbf{m}\cdot\mathbf{p}),$$6 and $$R = R_{\rm P}+\frac{\Delta R}{2}(1-\mathbf{m}\cdot\mathbf{p}),$$7 for $$R = R_{\rm P}+\frac{\Delta R}{2}(1-\mathbf{m}\cdot\mathbf{p}),$$8 mA, and $$R = R_{\rm P}+\frac{\Delta R}{2}(1-\mathbf{m}\cdot\mathbf{p}),$$9 and $$\frac{d\mathbf{m}}{dt} = -\gamma \mathbf{m}\times\mathbf{H} -\gamma a_J\,\mathbf{m}\times(\mathbf{p}\times\mathbf{m}) +\gamma b_J\,\mathbf{p}\times\mathbf{m} +\alpha\,\mathbf{m}\times\frac{d\mathbf{m}}{dt},$$0 for $$\frac{d\mathbf{m}}{dt} = -\gamma \mathbf{m}\times\mathbf{H} -\gamma a_J\,\mathbf{m}\times(\mathbf{p}\times\mathbf{m}) +\gamma b_J\,\mathbf{p}\times\mathbf{m} +\alpha\,\mathbf{m}\times\frac{d\mathbf{m}}{dt},$$1 mA; the maximum diode voltage is then about $$\frac{d\mathbf{m}}{dt} = -\gamma \mathbf{m}\times\mathbf{H} -\gamma a_J\,\mathbf{m}\times(\mathbf{p}\times\mathbf{m}) +\gamma b_J\,\mathbf{p}\times\mathbf{m} +\alpha\,\mathbf{m}\times\frac{d\mathbf{m}}{dt},$$2 (Taniguchi et al., 2013).

Electrical bias can tune not only the torque amplitude but also the resonance condition. In a voltage-tunable RF detector based on a weakly in-plane-anisotropic CoFeB/MgO free layer, the bias voltage modifies the perpendicular anisotropy through VCMA, shifting the detection frequency by about $$\frac{d\mathbf{m}}{dt} = -\gamma \mathbf{m}\times\mathbf{H} -\gamma a_J\,\mathbf{m}\times(\mathbf{p}\times\mathbf{m}) +\gamma b_J\,\mathbf{p}\times\mathbf{m} +\alpha\,\mathbf{m}\times\frac{d\mathbf{m}}{dt},$$3. Micromagnetic modeling required a change in perpendicular anisotropy of about $$\frac{d\mathbf{m}}{dt} = -\gamma \mathbf{m}\times\mathbf{H} -\gamma a_J\,\mathbf{m}\times(\mathbf{p}\times\mathbf{m}) +\gamma b_J\,\mathbf{p}\times\mathbf{m} +\alpha\,\mathbf{m}\times\frac{d\mathbf{m}}{dt},$$4 at $$\frac{d\mathbf{m}}{dt} = -\gamma \mathbf{m}\times\mathbf{H} -\gamma a_J\,\mathbf{m}\times(\mathbf{p}\times\mathbf{m}) +\gamma b_J\,\mathbf{p}\times\mathbf{m} +\alpha\,\mathbf{m}\times\frac{d\mathbf{m}}{dt},$$5 V to reproduce the measured resonance shift, and the peak detection frequency was shifted from about $$\frac{d\mathbf{m}}{dt} = -\gamma \mathbf{m}\times\mathbf{H} -\gamma a_J\,\mathbf{m}\times(\mathbf{p}\times\mathbf{m}) +\gamma b_J\,\mathbf{p}\times\mathbf{m} +\alpha\,\mathbf{m}\times\frac{d\mathbf{m}}{dt},$$6 GHz to $$\frac{d\mathbf{m}}{dt} = -\gamma \mathbf{m}\times\mathbf{H} -\gamma a_J\,\mathbf{m}\times(\mathbf{p}\times\mathbf{m}) +\gamma b_J\,\mathbf{p}\times\mathbf{m} +\alpha\,\mathbf{m}\times\frac{d\mathbf{m}}{dt},$$7 GHz by changing bias polarity (Skowroński et al., 2014).

In STVOs, the control variables are partly topological. The sign of the rectified voltage reverses with vortex chirality because the Ampère-Oersted field couples differently to clockwise and counterclockwise circulation. The same system also exhibits strong dependence on rf input power, reported for input powers from $$\frac{d\mathbf{m}}{dt} = -\gamma \mathbf{m}\times\mathbf{H} -\gamma a_J\,\mathbf{m}\times(\mathbf{p}\times\mathbf{m}) +\gamma b_J\,\mathbf{p}\times\mathbf{m} +\alpha\,\mathbf{m}\times\frac{d\mathbf{m}}{dt},$$8 down to $$\frac{d\mathbf{m}}{dt} = -\gamma \mathbf{m}\times\mathbf{H} -\gamma a_J\,\mathbf{m}\times(\mathbf{p}\times\mathbf{m}) +\gamma b_J\,\mathbf{p}\times\mathbf{m} +\alpha\,\mathbf{m}\times\frac{d\mathbf{m}}{dt},$$9, and a half-frequency spectral feature identified by DD-TEA as a fractional synchronization pattern (Chopin et al., 2023).

Additional interactions can restructure the resonance entirely. In an MTJ built on a Ta strip, sufficiently strong microwave drive together with interfacial Dzyaloshinskii-Moriya interaction (DMI) splits a single FMR peak into two. For $a_J$0 and $a_J$1, the no-DMI response has one peak at about $a_J$2 GHz, while for $a_J$3 the peak splits into two resonances at about $a_J$4 GHz and $a_J$5 GHz, with non-uniform spatial mode distributions (Tomasello et al., 2014).

5. Nonlinear regimes, synchronization, and localized or topological dynamics

A recurring simplification is to identify the spin-torque diode effect exclusively with linear FMR. The literature does not support that restriction. In a forced-synchronized spin-torque oscillator, the diode voltage is fundamentally nonlinear and is governed by the phase difference $a_J$6 between the auto-oscillation and the ac drive:

$a_J$7

Because $a_J$8 is constant only in the injection-locked state, a finite long-time-averaged diode voltage appears only within the locking region. In the reported macrospin example, injection locking occurs for $a_J$9 to $b_J=\beta a_J$0 GHz, the voltage is nearly zero outside that interval, and inside it the voltage varies approximately linearly with detuning, enabling experimental phase estimation (Yamaguchi et al., 2020).

Injection locking also underlies the highest reported low-power sensitivities in bias-field-free MTJ detectors. In a perpendicularly magnetized free-layer device, the measured room-temperature sensitivity reaches $b_J=\beta a_J$1 at an input power of $b_J=\beta a_J$2 nW under $b_J=\beta a_J$3 mA, while even at $b_J=\beta a_J$4 the bias-field-free sensitivity is $b_J=\beta a_J$5. Micromagnetic simulations and microwave-emission measurements associate the enhancement with locking between self-oscillation and the injected rf signal, including a reported frequency jump of about $b_J=\beta a_J$6 MHz near $b_J=\beta a_J$7 mA (Fang et al., 2014).

Localized magnetic textures provide additional departures from the linear-uniform picture. In skyrmion-based MTJs, the diode signal in the skyrmion state shows a resonance near $b_J=\beta a_J$8 GHz with linewidth about $b_J=\beta a_J$9 MHz, compared with about $$V_{\rm dc}(f_{\rm res}) = \frac{\Delta R I_{\rm ac}}{4} \frac{\tilde{a}_{J({\rm ac})}\sin^{2}(\varphi-\varphi_{\rm p})} {\alpha(H_X+H_Y)-2a_{J({\rm dc})}\cos(\varphi-\varphi_{\rm p})},$$0 MHz for the high-frequency uniform mode and $$V_{\rm dc}(f_{\rm res}) = \frac{\Delta R I_{\rm ac}}{4} \frac{\tilde{a}_{J({\rm ac})}\sin^{2}(\varphi-\varphi_{\rm p})} {\alpha(H_X+H_Y)-2a_{J({\rm dc})}\cos(\varphi-\varphi_{\rm p})},$$1 MHz for the low-frequency mode in the uniform antiparallel state. After a field protocol that stabilizes a skyrmion at zero applied field, the device also exhibits a zero-field spectrum around $$V_{\rm dc}(f_{\rm res}) = \frac{\Delta R I_{\rm ac}}{4} \frac{\tilde{a}_{J({\rm ac})}\sin^{2}(\varphi-\varphi_{\rm p})} {\alpha(H_X+H_Y)-2a_{J({\rm dc})}\cos(\varphi-\varphi_{\rm p})},$$2 GHz, with both positive and negative voltage peaks attributed to a field-induced phase change between resistance oscillation and injected current (Fang et al., 2024).

In domain-wall devices, the spin-torque diode effect serves as a quantitative probe of damping in confined, large-amplitude modes. The measured average linewidth is $$V_{\rm dc}(f_{\rm res}) = \frac{\Delta R I_{\rm ac}}{4} \frac{\tilde{a}_{J({\rm ac})}\sin^{2}(\varphi-\varphi_{\rm p})} {\alpha(H_X+H_Y)-2a_{J({\rm dc})}\cos(\varphi-\varphi_{\rm p})},$$3 GHz for both the domain-wall mode and the domain mode, corresponding to an effective damping parameter $$V_{\rm dc}(f_{\rm res}) = \frac{\Delta R I_{\rm ac}}{4} \frac{\tilde{a}_{J({\rm ac})}\sin^{2}(\varphi-\varphi_{\rm p})} {\alpha(H_X+H_Y)-2a_{J({\rm dc})}\cos(\varphi-\varphi_{\rm p})},$$4, compared with $$V_{\rm dc}(f_{\rm res}) = \frac{\Delta R I_{\rm ac}}{4} \frac{\tilde{a}_{J({\rm ac})}\sin^{2}(\varphi-\varphi_{\rm p})} {\alpha(H_X+H_Y)-2a_{J({\rm dc})}\cos(\varphi-\varphi_{\rm p})},$$5 in the unpatterned host film. Simulations show that neither dipolar coupling to the reference layer nor intralayer transverse spin pumping is sufficient to explain the full enhancement, leading to the conclusion that the increased damping is intrinsic to large-amplitude localized excitations such as vibrating or propagating domain walls (Lequeux et al., 2015).

6. Performance metrics, applications, and conceptual boundaries

The spin-torque diode effect is technologically relevant because it can combine sensitivity, tunability, compactness, and magnetic-state selectivity in a single nanoscale element. In asymmetric-electrode MgO MTJs, a diode sensitivity of $$V_{\rm dc}(f_{\rm res}) = \frac{\Delta R I_{\rm ac}}{4} \frac{\tilde{a}_{J({\rm ac})}\sin^{2}(\varphi-\varphi_{\rm p})} {\alpha(H_X+H_Y)-2a_{J({\rm dc})}\cos(\varphi-\varphi_{\rm p})},$$6 after impedance-matching correction was reported at $$V_{\rm dc}(f_{\rm res}) = \frac{\Delta R I_{\rm ac}}{4} \frac{\tilde{a}_{J({\rm ac})}\sin^{2}(\varphi-\varphi_{\rm p})} {\alpha(H_X+H_Y)-2a_{J({\rm dc})}\cos(\varphi-\varphi_{\rm p})},$$7 mT, with the large zero-bias anti-Lorentzian signal traced to an unusually strong out-of-plane torque. In voltage-tunable MTJ detectors, the main practical limitations were high resistance, impedance mismatch to a $$V_{\rm dc}(f_{\rm res}) = \frac{\Delta R I_{\rm ac}}{4} \frac{\tilde{a}_{J({\rm ac})}\sin^{2}(\varphi-\varphi_{\rm p})} {\alpha(H_X+H_Y)-2a_{J({\rm dc})}\cos(\varphi-\varphi_{\rm p})},$$8 environment, RC-limited bandwidth of about $$V_{\rm dc}(f_{\rm res}) = \frac{\Delta R I_{\rm ac}}{4} \frac{\tilde{a}_{J({\rm ac})}\sin^{2}(\varphi-\varphi_{\rm p})} {\alpha(H_X+H_Y)-2a_{J({\rm dc})}\cos(\varphi-\varphi_{\rm p})},$$9 GHz, and low quality factor (Matsumoto et al., 2011, Skowroński et al., 2014).

Bias-field-free operation is a second major application direction. In a PMA-based spintronic diode with first- and second-order anisotropy, sufficiently large rf currents can excite out-of-plane precession without any external magnetic field for frequencies $$\psi=\cos^{-1}(\mathbf{m}^{(0)}\!\cdot\!\mathbf{p})$$0 MHz. In that regime the device can operate as a broadband energy harvester with numerically obtained efficiency about $$\psi=\cos^{-1}(\mathbf{m}^{(0)}\!\cdot\!\mathbf{p})$$1, reduced to about $$\psi=\cos^{-1}(\mathbf{m}^{(0)}\!\cdot\!\mathbf{p})$$2 when impedance mismatch is included; the practical optimum was identified near $$\psi=\cos^{-1}(\mathbf{m}^{(0)}\!\cdot\!\mathbf{p})$$3, with threshold power $$\psi=\cos^{-1}(\mathbf{m}^{(0)}\!\cdot\!\mathbf{p})$$4 nW and maximum operational frequency around $$\psi=\cos^{-1}(\mathbf{m}^{(0)}\!\cdot\!\mathbf{p})$$5–$$\psi=\cos^{-1}(\mathbf{m}^{(0)}\!\cdot\!\mathbf{p})$$6 MHz (Artemchuk et al., 2020).

The effect is also migrating from sensing to computation. In STVO-based neuromorphic proposals, vortex chirality provides non-volatile state encoding, while nonlinear dynamical response supplies activation-function-like behavior; the spin-diode readout converts those dynamical states into electrical outputs, and the DD-TEA surrogate model is presented as a route toward simulation of full neuromorphic circuits rather than single oscillators בלבד (Chopin et al., 2023).

A broader implication is that the “diode” concept is no longer confined to tunnel-junction microwave rectifiers. In current-in-plane normal-metal/ferromagnet bilayers, the rectified output is a dc in-plane charge current generated by a quadratic SHE/ISHE response rather than by direct TMR or GMR mixing. The reported theory predicts especially large signals in metallic systems such as Au$$\psi=\cos^{-1}(\mathbf{m}^{(0)}\!\cdot\!\mathbf{p})$$7Fe, up to two orders of magnitude larger than in Pt$$\psi=\cos^{-1}(\mathbf{m}^{(0)}\!\cdot\!\mathbf{p})$$8YIG because of stronger longitudinal spin transport and longer spin-diffusion length (Gems et al., 10 Sep 2025).

A final conceptual boundary concerns terminology. Not every “spin diode” is a spin-torque diode. The ferromagnetic STM-tip/adatom system described as a spin diode is a dc transport asymmetry controlled by Coulomb blockade, tip polarization, and tip position; it does not rely on microwave-driven magnetization dynamics or on the current-resistance mixing mechanism that defines the conventional spin-torque diode effect (Penteado et al., 2011). This distinction matters because it separates spin-dependent rectification in general from the specific resonant spin-dynamical phenomenon denoted by the spin-torque diode effect.