Vertical Magnetic Tunnel Junctions

• Vertical magnetic tunnel junctions are current-perpendicular-to-plane heterostructures with two magnetic electrodes separated by an ultrathin insulating barrier enabling quantum tunneling and TMR effects.
• They exploit coherent Bloch-state filtering and symmetry selection in crystalline barriers like MgO to achieve high tunneling magnetoresistance for memory, logic, and oscillator applications.
• Recent research advances include barrier and electrode engineering, van der Waals and multiferroic integrations, and novel switching modes that broaden functionality beyond traditional ferromagnetic systems.

Vertical magnetic tunnel junctions are current-perpendicular-to-plane heterostructures in which two magnetic electrodes are separated by an ultrathin insulating barrier, so that quantum tunneling converts magnetic order into an electrical resistance or conductance contrast. In the canonical case, the central observable is tunneling magnetoresistance, defined by the difference between the parallel and antiparallel magnetic configurations of the electrodes. The class now encompasses crystalline Fe/MgO/Fe junctions governed by coherent Bloch-state tunneling, textured oxide stacks, van der Waals heterostructures, antiferromagnetic tunnel junctions, multiferroic tunnel junctions, and voltage- or current-driven devices for memory, logic, oscillation, and thermospin conversion (Duluard et al., 2011, Li et al., 2019, Yang et al., 15 Jun 2025).

1. Fundamental transport framework

The defining geometry of a vertical magnetic tunnel junction is current-perpendicular-to-plane transport through a thin barrier. In its standard form, the resistance-based and conductance-based tunneling magnetoresistance are written as

$$\mathrm{TMR}=\frac{R_{\mathrm{AP}}-R_{\mathrm{P}}}{R_{\mathrm{P}}}, \qquad \mathrm{TMR}=\frac{G_{\mathrm{P}}-G_{\mathrm{AP}}}{G_{\mathrm{AP}}},$$

with $R_{\mathrm{P,AP}}$ and $G_{\mathrm{P,AP}}$ denoting the parallel and antiparallel magnetic states. For coherent tunneling, the conductance is naturally expressed in a Landauer form,

$$G=\frac{e^2}{h}\sum_{\mathbf{k}_{\parallel},\sigma}T_{\sigma}(\mathbf{k}_{\parallel}),$$

or, in thermal generalizations, through transport integrals over the energy-dependent conductance density $g(E)$ (Duluard et al., 2011, López-Monís et al., 2013).

Two transport pictures coexist in the literature. The phenomenological Jullière model,

$\mathrm{TMR}=\frac{2P_1P_2}{1-P_1P_2},$

treats tunneling as the product of independent spin polarizations of the two electrodes. It remains useful for barriers such as $\mathrm{AlO_x}$, and it was used, for example, to estimate $P\approx 17\%$ in a $\mathrm{Fe_{0.25}TaS_2}$-based van der Waals junction with $\mathrm{TMR}\approx 6\%$ at $R_{\mathrm{P,AP}}$0 (Arai et al., 2015). By contrast, crystalline barriers require a symmetry-resolved description: in-plane momentum $R_{\mathrm{P,AP}}$1 and Bloch-state symmetry determine which propagating electrode states can couple to the slowest-decaying evanescent states in the barrier, and hence which spin channel dominates (Duluard et al., 2011, Li et al., 2019).

This distinction is central to the modern theory of vertical MTJs. In MgO-based junctions the slowest-decaying evanescent channel has $R_{\mathrm{P,AP}}$2 symmetry; in ScN-based junctions both $R_{\mathrm{P,AP}}$3 at $R_{\mathrm{P,AP}}$4 and $R_{\mathrm{P,AP}}$5 at $R_{\mathrm{P,AP}}$6 contribute low-decay channels; in several van der Waals structures the dominant mechanism is not conventional symmetry filtering but $R_{\mathrm{P,AP}}$7-resolved Bloch-state matching or mismatch between the spin channels of the electrodes (Karki et al., 2020, Li et al., 2019). Thermally driven transport preserves the same basic formalism: in the analytical treatment of tunneling magneto-thermopower, the junction thermopower and tunneling magneto-thermopower inherit the angle dependence of the spin-dependent transmission, and the tunneling magneto-thermopower can even change sign or vanish at a characteristic barrier height (López-Monís et al., 2013).

2. Fe/MgO/Fe(100) and the coherent-tunneling paradigm

The archetypal vertical MTJ is Fe/MgO/Fe(100). Its status derives from the conjunction of a majority-spin $R_{\mathrm{P,AP}}$8 Bloch state in bcc Fe at the Fermi level and a MgO(100) complex band structure whose slowest-decaying evanescent state also has $R_{\mathrm{P,AP}}$9 symmetry. This symmetry filtering strongly enhances parallel-state conductance and suppresses antiparallel transmission, establishing the modern coherent-tunneling picture for oxide MTJs (Duluard et al., 2011).

The structural distinction between epitaxial and textured Fe/MgO/Fe junctions is important but not absolute. In epitaxial junctions, coherence is supported by crystalline registry across the interfaces and is verified by XRD, pole figures, RHEED, TEM, and AFM. In sputtered, annealed $G_{\mathrm{P,AP}}$0-textured junctions, the out-of-plane orientation is preserved but the in-plane orientation varies from grain to grain. The critical result is that strong similarities in the transport properties of $G_{\mathrm{P,AP}}$1-textured and $G_{\mathrm{P,AP}}$2-epitaxial Fe/MgO/Fe show that structural coherence and magnetic quality at the $G_{\mathrm{P,AP}}$3 grain scale are sufficient to issue signatures of the spin-polarized transport specific to a single-crystal junction. The textured junction can therefore be modeled as a juxtaposition of nanometer-sized single-crystal junctions placed in parallel, with

$G_{\mathrm{P,AP}}$4

This establishes that the lateral coherence of the Bloch tunneling wave function can be identically limited in epitaxial and textured systems once the relevant disorder length scales become comparable (Duluard et al., 2011).

Representative Fe/MgO/Fe transport behavior reported in the supplied synthesis is correspondingly strong. Room-temperature TMR of at least $G_{\mathrm{P,AP}}$5–$G_{\mathrm{P,AP}}$6 is common in well-textured sputtered Fe/MgO/Fe, while epitaxial junctions and low-temperature measurements can exceed $G_{\mathrm{P,AP}}$7–$G_{\mathrm{P,AP}}$8. For MgO thicknesses of about $G_{\mathrm{P,AP}}$9–$$G=\frac{e^2}{h}\sum_{\mathbf{k}_{\parallel},\sigma}T_{\sigma}(\mathbf{k}_{\parallel}),$$0, resistance-area products of about $$G=\frac{e^2}{h}\sum_{\mathbf{k}_{\parallel},\sigma}T_{\sigma}(\mathbf{k}_{\parallel}),$$1–$$G=\frac{e^2}{h}\sum_{\mathbf{k}_{\parallel},\sigma}T_{\sigma}(\mathbf{k}_{\parallel}),$$2 are typical, with exponential scaling in thickness through the barrier decay constant. The bias dependence is smooth in the parallel state and more strongly suppressed in the antiparallel state; $$G=\frac{e^2}{h}\sum_{\mathbf{k}_{\parallel},\sigma}T_{\sigma}(\mathbf{k}_{\parallel}),$$3 and $$G=\frac{e^2}{h}\sum_{\mathbf{k}_{\parallel},\sigma}T_{\sigma}(\mathbf{k}_{\parallel}),$$4 frequently show a zero-bias anomaly and weak features at biases of hundreds of millivolts (Duluard et al., 2011).

The same MgO platform also underlies vertical domain-wall manipulation. In an MgO-based MTJ with TMR $$G=\frac{e^2}{h}\sum_{\mathbf{k}_{\parallel},\sigma}T_{\sigma}(\mathbf{k}_{\parallel}),$$5 and $$G=\frac{e^2}{h}\sum_{\mathbf{k}_{\parallel},\sigma}T_{\sigma}(\mathbf{k}_{\parallel}),$$6, vertical current injection moved a domain wall in the free layer at current densities of order $$G=\frac{e^2}{h}\sum_{\mathbf{k}_{\parallel},\sigma}T_{\sigma}(\mathbf{k}_{\parallel}),$$7–$$G=\frac{e^2}{h}\sum_{\mathbf{k}_{\parallel},\sigma}T_{\sigma}(\mathbf{k}_{\parallel}),$$8, about $$G=\frac{e^2}{h}\sum_{\mathbf{k}_{\parallel},\sigma}T_{\sigma}(\mathbf{k}_{\parallel}),$$9 times smaller than in conventional in-plane geometries. The decisive torque was the out-of-plane, field-like spin-transfer torque, which reached up to about $g(E)$0 of the in-plane torque at low bias and produced effective fields of about $g(E)$1 at $g(E)$2 (Chanthbouala et al., 2011).

3. Barrier and electrode engineering beyond the canonical MgO stack

A major trajectory in vertical-MTJ research is the search for barriers and electrodes that preserve high spin contrast while relaxing the materials restrictions of the Fe/MgO/Fe paradigm. One route is narrower-gap barrier engineering. In Fe/ScN/Fe(001), first-principles transport shows that ScN supports two low-decay channels, $g(E)$3 at $g(E)$4 and $g(E)$5 at $g(E)$6, both with $g(E)$7, compared with $g(E)$8 for MgO. At six barrier layers, the calculated Fe/ScN/Fe junction has $g(E)$9, $\mathrm{TMR}=\frac{2P_1P_2}{1-P_1P_2},$0, $\mathrm{TMR}=\frac{2P_1P_2}{1-P_1P_2},$1, and $\mathrm{TMR}=\frac{2P_1P_2}{1-P_1P_2},$2, whereas the comparison Fe/MgO/Fe junction at the same thickness has $\mathrm{TMR}=\frac{2P_1P_2}{1-P_1P_2},$3 but $\mathrm{TMR}=\frac{2P_1P_2}{1-P_1P_2},$4. The proposed significance is not a larger TMR than MgO, but high spin filtering at much lower resistance-area product (Karki et al., 2020).

A second route is crystallographic diversification. Fully epitaxial fcc(111) Co$\mathrm{TMR}=\frac{2P_1P_2}{1-P_1P_2},$5Fe$\mathrm{TMR}=\frac{2P_1P_2}{1-P_1P_2},$6/MgAlO/Co$\mathrm{TMR}=\frac{2P_1P_2}{1-P_1P_2},$7Fe$\mathrm{TMR}=\frac{2P_1P_2}{1-P_1P_2},$8 junctions on Ru(0001) demonstrate that close-packed, polar $\mathrm{TMR}=\frac{2P_1P_2}{1-P_1P_2},$9 barriers can be stabilized by interface engineering. At the CoFe(111)/MgAlO(111) interface, a $\mathrm{AlO_x}$0 in-plane domain-matching structure with periodic dislocations reduces the effective mismatch to about $\mathrm{AlO_x}$1, and the resulting MTJs exhibit $\mathrm{AlO_x}$2 TMR at room temperature and $\mathrm{AlO_x}$3 at $\mathrm{AlO_x}$4. Their $\mathrm{AlO_x}$5 is about $\mathrm{AlO_x}$6, and the differential conductance is highly symmetric with respect to bias polarity, indicating nearly identical upper and lower interface qualities (Song et al., 2023).

A third route is electrode engineering. In the theoretical Fe/MgO superlattice proposal, inserting Fe/MgO multilayers into both fixed and free electrodes transforms the barrier-proximal projected density of states into a half-metallic one: majority states become gapped at $\mathrm{AlO_x}$7, minority states remain available, and transport evolves from direct majority-spin $\mathrm{AlO_x}$8 tunneling to resonant minority-spin tunneling as the number of Fe/MgO repeats increases. In the ideal ballistic calculations, the TMR rises from $\mathrm{AlO_x}$9 for the reference $P\approx 17\%$0 junction to $P\approx 17\%$1 for $P\approx 17\%$2, while the same architecture simultaneously lowers the effective $P\approx 17\%$3 and produces robust perpendicular magnetic anisotropy for most $P\approx 17\%$4 multilayers examined (Lanzillo et al., 27 Feb 2025).

These examples mark a broader shift in the field. The design problem is no longer restricted to maximizing a single spin-filtering channel in one crystallographic family; it increasingly involves the co-optimization of symmetry filtering, decay rate, barrier polarity, interface matching, anisotropy, and write-energy constraints.

4. Van der Waals and multiferroic vertical junctions

Van der Waals materials extend the vertical-MTJ concept into a regime of atomically sharp yet weakly hybridized interfaces. In Fe$P\approx 17\%$5GeTe$P\approx 17\%$6/graphene/Fe$P\approx 17\%$7GeTe$P\approx 17\%$8 and Fe$P\approx 17\%$9GeTe$\mathrm{Fe_{0.25}TaS_2}$0/h-BN/Fe$\mathrm{Fe_{0.25}TaS_2}$1GeTe$\mathrm{Fe_{0.25}TaS_2}$2, the relaxed interface separation is about $\mathrm{Fe_{0.25}TaS_2}$3, much larger than the $\mathrm{Fe_{0.25}TaS_2}$4 spacing of about $\mathrm{Fe_{0.25}TaS_2}$5, and the projected band structures show negligible hybridization. The central mechanism is a severe mismatch between majority-spin Bloch states on one side and minority-spin states on the other in the antiparallel configuration. At zero bias, the calculated total transmissions are $\mathrm{Fe_{0.25}TaS_2}$6 and $\mathrm{Fe_{0.25}TaS_2}$7 for the graphene-spaced parallel and antiparallel states, giving $\mathrm{Fe_{0.25}TaS_2}$8; for h-BN the corresponding values are $\mathrm{Fe_{0.25}TaS_2}$9 and $\mathrm{TMR}\approx 6\%$0, giving $\mathrm{TMR}\approx 6\%$1. The current on/off ratio exceeds $\mathrm{TMR}\approx 6\%$2 over a broad bias range, with a characteristic half-width of about $\mathrm{TMR}\approx 6\%$3 (Li et al., 2019).

Experimentally, all-van-der-Waals MTJs can be much simpler and less ideal. A junction assembled by dry transfer from two exfoliated $\mathrm{TMR}\approx 6\%$4 flakes uses native $\mathrm{TMR}\approx 6\%$5 at the cleaved surfaces as the tunnel barrier. The junction area is $\mathrm{TMR}\approx 6\%$6, $\mathrm{TMR}\approx 6\%$7 at $\mathrm{TMR}\approx 6\%$8, and the TMR is about $\mathrm{TMR}\approx 6\%$9 at $R_{\mathrm{P,AP}}$00, decreasing with increasing $R_{\mathrm{P,AP}}$01 and nearly symmetric in bias polarity. This device established spin-polarized tunneling in a fully van der Waals ferromagnet/oxide/ferromagnet stack, but it also made clear that native-oxide uniformity and true-contact area are decisive for practical performance (Arai et al., 2015).

A more elaborate van der Waals architecture combines tunneling readout with spin-orbit-torque writing. In the monolayer stack $R_{\mathrm{P,AP}}$02, the semiconducting $R_{\mathrm{P,AP}}$03 monolayer acts as the tunnel barrier while metallic $R_{\mathrm{P,AP}}$04 supplies spin Hall conductivity. The equilibrium TMR at $R_{\mathrm{P,AP}}$05 is $R_{\mathrm{P,AP}}$06, but under bias the antiparallel resonance is shifted out of the transport window and the non-equilibrium TMR reaches $R_{\mathrm{P,AP}}$07 at $R_{\mathrm{P,AP}}$08 and $R_{\mathrm{P,AP}}$09, or $R_{\mathrm{P,AP}}$10 at $R_{\mathrm{P,AP}}$11. The intrinsic spin Hall conductivity of $R_{\mathrm{P,AP}}$12 is $R_{\mathrm{P,AP}}$13 at the Fermi level and can be increased up to about $R_{\mathrm{P,AP}}$14 at $R_{\mathrm{P,AP}}$15 (Zhou et al., 2019).

Van der Waals materials also make possible a “one-material” MTJ. In multilayer MnPX$R_{\mathrm{P,AP}}$16 ($R_{\mathrm{P,AP}}$17), an external electric field can drive only the outermost layers into half-metallic ferromagnetic states while the inner layers remain insulating antiferromagnets. The reported charge-transfer thresholds are about $R_{\mathrm{P,AP}}$18 carriers per atom for Li-like $R_{\mathrm{P,AP}}$19-doping and $R_{\mathrm{P,AP}}$20 per atom for F-like $R_{\mathrm{P,AP}}$21-doping, equivalent to surface charge densities of about $R_{\mathrm{P,AP}}$22 and $R_{\mathrm{P,AP}}$23, respectively. The resulting spin-filtering picture is idealized, and the stated TMR is $R_{\mathrm{P,AP}}$24 in the pessimistic definition used there (Jin et al., 2022).

Multiferroic functionality can be implemented in both oxide and van der Waals vertical junctions. An epitaxial $R_{\mathrm{P,AP}}$25 junction with a $R_{\mathrm{P,AP}}$26 ferroelectric HZO barrier exhibits TMR of $R_{\mathrm{P,AP}}$27–$R_{\mathrm{P,AP}}$28 at $R_{\mathrm{P,AP}}$29 and $R_{\mathrm{P,AP}}$30, inverse TMR of $R_{\mathrm{P,AP}}$31 at $R_{\mathrm{P,AP}}$32, resistance-area products above $R_{\mathrm{P,AP}}$33, and four non-volatile resistance states generated by combining magnetic and ferroelectric switching. The ON/OFF ratio is about $R_{\mathrm{P,AP}}$34, and the ferroelectric-state TER in the TMR measurement configuration is about $R_{\mathrm{P,AP}}$35 (Wei et al., 2019). In the theoretical $R_{\mathrm{P,AP}}$36 van der Waals multiferroic junction with asymmetric PtTe$R_{\mathrm{P,AP}}$37 and alkali-doped/intercalated $R_{\mathrm{P,AP}}$38 electrodes, the out-of-plane ferroelectric polarization is $R_{\mathrm{P,AP}}$39, the zero-bias TMR exceeds $R_{\mathrm{P,AP}}$40, TER reaches $R_{\mathrm{P,AP}}$41, and under bias they increase to $R_{\mathrm{P,AP}}$42 and $R_{\mathrm{P,AP}}$43, respectively; a peak-to-valley ratio of $R_{\mathrm{P,AP}}$44 is reported for the negative-differential-resistance response (Yan et al., 1 Jan 2025).

5. Antiferromagnetic and interface-controlled tunnel junctions

A persistent assumption in spintronics is that large TMR requires two ferromagnetic electrodes. Several recent vertical-MTJ proposals and experiments directly contradict that assumption. In the all-rutile RuO$R_{\mathrm{P,AP}}$45/TiO$R_{\mathrm{P,AP}}$46/CrO$R_{\mathrm{P,AP}}$47 design, CrO$R_{\mathrm{P,AP}}$48 is the sole ferromagnetic electrode, while RuO$R_{\mathrm{P,AP}}$49 is an antiferromagnetic altermagnet. In the $R_{\mathrm{P,AP}}$50-oriented junction, RuO$R_{\mathrm{P,AP}}$51 provides a spin-polarized conduction-channel distribution and the calculated TMR is about $R_{\mathrm{P,AP}}$52; in the $R_{\mathrm{P,AP}}$53 orientation, the mechanism is different and relies on a Néel spin current combined with sublattice-dependent barrier thickness, producing about $R_{\mathrm{P,AP}}$54 TMR for a $R_{\mathrm{P,AP}}$55-monolayer TiO$R_{\mathrm{P,AP}}$56 barrier and exceeding $R_{\mathrm{P,AP}}$57 at $R_{\mathrm{P,AP}}$58 (Samanta et al., 2023).

A second antiferromagnetic route is purely interfacial. In the interface-controlled A-type AFMTJ based on Fe$R_{\mathrm{P,AP}}$59GeTe$R_{\mathrm{P,AP}}$60, the bulk electrode is spin-degenerate by $R_{\mathrm{P,AP}}$61 symmetry, but the surface termination is magnetically uncompensated and generates a finite interfacial spin polarization. In a semi-infinite FGT/BN/PtTe$R_{\mathrm{P,AP}}$62 structure, the transmission spin polarization at the Fermi level is about $R_{\mathrm{P,AP}}$63 despite spin-degenerate bulk bands. In full PtTe$R_{\mathrm{P,AP}}$64/[FGT]$R_{\mathrm{P,AP}}$65/BN/[FGT]$R_{\mathrm{P,AP}}$66/PtTe$R_{\mathrm{P,AP}}$67 junctions with $R_{\mathrm{P,AP}}$68, the TMR is about $R_{\mathrm{P,AP}}$69 around $R_{\mathrm{P,AP}}$70, remaining strongly negative over a broad energy range and decreasing to about $R_{\mathrm{P,AP}}$71 at $R_{\mathrm{P,AP}}$72 bias. The negative sign is not anomalous within this mechanism: the conductance is higher when the Néel vectors are antiparallel because that state yields parallel interfacial moments across the barrier (Yang et al., 15 Jun 2025).

This interfacial picture is now experimentally realized in all-collinear van der Waals AFMTJs based on $R_{\mathrm{P,AP}}$73. The maximum TMR reaches $R_{\mathrm{P,AP}}$74 at $R_{\mathrm{P,AP}}$75, and the large peaks occur at $R_{\mathrm{P,AP}}$76, well inside the antiferromagnetic regime and distinctly separated from the AFM-to-FM transition fields of roughly $R_{\mathrm{P,AP}}$77 to $R_{\mathrm{P,AP}}$78. In an even-even device, the two characteristic interfacial coercive fields are $R_{\mathrm{P,AP}}$79 and $R_{\mathrm{P,AP}}$80 at $R_{\mathrm{P,AP}}$81, both vanishing near $R_{\mathrm{P,AP}}$82. By controlling even- or odd-layer parity in the electrodes, the devices select volatile or non-volatile TMR behavior; in odd-even devices, $R_{\mathrm{P,AP}}$83 crosses zero near $R_{\mathrm{P,AP}}$84, and above that temperature the TMR becomes non-volatile without a dedicated pinning layer (Zhao et al., 17 Jul 2025).

The conceptual consequence is substantial. Vertical MTJs can exploit not only bulk spin-split ferromagnets, but also altermagnetic band structures, uncompensated antiferromagnetic interfaces, and Néel-current physics. The operative spin polarization may therefore be bulk, interfacial, or sublattice-selective rather than uniformly ferromagnetic.

6. Switching modes, nonlinear operation, and device functions

Vertical MTJs have evolved from passive readout elements into actively reconfigurable devices with multiple write pathways. Voltage-controlled magnetic anisotropy is one such pathway. In MgO/CoFeB nanopillars with remote Ir doping near, but not at, the MgO/CoFeB interface, precessional VCMA switching in $R_{\mathrm{P,AP}}$85 pillars achieved $R_{\mathrm{P,AP}}$86 with a $R_{\mathrm{P,AP}}$87, $R_{\mathrm{P,AP}}$88 pulse and a switching energy of $R_{\mathrm{P,AP}}$89. The same remote-doped sample retained TMR of $R_{\mathrm{P,AP}}$90–$R_{\mathrm{P,AP}}$91 after $R_{\mathrm{P,AP}}$92 annealing and showed room-temperature thermal stability $R_{\mathrm{P,AP}}$93 (Zhang et al., 22 Nov 2025).

Spin-transfer-torque functionality can also be distributed over several pillars by sharing a continuous free layer. In the perpendicular-MTJ majority-gate architecture, blanket films had TMR $R_{\mathrm{P,AP}}$94 and $R_{\mathrm{P,AP}}$95, but the patterned $R_{\mathrm{P,AP}}$96 pillars suffered degraded device-level TMR of about $R_{\mathrm{P,AP}}$97–$R_{\mathrm{P,AP}}$98 because of sidewall “fencing” during fabrication. Micromagnetic simulations nevertheless showed that a downscaled cross-shaped free layer with $R_{\mathrm{P,AP}}$99 arm width and lateral size $G_{\mathrm{P,AP}}$00 can perform majority operation in about $G_{\mathrm{P,AP}}$01 at $G_{\mathrm{P,AP}}$02 (Wan et al., 2017). In a different MgO-based architecture, vertical current injection moved a domain wall step by step, producing memristive resistance changes and demonstrating why perpendicular injection is attractive for domain-wall devices (Chanthbouala et al., 2011).

Reference-layer reconfiguration adds a further degree of freedom. In MgO-based vortex MTJs with diameters from $G_{\mathrm{P,AP}}$03 to $G_{\mathrm{P,AP}}$04, reannealing the upper layer of a pinned synthetic antiferromagnet can transform the reference layer from a single-domain state into a vortex state with programmable core position. When the reference layer is vortex-like, the injected spin polarization becomes spatially nonuniform and supports field-free vortex oscillations in the free layer. For $G_{\mathrm{P,AP}}$05 pillars, zero-field oscillations appeared around $G_{\mathrm{P,AP}}$06, the frequency tuned from about $G_{\mathrm{P,AP}}$07 to $G_{\mathrm{P,AP}}$08, the reflected rf power rose from about $G_{\mathrm{P,AP}}$09 to $G_{\mathrm{P,AP}}$10, $G_{\mathrm{P,AP}}$11, and the TMR decreased from about $G_{\mathrm{P,AP}}$12 at $G_{\mathrm{P,AP}}$13 to about $G_{\mathrm{P,AP}}$14 at $G_{\mathrm{P,AP}}$15 (Stebliy et al., 14 Aug 2025).

Thermally assisted vertical switching constitutes yet another regime. In a $G_{\mathrm{P,AP}}$16 Curie-switch MTJ using a $G_{\mathrm{P,AP}}$17 weak-ferromagnet spacer with $G_{\mathrm{P,AP}}$18, the transport current through the tunnel junction heats the spacer above its Curie temperature, exchange-decouples the composite free layer, and enables reversal under a small bias field. Current densities of order $G_{\mathrm{P,AP}}$19 were sufficient for switching, and thermal modeling inferred an effective barrier area $G_{\mathrm{P,AP}}$20, indicating strong heat focusing by barrier hot spots (Kravets et al., 2017).

Finally, vertical MTJs support explicitly thermal readout modes. In the analytical treatment of tunneling magneto-thermopower, the thermopower $G_{\mathrm{P,AP}}$21 inherits the angular dependence of the spin-resolved conductance, and the tunneling magneto-thermopower can change sign as the barrier height is varied. For the Fe/Al$G_{\mathrm{P,AP}}$22O$G_{\mathrm{P,AP}}$23/Fe example treated there, the vanishing of the tunneling magneto-thermopower occurs when the barrier top lies about $G_{\mathrm{P,AP}}$24 above $G_{\mathrm{P,AP}}$25 for the stated parameters (López-Monís et al., 2013).

Taken together, these developments show that “vertical magnetic tunnel junction” no longer denotes a single device archetype. It denotes a transport geometry in which coherent tunneling, interfacial electronic structure, magnetic symmetry, and write physics can be combined in markedly different ways. The common architecture remains a magnetic electrode/barrier/magnetic electrode stack with out-of-plane transport, but the operative mechanisms now include $G_{\mathrm{P,AP}}$26 filtering, Bloch-state mismatch, interfacial uncompensation, magnetically controlled ferroelectricity, VCMA, field-like spin torque, thermo-magnetic decoupling, and spatially structured spin injection.