---
title: Vertical Magnetic Tunnel Junctions
url: https://www.emergentmind.com/topics/vertical-magnetic-tunnel-junctions
type: topic
---

# Vertical Magnetic Tunnel Junctions

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 [1111.6439][1904.06098][2506.12715].

## 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 \(g(E)\) [1111.6439][1310.7841].

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 \(5\ \mathrm{K}\) [1509.03525]. By contrast, crystalline barriers require a symmetry-resolved description: in-plane momentum \(\mathbf{k}_{\parallel}\) 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 [1111.6439][1904.06098].

This distinction is central to the modern theory of vertical MTJs. In MgO-based junctions the slowest-decaying evanescent channel has \(\Delta_1\) symmetry; in ScN-based junctions both \(\Delta_1\) at \(\Gamma\) and \(\Delta_2'\) at \(X\) contribute low-decay channels; in several van der Waals structures the dominant mechanism is not conventional symmetry filtering but \(\mathbf{k}_{\parallel}\)-resolved Bloch-state matching or mismatch between the spin channels of the electrodes [2008.12770][1904.06098]. 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 [1310.7841].

## 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 \(\Delta_1\) Bloch state in bcc Fe at the Fermi level and a MgO(100) complex band structure whose slowest-decaying evanescent state also has \(\Delta_1\) symmetry. This symmetry filtering strongly enhances parallel-state conductance and suppresses antiparallel transmission, establishing the modern coherent-tunneling picture for oxide MTJs [1111.6439].

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 \((100)\)-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 \((100)\)-textured and \((100)\)-epitaxial Fe/MgO/Fe show that structural coherence and magnetic quality at the \(25\ \mathrm{nm}\) 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
$$
G_{\mathrm{tot,P}}=\sum_i G_{i,\mathrm{P}},\qquad
G_{\mathrm{tot,AP}}=\sum_i G_{i,\mathrm{AP}}.
$$
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 [1111.6439].

Representative Fe/MgO/Fe transport behavior reported in the supplied synthesis is correspondingly strong. Room-temperature TMR of at least \(150\)–\(200\%\) is common in well-textured sputtered Fe/MgO/Fe, while epitaxial junctions and low-temperature measurements can exceed \(300\)–\(600\%\). For MgO thicknesses of about \(1.0\)–\(2.2\ \mathrm{nm}\), resistance-area products of about \(1\)–\(30\ \Omega\cdot\mu\mathrm{m}^2\) 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; \(dI/dV\) and \(G(V)\) frequently show a zero-bias anomaly and weak features at biases of hundreds of millivolts [1111.6439].

The same MgO platform also underlies vertical domain-wall manipulation. In an MgO-based MTJ with TMR \(\approx 65\%\) and \(RA\approx 3.5\ \Omega\cdot\mu\mathrm{m}^2\), vertical current injection moved a domain wall in the free layer at current densities of order \(4\)–\(5\times 10^6\ \mathrm{A\ cm^{-2}}\), about \(100\) 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 \(40\%\) of the in-plane torque at low bias and produced effective fields of about \(10\ \mathrm{Oe}\) at \(5\times 10^6\ \mathrm{A\ cm^{-2}}\) [1102.2106].

## 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, \(\Delta_1\) at \(\Gamma\) and \(\Delta_2'\) at \(X\), both with \(\kappa(E_F)\approx 0.07\ \mathrm{\AA^{-1}}\), compared with \(\kappa_{\Delta_1}(E_F)\approx 0.21\ \mathrm{\AA^{-1}}\) for MgO. At six barrier layers, the calculated Fe/ScN/Fe junction has \(G_{\mathrm{P}}=2.51\times 10^{-7}\ \mathrm{S}\), \(G_{\mathrm{AP}}=2.11\times 10^{-9}\ \mathrm{S}\), \(\mathrm{TMR}\approx 11{,}800\%\), and \(RA\approx 0.326\ \Omega\cdot\mu\mathrm{m}^2\), whereas the comparison Fe/MgO/Fe junction at the same thickness has \(\mathrm{TMR}\approx 30{,}600\%\) but \(RA\approx 10.5\ \Omega\cdot\mu\mathrm{m}^2\). The proposed significance is not a larger TMR than MgO, but high spin filtering at much lower resistance-area product [2008.12770].

A second route is crystallographic diversification. Fully epitaxial fcc(111) Co\(_{90}\)Fe\(_{10}\)/MgAlO/Co\(_{90}\)Fe\(_{10}\) junctions on Ru(0001) demonstrate that close-packed, polar \((111)\) barriers can be stabilized by interface engineering. At the CoFe(111)/MgAlO(111) interface, a \(5{:}6\) in-plane domain-matching structure with periodic dislocations reduces the effective mismatch to about \(-5\%\), and the resulting MTJs exhibit \(37\%\) TMR at room temperature and \(47\%\) at \(10\ \mathrm{K}\). Their \(RA\) is about \(2.7\times 10^6\ \Omega\cdot\mu\mathrm{m}^2\), and the differential conductance is highly symmetric with respect to bias polarity, indicating nearly identical upper and lower interface qualities [2308.04149].

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 \(E_F\), minority states remain available, and transport evolves from direct majority-spin \(\Delta_1\) tunneling to resonant minority-spin tunneling as the number of Fe/MgO repeats increases. In the ideal ballistic calculations, the TMR rises from \(\approx 10^5\) for the reference \(N=M=0\) junction to \(\approx 10^8\) for \(N=M=3\), while the same architecture simultaneously lowers the effective \(M_s\) and produces robust perpendicular magnetic anisotropy for most \(n[\mathrm{Fe}]/m[\mathrm{MgO}]\) multilayers examined [2502.20365].

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\(_3\)GeTe\(_2\)/graphene/Fe\(_3\)GeTe\(_2\) and Fe\(_3\)GeTe\(_2\)/h-BN/Fe\(_3\)GeTe\(_2\), the relaxed interface separation is about \(3.50\ \mathrm{\AA}\), much larger than the \(\mathrm{Fe\!-\!MgO}\) spacing of about \(2.16\ \mathrm{\AA}\), 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 \(2.60\times 10^{-2}\) and \(6.98\times 10^{-4}\) for the graphene-spaced parallel and antiparallel states, giving \(\mathrm{TMR}\approx 3621\%\); for h-BN the corresponding values are \(2.15\times 10^{-2}\) and \(3.38\times 10^{-4}\), giving \(\mathrm{TMR}\approx 6256\%\). The current on/off ratio exceeds \(10\) over a broad bias range, with a characteristic half-width of about \(0.04\ \mathrm{V}\) [1904.06098].

Experimentally, all-van-der-Waals MTJs can be much simpler and less ideal. A junction assembled by dry transfer from two exfoliated \(\mathrm{Fe_{0.25}TaS_2}\) flakes uses native \(\mathrm{Ta_2O_5}\) at the cleaved surfaces as the tunnel barrier. The junction area is \(67.8\ \mu\mathrm{m}^2\), \(RJA=580\ \mathrm{k}\Omega\cdot\mu\mathrm{m}^2\) at \(5\ \mathrm{K}\), and the TMR is about \(6\%\) at \(5\ \mathrm{K}\), decreasing with increasing \(|V|\) 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 [1509.03525].

A more elaborate van der Waals architecture combines tunneling readout with spin-orbit-torque writing. In the monolayer stack \(1T\ \mathrm{MoS_2}/1T\ \mathrm{VSe_2}/1H\ \mathrm{MoS_2}/1T\ \mathrm{VSe_2}/1T\ \mathrm{MoS_2}\), the semiconducting \(1H\ \mathrm{MoS_2}\) monolayer acts as the tunnel barrier while metallic \(1T\ \mathrm{MoS_2}\) supplies spin Hall conductivity. The equilibrium TMR at \(300\ \mathrm{K}\) is \(22\%\), but under bias the antiparallel resonance is shifted out of the transport window and the non-equilibrium TMR reaches \(846\%\) at \(0.5\ \mathrm{V}\) and \(300\ \mathrm{K}\), or \(1277\%\) at \(100\ \mathrm{K}\). The intrinsic spin Hall conductivity of \(1T\ \mathrm{MoS_2}\) is \(-34\ (\hbar/e)\ \mathrm{S/cm}\) at the Fermi level and can be increased up to about \(400\ (\hbar/e)\ \mathrm{S/cm}\) at \(E=-2.0\ \mathrm{eV}\) [1904.07499].

Van der Waals materials also make possible a “one-material” MTJ. In multilayer MnPX\(_3\) (\(X=\mathrm{S,Se}\)), 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 \(5.5\times 10^{-3}\) carriers per atom for Li-like \(n\)-doping and \(4.7\times 10^{-3}\) per atom for F-like \(p\)-doping, equivalent to surface charge densities of about \(1.5\times 10^{13}\ \mathrm{cm^{-2}}\) and \(1.3\times 10^{13}\ \mathrm{cm^{-2}}\), respectively. The resulting spin-filtering picture is idealized, and the stated TMR is \(100\%\) in the pessimistic definition used there [2204.14262].

Multiferroic functionality can be implemented in both oxide and van der Waals vertical junctions. An epitaxial \(\mathrm{La_{0.7}Sr_{0.3}MnO_3}/\mathrm{Hf_{0.5}Zr_{0.5}O_2}/\mathrm{Co}\) junction with a \(2\ \mathrm{nm}\) ferroelectric HZO barrier exhibits TMR of \(5\)–\(7\%\) at \(-0.2\ \mathrm{V}\) and \(50\ \mathrm{K}\), inverse TMR of \(-2.6\%\) at \(+0.2\ \mathrm{V}\), resistance-area products above \(600\ \mathrm{M}\Omega\cdot\mu\mathrm{m}^2\), and four non-volatile resistance states generated by combining magnetic and ferroelectric switching. The ON/OFF ratio is about \(440\%\), and the ferroelectric-state TER in the TMR measurement configuration is about \(190\%\) [1902.08021]. In the theoretical \(\mathrm{CrBr_3/MnPSe_3/CrBr_3}\) van der Waals multiferroic junction with asymmetric PtTe\(_2\) and alkali-doped/intercalated \(\mathrm{CrBr_3}\) electrodes, the out-of-plane ferroelectric polarization is \(0.33\ \mathrm{pC/m}\), the zero-bias TMR exceeds \(8.1\times 10^5\%\), TER reaches \(2499\%\), and under bias they increase to \(3.6\times 10^7\%\) and \(9990\%\), respectively; a peak-to-valley ratio of \(9.55\times 10^9\%\) is reported for the negative-differential-resistance response [2501.00761].

## 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\(_2\)/TiO\(_2\)/CrO\(_2\) design, CrO\(_2\) is the sole ferromagnetic electrode, while RuO\(_2\) is an antiferromagnetic altermagnet. In the \((110)\)-oriented junction, RuO\(_2\) provides a spin-polarized conduction-channel distribution and the calculated TMR is about \(965\%\); in the \((001)\) orientation, the mechanism is different and relies on a Néel spin current combined with sublattice-dependent barrier thickness, producing about \(125\%\) TMR for a \(7\)-monolayer TiO\(_2\) barrier and exceeding \(1000\%\) at \(E_F+0.3\ \mathrm{eV}\) [2310.02139].

A second antiferromagnetic route is purely interfacial. In the interface-controlled A-type AFMTJ based on Fe\(_4\)GeTe\(_2\), the bulk electrode is spin-degenerate by \(PT\) symmetry, but the surface termination is magnetically uncompensated and generates a finite interfacial spin polarization. In a semi-infinite FGT/BN/PtTe\(_2\) structure, the transmission spin polarization at the Fermi level is about \(84\%\) despite spin-degenerate bulk bands. In full PtTe\(_2\)/[FGT]\(_n\)/BN/[FGT]\(_m\)/PtTe\(_2\) junctions with \(n=m=2\), the TMR is about \(-300\%\) around \(E_F\), remaining strongly negative over a broad energy range and decreasing to about \(-40\%\) at \(0.1\ \mathrm{V}\) 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 [2506.12715].

This interfacial picture is now experimentally realized in all-collinear van der Waals AFMTJs based on \((\mathrm{Fe}_{0.6}\mathrm{Co}_{0.4})_5\mathrm{GeTe}_2/\mathrm{WSe_2}/(\mathrm{Fe}_{0.6}\mathrm{Co}_{0.4})_5\mathrm{GeTe}_2\). The maximum TMR reaches \(75\%\) at \(10\ \mathrm{K}\), and the large peaks occur at \(\pm 0.9\ \mathrm{T}\), well inside the antiferromagnetic regime and distinctly separated from the AFM-to-FM transition fields of roughly \(\pm 1.8\) to \(\pm 3.0\ \mathrm{T}\). In an even-even device, the two characteristic interfacial coercive fields are \(B_{c1}\approx 0.84\ \mathrm{T}\) and \(B_{c2}\approx 0.98\ \mathrm{T}\) at \(10\ \mathrm{K}\), both vanishing near \(230\ \mathrm{K}\). By controlling even- or odd-layer parity in the electrodes, the devices select volatile or non-volatile TMR behavior; in odd-even devices, \(B_{c1}\) crosses zero near \(80\ \mathrm{K}\), and above that temperature the TMR becomes non-volatile without a dedicated pinning layer [2507.12735].

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 \(160\ \mathrm{nm}\) pillars achieved \(P_{\mathrm{sw}}\approx 98\%\) with a \(2.04\ \mathrm{V}\), \(0.6\ \mathrm{ns}\) pulse and a switching energy of \(3.54\ \mathrm{fJ/bit}\). The same remote-doped sample retained TMR of \(160\)–\(165\%\) after \(400^\circ\mathrm{C}\) annealing and showed room-temperature thermal stability \(\Delta\approx 32\) [2511.18143].

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 \(\approx 110\%\) and \(RA\approx 10\ \Omega\cdot\mu\mathrm{m}^2\), but the patterned \(70\ \mathrm{nm}\) pillars suffered degraded device-level TMR of about \(2.5\)–\(15\%\) because of sidewall “fencing” during fabrication. Micromagnetic simulations nevertheless showed that a downscaled cross-shaped free layer with \(10\ \mathrm{nm}\) arm width and lateral size \(L=90\ \mathrm{nm}\) can perform majority operation in about \(3\ \mathrm{ns}\) at \(J=6\times 10^{11}\ \mathrm{A\ m^{-2}}\) [1711.03609]. 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 [1102.2106].

Reference-layer reconfiguration adds a further degree of freedom. In MgO-based vortex MTJs with diameters from \(400\) to \(1000\ \mathrm{nm}\), 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 \(1\ \mu\mathrm{m}\) pillars, zero-field oscillations appeared around \(I\approx +8\ \mathrm{mA}\), the frequency tuned from about \(70\) to \(120\ \mathrm{MHz}\), the reflected rf power rose from about \(-70\) to \(-40\ \mathrm{dBm}\), \(RA\approx 5\ \Omega\cdot\mu\mathrm{m}^2\), and the TMR decreased from about \(150\%\) at \(0.1\ \mathrm{mA}\) to about \(80\%\) at \(10\ \mathrm{mA}\) [2508.10739].

Thermally assisted vertical switching constitutes yet another regime. In a \(3\ \mu\mathrm{m}\) Curie-switch MTJ using a \(\mathrm{Ni_{72}Cu_{28}}(20\ \mathrm{nm})\) weak-ferromagnet spacer with \(T_C^f\approx 350\ \mathrm{K}\), 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 \(1\ \mathrm{mA}/\mu\mathrm{m}^2\) were sufficient for switching, and thermal modeling inferred an effective barrier area \(A_{\mathrm{TB}}^{\mathrm{eff}}\approx 0.035A_{\mathrm{cs}}\), indicating strong heat focusing by barrier hot spots [1710.06311].

Finally, vertical MTJs support explicitly thermal readout modes. In the analytical treatment of tunneling magneto-thermopower, the thermopower \(S(\phi)\) 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\(_2\)O\(_3\)/Fe example treated there, the vanishing of the tunneling magneto-thermopower occurs when the barrier top lies about \(0.7\ \mathrm{eV}\) above \(E_F\) for the stated parameters [1310.7841].

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 \(\Delta_1\) filtering, Bloch-state mismatch, interfacial uncompensation, magnetically controlled ferroelectricity, VCMA, field-like spin torque, thermo-magnetic decoupling, and spatially structured spin injection.

Source: https://www.emergentmind.com/topics/vertical-magnetic-tunnel-junctions