MRAM cells are built around an MTJ, which consists of a pinned RL and an FL, separated by a thin insulating tunnel barrier (TB), typically MgO. The magnetization orientation of both the FL and RL can either lie in-plane or
perpendicular to the film plane due to PMA, as illustrated in Figure 2.3. When both magnetic layers exhibit perpendicular anisotropy, the device is referred
to as a perpendicular magnetic tunnel junction (pMTJ), while in-plane magnetized structures are called in-plane magnetic tunnel junction (iMTJ).
The first MTJs used amorphous Al2O3 barriers and achieved modest TMR values in the double digits. However, theoretical predictions [18, 19] and experimental
demonstrations [39, 70] of crystalline MgO-based MTJs showed a dramatic increase in TMR, reaching up to 600 % at room-temperature under optimized laboratory conditions. Today, typical TMR
values in perpendicularly magnetized MTJs range between 100 % and 200 % [71].
Figure 2.3: Illustration of magnetization orientations in MTJs, showing the difference between in-plane (iMTJ) and perpendicular (pMTJ) configurations.
Data retrieval from MTJs is performed by applying a small bias voltage (typically 0.1-0.2 V) and measuring the resulting resistance across the junction. Operating at such low bias
avoids disturbing the stored magnetization state and ensures that the TMR remains high. The TMR is highest at low bias and decreases with increasing voltage [49].
The thermal stability of MRAM cells, critical for ensuring data retention over years, is characterized by the dimensionless factor:
where \(\mathcal {E}_\mathrm {B}\) is the energy barrier separating the P and AP states for out-of-plane magnetized films, \(V\) is the volume of the free layer, \(k_{\mathrm {B}}\) is the Boltzmann constant, \(T\) the
absolute temperature, and \(H_k = 2K_\mathrm {eff}/(\mu _0 M_s)\) is the effective anisotropy field. The effective anisotropy \(K_\mathrm {eff} = K_u - \mu _0 M_s^2/2\) accounts for the reduction of the intrinsic uniaxial
anisotropy \(K_u\) by the demagnetizing field \(H_d = M_s\) of the thin film. To ensure reliable retention over ten years, \(\Delta \) values of 40–60 are typically required [72, 73].
where \(\mu _0\) is the vacuum permeability, \(M_{\mathrm {S}}\) the saturation magnetization, \(H_K\) the anisotropy field, and \(V\) the volume of the FL. Optimization of \(H_K\) and the FL geometry are thus key to
MRAM scaling [28].
STT was independently proposed by Slonczewski and Berger in 1996 [18, 19], enabling purely electrical switching in MTJs. The first experimental demonstration followed in metallic spin valves [39] and later
in MTJs [70].
Electrons flowing from the RL to the FL generate a spin-polarized current. Due to exchange interaction, spins align with the magnetization of the FL, and conservation of angular momentum causes the transverse spin component to
exert a torque on the FL, known as STT [18, 19]. Reversing current direction induces AP switching via reflected minority spins [35]. Since STT depends on transverse spin components, no torque is exerted
when the magnetizations are collinear, leading to an incubation delay. Thermal fluctuations are then needed to initiate switching, and if the current pulse duration is shorter than the delay, write errors may
occur [74, 75].
where \(A\) is the cross-sectional area, \(t_{\mathrm {FL}}\) is the FL thickness, \(\alpha \) is the Gilbert damping factor, and \(\eta \) is the STT efficiency parameter [72, 76].
In addition to the switching current, data retention in STT-MRAM is governed by the thermal stability factor:
where \(\mathcal {E}_{\mathrm {B}}\) is the energy barrier separating stable magnetization states P or AP, \(k_B\) is the Boltzmann constant, and \(T\) is the absolute temperature. A larger \(\Delta \) ensures long-term
retention by suppressing the thermally activated switching. However, increasing \(\Delta \) also raises the required switching current, creating a trade-off between writability and data retention.
Early MRAM relied on in-plane magnetic anisotropy (IMA), with shape anisotropy in elliptical FLs [11, 77]. In these devices, switching involves out-of-plane precession, as the reversal trajectory does not follow
thermal fluctuations directly [78]. In-plane devices were typically larger than 60 nm to maintain thermal stability, which limited scaling [79]. Furthermore, both the
anisotropy field \(H_K\) and the demagnetizing field \(H_d\) must be overcome for switching, increasing energy requirements [80]:
Additionally, stronger in-plane demagnetizing fields oppose magnetization changes, necessitating higher switching currents [72]. These drawbacks hindered scalability and energy efficiency.
PMA structures, dominated by interfacial anisotropy (e.g., CoFeB/MgO), mitigate these challenges by eliminating \(H_d\) and orienting magnetization perpendicular to the film plane, allowing for more energy-efficient STT
switching [81]. PMA devices also support cylindrical layouts with better scaling and enhanced thermal stability. The corresponding critical current density simplifies to:
Ikeda et al. [28, 53] demonstrated robust PMA in MgO/CoFeB stacks, a pivotal advancement in the development of scalable and high-performance MRAM [82]. Annealing also
significantly impacts both TMR and anisotropy optimization [83].
Further improvements arise from engineering multilayered FLs, SAF RLs [84, 85, 86], and tailored interfaces. SAF designs use two antiferromagnetically coupled ferromagnetic layers (often separated by Ru)
to cancel dipolar stray fields on the FL, enhancing write symmetry and error tolerance [34]. Advanced strategies include ionic liquid gating to modulate Rudderman–Kittel–Kasuya–Yosida (RKKY) coupling [87] and
re-doped Ru spacers for thermal robustness during CMOS backend processing [88].
