Reliability physics of scaled MgO-barrier magnetic tunnel junctions: write-error statistics, barrier endurance under repeated current-induced switching, and the microscopic origin of bias-dependent TMR quenching

A 2007 review of MgO-barrier magnetic tunnel junctions (MTJs) for spintronic memories (Ikeda et al., IEEE Trans. Electron Devices 54, 991, 2007) identified five linked further-work issues: reducing the intrinsic switching current density toward MOSFET-compatible values, establishing MgO barrier endurance and failure modes under repeated current-induced switching, understanding stochastic switching statistics at short pulse durations, explaining the bias-voltage dependence of the tunnel magnetoresistance (TMR), and wafer-scale CMOS integration. Later literature has substantially closed the first and last of these: interfacial perpendicular anisotropy in Ta/CoFeB/MgO stacks (Ikeda et al., Nat. Mater. 2010) and foundry-level production of embedded spin-transfer-torque MRAM at 28 nm and below. What survives as a precise open core is the reliability physics of scaled MgO/CoFeB MTJs: a predictive, material-parameter-level description and control of write-error (switching-probability) distributions at short pulses, endurance and failure modes of MgO barriers beyond the demonstrated 10^12-cycle class toward effectively unlimited cycling, and a first-principles account of TMR quenching under bias voltage in practical devices.

Background

A magnetic tunnel junction (MTJ) is a two-terminal device consisting of two ferromagnetic metal electrodes separated by a ~1 nm insulating barrier; its resistance is low or high depending on whether the two magnetizations are parallel or antiparallel. The resistance change is quantified by the tunnel magnetoresistance (TMR) ratio. Sputtered MgO barriers with Co–Fe–B electrodes, crystallized by annealing into a (001)-textured bcc structure, produce giant TMR (room-temperature ratios of 361% and 500% were reported in the 2007 review by Ikeda and co-workers) because coherent electronic states of the $\Delta_1$ symmetry class — the bcc band-structure symmetry whose wave functions match the slowly decaying, symmetry-compatible channel of the MgO(001) barrier — tunnel across the junction with little scattering while other symmetry channels are filtered out. Writing the memory state is done by spin-transfer torque: a spin-polarized current passing through the junction exerts a torque on the free-layer magnetization and switches it, which is scalable in a way that external-field writing is not. The intrinsic (zero-temperature) critical current density for switching, $J_{C0}$, ranged from roughly $10^{6}$ to $10^{8}\ \mathrm{A/cm^2}$ across early devices, with the 2007 review's best synthetic-ferrimagnetic free layer — two antiferromagnetically coupled ferromagnetic films acting as a single free layer — reaching $J_{C0}=8.7\times10^{6}\ \mathrm{A/cm^2}$ at a thermal-stability factor $E/k_B T=67$ ($E/k_B T$ is the ratio of the magnetic anisotropy energy barrier $E$ to thermal energy, which must be ~40–60 for ten-year data retention); because a scaled MOS transistor can only deliver currents corresponding to roughly $10^{6}\ \mathrm{A/cm^2}$, the 2007 review posed reduction of $J_{C0}$ toward or below $10^{6}\ \mathrm{A/cm^2}$ at practical pulse widths while preserving high $E/k_B T$ as a central requirement. In 2010 the same group showed that the interfacial perpendicular magnetic anisotropy of the CoFeB/MgO interface itself (Ta/CoFeB/MgO/CoFeB/Ta stacks) supplies the out-of-plane easy axis: this simultaneously gave TMR above 120%, stability at 40 nm diameter, and a switching current of 49 μA (Ikeda et al., Nat. Mater. 9, 721, 2010), and the perpendicular route became the basis of production spin-transfer-torque MRAM (STT-MRAM), now a standard foundry embedded-memory offering at the 28 nm node and below (Apalkov, Dieny and Slaughter, Proc. IEEE 2016; Worledge, IRPS 2023). This closes the engineering aims of low-current writing and wafer-scale CMOS integration, but it exposes the surviving scientific core of the 2007 agenda, which concerns how reliably and how far the writing and reading physics scales. Three issues remain. First, at practical (nanosecond) pulse widths, switching is a stochastic process: the switching probability versus pulse current, duration, and polarity is measured as a write-error rate, which is the key reliability metric of STT-MRAM; the standard thermal-activation (Arrhenius) and macrospin models — in which the free layer is treated as a single rigid magnetic moment — are known to break down in this regime, and recent work shows that empirical fits to the measured distributions no longer yield parameters with material meaning. Second, the MgO barrier must survive repeated current-induced switching; demonstrated endurance sits in the $10^{5}$–$10^{12}$-cycle class depending on the application specification, the failure modes are barrier- and process-dependent, and the original target of establishing endurance beyond $10^{15}$ cycles with identified and controlled failure mechanisms has not been met. Third, the TMR ratio falls rapidly with read bias voltage (TMR quenching), which limits the read signal at the low resistance-area (RA) products (the product of junction resistance and junction area, the stack figure of merit that for a given current sets the read voltage) needed for dense, low-voltage devices; magnon (spin-wave) excitation and other inelastic tunneling channels have been shown to contribute, but a complete microscopic account for practical CoFeB/MgO junctions is still missing.

Problem Statement

For sputtered MgO-barrier magnetic tunnel junctions with CoFeB-based electrodes of the kind introduced for spintronic memories, establish the reliability physics that governs how far current-induced writing and tunneling readout can be scaled. Specifically, as posed in the 2007 review of Ikeda and co-workers and narrowed by subsequent progress: (1) determine, with predictive rather than empirical accuracy, the probability distributions of current-induced magnetization switching versus pulse magnitude, polarity, and duration at the practical nanosecond pulse widths where thermal-activation and macrospin models break down, so that write-error rates can be engineered from material and device parameters; (2) establish the endurance and the dominant failure mechanisms of MgO tunnel barriers under repeated current-induced switching, extending demonstrated cycle counts beyond the 10^5-10^12 class toward effectively unlimited endurance with identified and controlled failure modes; and (3) establish the microscopic origin of the bias-voltage dependence of the tunnel magnetoresistance, explaining TMR quenching under bias in practical CoFeB/MgO devices and how it limits read margin at the low resistance-area products required for dense, low-voltage operation. These three items share one scientific objective — quantifying and controlling the physics that sets the ultimate reliability and scalability of MgO-based spintronic memory — and a completion boundary is recognizable when each item's stated target is met at the level of mechanism rather than empirical fit: for (1), a switching-probability description that predicts write-error distributions from material and device parameters without per-device re-fitting; for (2), demonstrated endurance beyond the 10^15-cycle target of the 2007 review with the dominant MgO failure modes identified and controlled; and for (3), a microscopic account of bias-dependent TMR quenching that explains the measured quenching in practical CoFeB/MgO devices and its read-margin limit. The two remaining 2007 issues (reduction of the intrinsic switching current density $J_{C0}$ toward MOSFET-compatible values, and wafer-scale CMOS integration of high-TMR stacks) are documented as substantially achieved by 2010-2023 literature and are part of the problem's lineage, not of its surviving core.

References

1. S. Ikeda, J. Hayakawa, Y. M. Lee, F. Matsukura, Y. Ohno, T. Hanyu, H. Ohno, "Magnetic Tunnel Junctions for Spintronic Memories and Beyond," IEEE Transactions on Electron Devices 54, 991-1002 (2007). DOI: 10.1109/ted.2007.894617, https://doi.org/10.1109/ted.2007.894617
2. S. Ikeda, K. Miura, H. Yamamoto, K. Mizunuma, H. D. Gan, M. Endo, S. Kanai, J. Hayakawa, F. Matsukura, H. Ohno, "A perpendicular-anisotropy CoFeB-MgO magnetic tunnel junction," Nature Materials 9, 721-724 (2010). DOI: 10.1038/nmat2804, https://doi.org/10.1038/nmat2804
3. Y. J. Lee, G. Jan, Y. J. Wang, K. Pi, T. Zhong, R. Y. Tong, V. Lam, J. Teng, K. Huang, R. R. He, S. Le, T. Torng, J. DeBrosse, T. Maffitt, C. Long, W. J. Gallagher, P. K. Wang, "Demonstration of chip level writability, endurance and data retention of an entire 8Mb STT-MRAM array," 2013 International Symposium on VLSI Technology, Systems and Application (VLSI-TSA) (2013). DOI: 10.1109/vlsi-tsa.2013.6545595, https://doi.org/10.1109/vlsi-tsa.2013.6545595
4. G. Jan, L. Thomas, S. Le, Y.-J. Lee, H. Liu, J. Zhu, R. Tong, K. Pi, Y.-J. Wang, D. Shen, R. He, J. Haq, J. Teng, V. Lam, K. Huang, T. Zhong, T. Torng, P.-K. Wang, "Demonstration of fully functional 8Mb perpendicular STT-MRAM chips with sub-5ns writing for non-volatile embedded memories," 2014 Symposium on VLSI Technology (2014). DOI: 10.1109/vlsit.2014.6894357, https://doi.org/10.1109/vlsit.2014.6894357
5. D. Apalkov, B. Dieny, J. M. Slaughter, "Magnetoresistive Random Access Memory," Proceedings of the IEEE 104, 1796-1830 (2016). DOI: 10.1109/jproc.2016.2590142, https://doi.org/10.1109/jproc.2016.2590142
6. S. Bhatti, R. Sbiaa, A. Hirohata, H. Ohno, S. Fukami, S. N. Piramanayagam, "Spintronics based random access memory: a review," Materials Today 20, 530-548 (2017). DOI: 10.1016/j.mattod.2017.07.007, https://doi.org/10.1016/j.mattod.2017.07.007
7. C.-Y. Wang, M.-C. Shih, C.-H. Weng, C.-H. Chen, C.-Y. Chang, W. Wang, T.-W. Chiang, A. Hung, H. Chuang, W. J. Gallagher, "Reliability Demonstration of Reflow Qualified 22nm STT-MRAM for Embedded Memory Applications," 2020 IEEE Symposium on VLSI Technology (2020). DOI: 10.1109/vlsitechnology18217.2020.9265054, https://doi.org/10.1109/vlsitechnology18217.2020.9265054
8. J. Z. Sun, "Spin-transfer torque switching probability of CoFeB/MgO/CoFeB magnetic tunnel junctions beyond macrospin," Physical Review B 104, 104428 (2021). DOI: 10.1103/physrevb.104.104428, https://doi.org/10.1103/physrevb.104.104428
9. D. C. Worledge, "Write-error-rate of Spin-Transfer-Torque MRAM (Invited)," 2023 IEEE International Reliability Physics Symposium (IRPS) (2023). DOI: 10.1109/irps48203.2023.10117666, https://doi.org/10.1109/irps48203.2023.10117666
10. V. Drewello, J. Schmalhorst, A. Thomas, G. Reiss, "Evidence for strong magnon contribution to the TMR temperature dependence in MgO based tunnel junctions," Physical Review B 77, 014440 (2008). DOI: 10.1103/physrevb.77.014440, https://doi.org/10.1103/physrevb.77.014440
11. Q. L. Ma, S. G. Wang, H. X. Wei, H. F. Liu, X.-G. Zhang, X. F. Han, "Evidence for magnon excitation contribution to the magnetoresistance behavior during thermal annealing in CoFeB/MgO/CoFeB magnetic tunnel junctions," Physical Review B 83, 224430 (2011). DOI: 10.1103/physrevb.83.224430, https://doi.org/10.1103/physrevb.83.224430
12. K. Li, X. Yang, J. Gong, S. He, "Realizing high-performance, enhanced write endurance of low-RA STT-MRAM through MgO tunnel barrier engineering," Journal of Semiconductors 47, 042302 (2026). DOI: 10.1088/1674-4926/25080016, https://doi.org/10.1088/1674-4926/25080016