Thanks to their unique set of assets (non-volatility, speed, density, endurance), STT-MRAM are seen as a unique candidate for DRAM and/or Cache SRAM replacement allowing to drastically reduce the power consumption of electronic circuits thanks to new power gating strategies made possible by the non-volatility of these memories. For these applications, fast write speed (in the range 0.3ns-5ns) with low bit error rate are required.

A well-known problem concerning the switching speed in conventional STT-MRAM is that of stochasticity of the storage layer magnetization switching. Indeed, the spin transfer torque (STT) is proportional to the cross product (M^P) of the storage layer magnetization M and of the current polarization which is parallel to the reference layer magnetization P. It turns out that in conventional STT-MRAM embodiments, either in-plane or out-of-plane magnetized, the two magnetizations M and P are either parallel or antiparallel at rest which means that when the write current is switched on, the STT is initially zero. One has to wait for a thermal fluctuation to create an initial angle between the storage layer magnetization and the current polarization which then triggers the storage layer magnetization reversal. Because by essence thermal fluctuations randomly appear, this yields a non-reproducibility in the switching as illustrated in Fig.1. As a result, to achieve a low bit error rate, it is necessary to wait a sufficiently long time to make sure that a large enough thermal fluctuation has happened to trigger the magnetization reversal. This is detrimental for the switching speed. To circumvent this issue, new STT-MRAM stack designs were investigated in this WP. New concepts of MTJ stacks were proposed and studied allowing ultrafast magnetic switching (100ps) with lowest energy ever reported in this kind of systems (90fJ).

*Fig.1 : Illustration of the stochastic switching observed in in-plane magnetized STT-MRAM (from Devolder et al, Phys. Rev. Lett.100 (2008))*

Our most promising 2-terminal stacks for ultrafast switching memories are illustrated in Fig.2. They comprise a storage layer sandwiched between two orthogonal polarizing layers: a perpendicular polarizer and an in-plane analyzer. This concept was patented by SPINTEC in 2000 and much further investigated within HYMAGINE as explained below. It can be used either with in-plane magnetized storage layer (possibly in combination with reduced demagnetizing field to lower the switching current) or with out-of-plane magnetized layer. In this study, we mostly focused on in-plane magnetized storage layer with reduced demagnetizing field thanks to the perpendicular anisotropy existing at CoFe/MgO interfaces. In these structures, the storage layer magnetization is submitted to two STT contributions: one from the perpendicular polarizer, the other from the in-plane analyzer. These two contributions tend to induce very different magnetization dynamics in the storage layer as illustrated in Fig.2. Above a certain current threshold proportional to the in-plane anisotropy, the perpendicular polarizer tends to induce an out-of-plane steady state precessional motion of the storage layer (SL) magnetization. In contrast, the in-plane analyzer tends to induce a stochastic bipolar switching, the final state being controlled by the current direction throughout the stack.

*Fig.2: Concept of ultrafast STT-MRAM comprising a storage free layer sandwiched between two polarizing layers of orthogonal magnetization.*

The actual storage layer switching dynamics in these structures which are sometimes called OST-MRAM (orthogonal spin-transfer MRAM), depends on the relative amplitude of the two STT contributions.

If the STT from the perpendicular polarizer dominates, due to the induced precessional motion of the magnetization, one may expect an oscillatory probability of switching of the SL magnetization as a function of the current pulse width. Since the frequency of the oscillations is typically in the range of a few GHZ, the period of the switching probability oscillations is in the range of a few 100ps. This implies that properly controlling the writing in such OST-MRAM would require controlling the current pulse duration with an accuracy of the order of 100ps. This is possible to achieve at the single bit scale but very difficult at the scale of an entire MRAM chip. It is therefore desirable not to work in this regime where the STT from the perpendicular polarizer dominates.

If the STT from the in-plane analyzer dominates, a stochastic bipolar switching is expected wherein the final state is controlled by the current direction throughout the stack but is independent on the pulse duration provided the pulse is long enough and the current density large enough. However, this switching is stochastic if this STT contribution too much dominates the STT contribution from the perpendicular polarizer.

**It is therefore desirable to be able to tune the relative influence of these two STT contributions so that the final state is controlled by the current direction through the stack whereas the perpendicular polarizer still provides the initial impulse on the magnetization which allows to make the switching faster and more deterministic.** Within HYMAGINE, we demonstrated that **this tuning can be achieved by playing with the cell aspect ratio** which influences the in-plane shape anisotropy. This shape anisotropy turns out to have a strong influence on the magnetization dynamics induced by the perpendicular polarizer whereas it has a weak influence on the magnetization dynamics induced by the in-plane analyzer as explained in Fig.3.

*Fig.3: Explanation of the influence of cell aspect ratio on the switching dynamics. Increasing the cell aspect ratio favors the STT contribution of the in-plane analyzer relatively to that of the out-of-plane polarizer. Indeed, the critical current to trigger the magnetization switching due to STT from in-plane analyzer is almost independent on the in-plane anisotropy because it is mostly controlled by the demagnetizing energy. In contrast, the critical current to trigger the out-of-plane precession of magnetization due to STT from perpendicular polarizer is directly proportional to the in-plane anisotropy. Therefore tuning the in-plane anisotropy by playing with the cell aspect ratio appears to be an efficient way to vary the relative amplitude of the two STT contributions*

The possibility to tune the relative influence of the two STT contributions from perpendicular polarizer and in-plane analyzer was demonstrated experimentally within HYMAGINE as described below. The samples are magnetic tunnel junctions comprising synthetic antiferromagnetic (SyAF) storage layer and polarizing/analyzing layers. The purpose of using SyAF structure is to minimize the magnetostatic interactions between layers. The structures were deposited and patterned at SPINTEC. Cells of various aspect ratio ranging from 1.5 to 5 were realized and studied.

*Fig.4: Typical stacks deposited and patterned at SPINTEC for this HYMAGINE study*

Below, a comparison of the switching dynamics in these structures with low aspect ratio (AR=2) and high aspect ratio (AR=5) is presented.

__Case of low aspect ratio (AR=2, elliptical cell 90nm*180nm) : __

In this case, we expect a dominant role of the STT from the perpendicular polarizer yielding an oscillatory behaviour of the switching probability as a function of pulse duration. This was indeed observed.

Fig.5 shows real time measurements of transmitted voltage across the MTJ versus time following the onset of the voltage pulse. The experiments were repeated for various amplitudes of the voltage pulses. An oscillatory behaviour of the transmitted voltage is observed due to the combination of the precessional dynamics of the storage layer magnetization and tunnel magnetoresistance of the MTJ. The higher the voltage, the higher the precession frequency which is consistent with our earlier theoretical study (Lee et al, APL 86, 022505 (2005)).

*Fig.5: Real time measurements of transmitted voltage across the MTJ for various amplitudes of the excitation voltage. Upper right: precession frequency versus current density and linear fit according to the formula from Lee et al, APL 86, 022505 (2005).*

Fig.6 shows the theoretical phase diagram in the (pulse duration, current density throughout the MTJ) plane (a) as well as the experimentally obtained switching probability versus pulse duration (b). This curve is obtained by averaging the real time oscillations observed in Fig.5 over 100 successive traces. The damping of the oscillations of probability is due to a gradual decoherence of the precessional motion caused by the thermal fluctuations in the storage layer magnetization. As expected, Fig.6b demonstrates that ultrafast switching in STT-MRAM is possible (here switching in 500ps corresponding to the position of the first peak). However, achieving this switching with 100% probability requires controlling the pulse duration with an accuracy of ±100ps which is very difficult to achieve at chip level.

*Fig.6: (a) Switching phase diagram in the (current, pulse duration) plane. (b) Switching probability versus pulse duration for the structure shown in Fig.4 and patterned at dimension of 90nm*180nm. (blue=100% probability of ending in the P state, orange=100% probability of ending in the AP state)*

__Case of high aspect ratio (AR=5, elliptical cell 50nm*250nm) : __

In this more favourable case in terms of practical use in STT-MRAM chip, the STT influence from the in-plane analyser dominates the STT influence from the out-of-plane polarizer. Fig.7 shows the expected behaviour of the switching dynamics for the parallel (P) to antiparallel (AP) switching and for the AP to P switching (P and AP refer here to the relative orientation of the storage layer and in-plane analyser magnetizations). Fig.7 shows that there indeed exists an interval of current density wherein one can switch deterministically to a final state which is controlled by the current direction independently of the pulse duration. However, it is also observed that if the current is too large, an oscillatory behaviour of the switching probability is recovered which must be avoided.

*Fig.7: Theoretical phase diagrams similar to that of Fig.6a showing the switching probability as a function of current density and pulse duration (blue=100% probability of ending in the P state, orange=100% probability of ending in the AP state). Left: Starting from the P state. Right: Starting from the AP state.*

This expected behaviour was confirmed experimentally as shown in Fig.8. Fig.8 shows the switching probability as a function of pulse duration for various voltage pulse amplitude. A non-oscillatory switching is observed. The larger the voltage pulse amplitude, the faster the switching. Here **switching in sub-ns regime could be achieved with power consumption as low as 90fJ per switching event.**

*Fig.8: Switching probability as a function of pulse duration measured for various voltage pulse amplitude on the patterned stack of Fig.4 with dimensions 50nm*250nm*

Besides being fast and deterministic, the switching of the storage layer magnetization is also very reproducible under these conditions. This is illustrated in Fig.9 which shows a screen capture of 1000 traces on a digital oscilloscope of the transmitted signal across the MTJ of Fig.4 submitted to pulse of voltage 891mV in amplitude. The intense red contrast indicates that all these traces fall on top of each other indicating that the switching is very reproducible.

*Fig.9: Superposition of 1000 traces of transmitted voltage across the MTJ of Fig.4 (collaboration with T.Devolder at IEF).*

**The possibility to achieve sub-ns deterministic writing in these STT-MRAMs is quite promising for their use in SRAM-type applications.**

**Publication associated with WP2 :**

**Spin transfer torque switching assisted by thermally induced anisotropy reorientation in perpendicular magnetic tunnel junctions**

Bandiera, S., R.C. Sousa, M. Marins de Castro Souza, C. Ducruet, C. Portemont, S. Auffret, L. Vila, I.L. Prejbeanu, B. Rodmacq and B. Dieny

Applied Physics Letters **99** (2011) 202507

http://dx.doi.org/10.1063/1.3662971

**Parametric oscillator based on non-linear vortex dynamics in low resistance magnetic tunnel junctions**

Martin, S., N. De Mestier, C. Thirion, C. Hoarau, Y. Conraux, C. Baraduc and B. Dieny

Physical Review B **84** (2011) 14434

http://journals.aps.org/prb/pdf/10….

**Improved coherence of ultrafast spin-transfer-driven precessional switching with synthetic antiferromagnet perpendicular polarizer**

Vaysset, A., C. Papusoi, L.D. Buda-Prejbeanu, S. Bandiera, M. Marins de Castro Souza, Y. Dahmane, J.-C. Toussaint, U. Ebels, S. Auffret, R.C. Sousa, L. Vila and B. Dieny

Applied Physics Letters **98** (2011) 242511

http://dx.doi.org/10.1063/1.3597797

**Spin torque nano-oscillator based on a synthetic antiferromagnet free layer and a perpendicular-to-plane polarizer**

Firastrau, I., L.D. Buda-Prejbeanu, B. Dieny and U. Ebels

Journal of Applied Physics **113** (2013) 113908http://dx.doi.org/10.1063/1.4795160

**Out-of-plane precession of an in-plane magnetized free layer submitted to the spin transfer torque of a perpendicular polarizer: An analytical perturbative approach**

Lacoste, B., L.D. Buda-Prejbeanu, U. Ebels and B. Dieny

Physical Review B **88** (2013) 0544250

http://dx.doi.org/10.1103/PhysRevB….

**Magnetization dynamics of an in-plane magnetized synthetic ferrimagnetic free layer submitted to spin transfer torques and applied field**

Lacoste, B., L. D. Buda-Prejbeanu, U. Ebels and B. Dieny

Physical Review B **89** (2014) 064408

http://dx.doi.org/10.1103/PhysRevB….

**Tunability versus deviation sensitivity in a non-linear vortex oscillator**

Martin, S., C. Thirion, C. Hoarau, C. Baraduc and B. Dieny

Physical Review B **88** (2013) 024421

http://dx.doi.org/10.1103/PhysRevB….

**Sub-nanosecond precessional switching in a MRAM cell with a perpendicular polarizer**

Marins de Castro Souza, M., B. Lacoste, R.C. Sousa, T. Devolder, L.D. Buda-Prejbeanu, A. Chavent, A. Mejdoubi, S. Auffret, U. Ebels, C. Ducruet, I.L. Prejbeanu, L. Vila, B. Rodmacq and B. Dieny

Proceedings of the 4th IEEE International Memory Workshop (2012) 6213651

http://ieeexplore.ieee.org/xpls/abs…

**Precessional spin-transfer switching in a magnetic tunnel junction with a synthetic antiferromagnetic perpendicular polarizer**

Marins de Castro Souza, M., R.C. Sousa, S. Bandiera, C. Ducruet, A. Chavent, S. Auffret, C. Papusoi, I.L. Prejbeanu, C. Portemont, L. Vila, U. Ebels, B. Rodmacq and B. Dieny

Journal of Applied Physics **111** (2012) 07C912

http://link.aip.org/link/doi/10.106…

**Effects of the heating current polarity on the writing of thermally assisted switching-MRAM**

Chavent, A. ; Alvarez-Hérault, J. ; Portemont, C. ; Creuzet, C. ; Pereira, J. ; Vidal, J. ; Mackay, K. ; Sousa, R.C. ; Prejbeanu, I.L. ; Dieny, B.

IEEE Transactions on Magnetics **50 **(2014) 3401504

http://ieeexplore.ieee.org/stamp/st…

**Field dependence of spin transfer torque switching current in perpendicular magnetic tunnel junctions **

Cuchet, L. ; Sousa, R.C. ; Vila, L. ; Auffret, S. ; Rodmacq, B. ; Dieny, B.

IEEE Transactions on Magnetics **50 **(2014) 4401404

http://ieeexplore.ieee.org/stamp/st…

**Magnetization dynamics of an in-plane magnetized synthetic ferrimagnetic free layer submitted to spin transfer torques and applied field **

Lacoste, B. ; Buda-Prejbeanu, L.D. ; Ebels, U. ; Dieny, B.

Physical Review B **89 **(2014) 064408

http://journals.aps.org/prb/pdf/10….

**Modulating spin transfer torque switching dynamics with two orthogonal spin-polarizers by varying the cell aspect ratio **

Lacoste, B. ; Marins de Castro Souza, M. ; Devolder, T. ; Sousa, R.C. ; Buda-Prejbeanu, L.D. ; Auffret, S. ; Ebels, U. ; Ducruet, C. ; Prejbeanu, I.L. ; Vila, L. ; Rodmacq, B. ; Dieny, B.

Physical Review B **90 **(2014) 224404

http://journals.aps.org/prb/pdf/10….