Intrinsic back-switching phenomenon in SOT-MRAM devices

TL;DR

This work tackles the problem of deterministic switching in SOT-MRAM by systematically characterizing back-switching (BSW) as an intrinsic spin-orbit torque effect. It combines statistical WER mapping on sub-100 nm CoFeB/MgO pillars within β-W Hall crosses with macrospin LLg simulations that incorporate DL and FL torques and thermal fluctuations, achieving good agreement with experimental results. The study reveals that BSW is not random but tied to the magnetization direction at the end of a write pulse and the relaxation trajectory, and it demonstrates practical mitigation via pulse shaping that extends the deterministic write current window and achieves very low WER (below $2 \times 10^{-6}$ in a complete cell). These insights enable more reliable SOT-MRAM operation and inform design strategies for embedded memory applications. The results highlight the value of compact macrospin models for rapid WER mapping and device optimization, while pointing to nucleation-dominated mechanisms and the potential need for micromagnetic modeling to capture domain-wall dynamics in some regimes.

Abstract

The writing process of SOT-MRAMs is considered deterministic when additional symmetry-breaking factors, such as the application of an external magnetic field aligned with the current, are present. Notably, the write probability exhibits a unique behavior as a function of the current: it drops to zero at high currents or even oscillates with the current. This phenomenon is attributed to back-switching, an intrinsic effect of magnetization reversal driven by spin-orbit torques. A systematic investigation of this back-switching phenomenon is conducted on sub-100 nm CoFeB magnetic pillars positioned at the center of $β$-W Hall crosses. Using a statistical approach, the study examines the impact of various parameters, including the amplitude of current pulses and the application of magnetic fields in different directions. The findings reveal that the back-switching phenomenon is not statistically random. Macrospin simulations, employing realistic magnetic parameter values, accurately replicate the experimental observations and provide insights into the underlying mechanisms of back-switching. These simulations also explore strategies to mitigate the phenomenon, such as optimizing the shape of the writing pulses. Applying this approach to complete SOT-MRAM single cells achieves a write error rate below $2 \times 10^{-6}$, demonstrating the effectiveness of this strategy in expanding the operational current range for write operations in SOT-MRAMs.

Figures (15)

• Figure 1: (a) To design a memory array, the statistical variation in device properties must be taken into account. To ensure reliable writing, the distribution of the current supplied by the write transistor (I$_{\text{Transistor}}$) should be higher than the distribution of SOT-MRAM write current (I$_{\text{Write}}$) and should not overlap with the distribution of the current inducing a back-switching phenomenon (I$_{\text{Back-switching}}$). Ideally, I$_{\text{Transistor}}$ should be in the middle of I$_{\text{Write}}$ and I$_{\text{Back-switching}}$. (b) A schematic representation of the studied magnetic stack and the device used for electrical measurements is shown: a magnetic pillar patterned onto a $\beta$-tungsten Hall cross (see the representative SEM image). The applied writing current I$_{\text{SOT}}$ is injected through the longitudinal arms of the Hall cross, while the anomalous Hall voltage is measured across the transverse arms. H$_{\text{X}}$, H$_{\text{Y}}$, and H$_{\text{Z}}$ denote the applied magnetic field components along the respective axes.
• Figure 2: (a) WER measurement scheme and definition. (b) Typical hysteresis curve for current-induced magnetization reversal. The current pulse is 10 ns long, with a rise/fall time of 2 ns, and a permanent field (H$_{\text{X}}$) of 950 Oe is applied. The areas colored green (respectively red) define the current ranges for which switching (respectively back-switching) is observed. V$_0$ and V$_1$ are the positive voltages that define the boundaries of these areas (see text). (c) Corresponding WER measurement curve where larger voltages were applied. A deterministic back-switching is observed while increasing the applied voltage.
• Figure 3: Experimental and simulated WER color maps as a function of applied SOT current using 5 ns-long pulses with 70 ps rise/fall times. The experimental results are shown in (a) for a variable H$_{\text{X}}$ field, (b) for a variable H$_{\text{Y}}$ field in the presence of a fixed H$_{\text{X}}$ field of 600 Oe, and (c) for a variable H$_{\text{Z}}$ field and a fixed H$_{\text{X}}$ field of 800 Oe. Simulation results are shown opposite for the same applied fields (d), (e) and (f). These WER maps were obtained using 100 write/read iterations. The blue color, corresponding to a value of 1, indicates an error for each of the procedures, and the yellow color indicates no writing error.
• Figure 4: Magnetization switching orbits and temporal evolution of the magnetization components obtained from macrospin simulations at (a) H$_{\text{X}}=500$ Oe for different I$_{\text{SOT}}=550~\mu$A (sub-critical)(red), $650~\mu$A (deterministic)(green) and $750~\mu$A (non-deterministic)(blue); (b) I$_{\text{SOT}}=750~\mu$A for H$_{\text{X}}=500$ Oe (non-deterministic)(blue) and H$_{\text{X}}=800$ Oe (deterministic)(orange); (c) H$_{\text{X}}=600$ Oe, I$_{\text{SOT}}=750~\mu$A in presence of an in-plane field along the y axis, H$_{\text{Y}}=-600$ Oe (non-deterministic) (yellow) and H$_{\text{Y}}=600$ Oe (deterministic) (brown); (d) H$_{\text{X}}=500$ Oe, I$_{\text{SOT}}=750~\mu$A for different $\alpha=0.01$ (non-deterministic)(light blue), $0.035$ (non-deterministic)(blue) and $0.2$ (deterministic)(red).
• Figure 5: Experimental WER curves as a function of applied SOT current using a 10ns long write pulse (a) for a 100 nm diameter CoFeB (0.9 nm) pillars at the center of $\beta$-W Hall crosses, annealed at 410° C for 30 minutes, with fall times of (i) 2 ns, (ii) 3 ns and (iii) 4 ns and an in-plane field of 950 Oe; and (b) for a 75 nm [CoFeB (0.72 nm)/MgO (RA$\sim$20 $\Omega\mu$m$^2$)/Pinned Layer] dot sitting on a 130 nm wide W(4 nm) SOT track, annealed at 410° C for 30 minutes, for a fall time of 2 ns and an in-plane field of 800 Oe.