Resonant spin-transfer-driven switching of magnetic devices assisted by microwave current pulses

The torque generated by the transfer of spin angular momentum from a spin-polarized current to a nanoscale ferromagnet can switch the orientation of the nanomagnet much more efficiently than a current-generated magnetic field and is therefore in development for use in next-generation magnetic random access memory (MRAM). Up to now, experiments have focused on spin-torque switching driven by simple square-wave current pulses. Here we present measurements showing that spin transfer from a microwave-frequency current pulse can produce a resonant excitation of a nanomagnet and improved switching characteristics in combination with a square current pulse. With the assistance of a microwave-frequency pulse, the switching time is reduced and achieves a narrower distribution than when driven by a square current pulse alone, and this can permit significant reductions in the integrated power required for switching. Resonantly excited switching may also enable alternative, more compact MRAM circuit architectures.

Figure 1

(Color online) Enhancement of magnetic switching using microwave current pulses at a background temperature of 20 K. (a) Resistance measured as a function of magnetic field applied along the exchange bias direction. Upper right inset: layer structure of the samples. Lower left inset: orientation of the free- and pinned-layer moments for the low-resistance equilibrium state. (b) For a precession amplitude less than the offset angle between the pinned-layer moment and the free-layer precession axis (smaller trajectory), a dc current produces a spin torque (black arrows) that increases the precession amplitude over half the cycle but decreases it for the other half. Only when the precession amplitude is greater than the offset angle (larger trajectory) will the spin torque from a dc current increase the precession amplitude on both halves of the precession cycle. (c) Examples of our excitation wave form, which combines microwave-frequency and square-wave pulses. $\varphi$ is defined as the phase of the microwave signal at the onset time of the square pulse. (d) Switching probability as a function of frequency after optimizing the phase $\varphi$. rf pulse parameters: 1.4 mA, 1.7 ns; square-pulse parameters: 7.7 mA, 0.7 ns. (e) Switching probability as a function of the phase $\varphi$ for $f=4.9\text{GHz}$ using the same pulse parameters as in (d). (f) Solid lines: switching probability as a function of square-pulse duration without a microwave pulse (black/darker curve) and with a 1.4 mA, 0.3 ns, $\varphi =50°$ microwave pulse (red/lighter curve). The square-pulse amplitudes used are both 7.7 mA. Dotted lines: corresponding distributions of switching pulse duration, corresponding to the derivative of the switching probability versus pulse length.

Figure 2

(Color online) Macrospin simulations of resonantly enhanced spin-torque switching at 20 K. (a) Dependence of the switching probability on the phase $\varphi$ for a 4.5 mA, 2 ns rf pulse with a 7 mA, 0.6 ns square-wave pulse. (b) Free-layer moment precession in response to the current pulse. The two curves in each panel show the applied current pulse and the resulting precession angle of the free-layer moment. The top panel shows the case with the optimum phase for switching, $\varphi =50°$. The bottom panel shows the least-optimum phase, $\varphi =230°$. In both cases, the precession of the free layer initially responds in phase with the applied rf current and the precession amplitude grows with time. For $\varphi =50°$ the square pulse is applied at the correct moment to optimally enhance the precession amplitude, leading to efficient switching, while for $\varphi =230°$ the torque from the square pulse decreases the precession amplitude over the first half-cycle of precession after the onset of the square pulse and switching is suppressed.

Figure 3

(Color online) Dependence of switching on the parameters of the microwave pulse at a background temperature of 20 K. (a) Switching probability versus frequency at the optimum phase $\varphi$, measured using rf pulse amplitudes from 0.7 to 1.8 mA. The rf pulse duration is 1.7 ns and the square-pulse parameters are 7.7 mA and 0.7 ns. (b) Switching probability for $f=4.9\text{GHz}$ as a function of rf pulse duration measured at different rf pulse amplitudes from 0.7 to 1.8 mA. The square-pulse parameters are the same as in (a). (c) The square-pulse duration required for a 95% switching probability as a function of rf amplitude for both the optimum phase $\varphi$ for switching and the least-favorable phase. The black dotted horizontal line indicates the pulse duration required to give 95% switching probability for a square pulse alone. The rf pulse duration is 1.7 ns. The square-pulse amplitude is 7.7 mA. (d) Wave forms for two pulses that both give a 95% switching probability.

Figure 4

(Color online) Effects of thermal fluctuations on resonantly enhanced switching. (a) Switching probability as a function of the phase $\varphi$ at a background temperature of 20 K for 1.4 mA, 0.8 ns rf pulses with 4.8 mA, 2 ns square pulses. (b) Solid lines: switching probability as a function of square-pulse duration without an rf pulse and with a 1.4 mA, 0.8 ns rf pulse. The square-pulse amplitudes are both 4.8 mA. Dotted lines: corresponding distributions of switching pulse duration. (c) Room-temperature switching probability as a function of the phase $\varphi$ for 1.5 mA, 1.5 ns rf pulses with 10.8 mA, 0.5 ns square pulses. (d) Solid lines: room-temperature switching probability as a function of square-pulse duration without an rf pulse and with a 1.5 mA, 1.5 ns rf pulse. The square amplitudes are both 10.8 mA. Dotted lines: corresponding distributions of switching pulse duration.

Figure 5

(Color online) Circuit schematic and timing diagram for combining rf and square-wave pulses to make our excitation wave forms.

Figure 6

(a) Switching probability as a function of pulse amplitude. $V_0$ is chosen as a reference point, with $P_0$ close to 50%. (b) Switching probability at pulse voltage $V_1$ as a function of a steady-state background current, $I$. When $I$ is large enough to return the switching probability to $P_0$, the steady-state current is equal to the difference in the currents generated at the sample by the pulses $V_0$ and $V_1$.