Size dependence of nanosecond-scale spin-torque switching in perpendicularly magnetized tunnel junctions

We have time resolved the spin-transfer-torque-induced switching in perpendicularly magnetized tunnel junctions of diameters from 50 to 250 nm in the subthreshold thermally activated regime. When the field and the spin torque concur to both favor the P to AP transition, the reversal yields monotonic resistance ramps that can be interpreted as a domain wall propagation through the device at velocities of the order of 17 to 30 nm/ns; smaller cells switch faster, and proportionally to their diameter. At the largest sizes, transient domain wall pinning can occasionally occur. When the field hinders the P to AP transition triggered by the spin torque, the P to AP switching is preceded by repetitive switching attempts, during which the resistance transiently increases until successful reversal occurs. At 50 nm, the P to AP switching proceeds reproducibly in 3 ns, with a monotonic featureless increase of the device resistance. In the reverse transition (AP to P), the variability of thermally activated reversal is not restricted to stochastic variations of incubation delays before the onset of reversal: several reversal paths are possible even in the smallest perpendicularly magnetized junctions. Besides, the nonuniform nature of the magnetic response seems still present at the nanoscale, with sometimes electrical signatures of strong disorder during the AP to P reversal. The AP to P transition is preceded by a strong instability of the AP states in devices larger than 100 nm. The resistance becomes extremely agitated before switching to P in a path yielding a slow (20 to 50 ns) and irregular increase of the conductance with substantial event-to-event variability. Unreversed bubbles of typical diameter 60 nm can persist a few additional microseconds in the largest junctions. The complexity of the AP to P switching is reduced but not suppressed when the junctions are downsized below 60 nm. The instability of the initial AP state is no longer detected but the other features are maintained. In the smallest junctions (50 nm) we occasionally observe much faster (sub-1 ns) AP to P switching events that could result from a macrospin process. We discuss the origin of the switching asymmetry and the size dependence, with an emphasis on the role of the nonuniformities of the stray field emanating from the reference layers of the tunnel junction, which affects the zones in which nucleation is favored.

Figure 1

Panel (a): simplified sketch of the experimental setup. The signal is the voltage delivered to a $50\mathrm{\Omega }$ oscilloscope by the current flowing into the device when submitted to a slow (dc) voltage ramp and coupled to charge-pump capacitor loaded by a $50\mathrm{\Omega }$ matching system. Panel (b): fastest observed switching events for junctions of diameter 60 nm and resistance of $4.2\mathrm{k}\mathrm{\Omega }$ in the parallel state. The two vertical bars are separated by 1 ns. The AP to P switching event is measured to last less than 1 ns; however, the assessment of its exact duration is impeded by the capacitive part of the device that filters out any faster evolution. Panel (c): quasistatic resistance vs field hysteresis loops for circular devices of nominal sizes 50, 120, and 200 nm. The actual sizes are deduced from the measured device resistance in the remanent parallel state.

Figure 2

Quasistatic resistance vs voltage hysteresis loops for circular devices of sizes 120 nm (top panel) and 250 nm (bottom panel) for various applied fields. Notice the plateaus at intermediate resistance level at high fields for the largest junction.

Figure 3

State diagrams of 50 nm and 250 nm diameter MTJs, as constructed from resistance vs voltage loops recorded at constant fields. U stands for “unknown” state of time-average conductance intermediate between that of P and AP states.

Figure 4

Time-resolved measurements of subthresholds STT-induced $\mathrm{P}\to \mathrm{AP}$ (left panels) and $\mathrm{AP}\to \mathrm{P}$ (right panels) transitions for junctions of diameter 250 nm. The curves are vertically and horizontally offset for readability. The signal is the voltage delivered to a fixed 50 Ohm load by the (signed) current flowing into the sample, i.e., it has a shape reflecting the opposite (left panels) or similar to (right panels) the conductance evolution. The arrows in the $\mathrm{AP}\to \mathrm{P}$ transition indicate the moment at which reversal gets complete. The amplitude of the corresponding signal change is equivalent to the annihilation of a bubble of diameter 60 nm.

Figure 5

Time-resolved measurements of subthresholds STT-induced $\mathrm{AP}\to \mathrm{P}$ transitions for junctions of diameter 150 nm (left panels) and 90 nm (right panels). The curves are offset for readability. The time scales are expended in the bottom panels.

Figure 6

Time-resolved measurements of subthresholds STT-induced transitions for a junction of diameter 60 nm. The curves are vertically and horizontally offset for readability. The black curves are the raw data, while the colored curves are their counterparts after mathematical low pass filtering of 3 dB cutoff at 1 GHz. (a) Examples of $\mathrm{P}\to \mathrm{AP}$ transitions. The curve labeled “mean” is an average of 50 $\mathrm{P}\to \mathrm{AP}$ events, meant to evidence the very high reproducibility of the electrical signature of this transition at small junction sizes. (b) Examples of $\mathrm{AP}\to \mathrm{P}$ transitions. Note the different scales on the time axes.

Figure 7

Proposed spin-torque switching scenario in large (diameter $\ge 100$ nm) junctions in globally compensated field conditions. The $\mathrm{P}\to \mathrm{AP}$ transition proceed through easy domain wall motion once a wall has emerged (nucleation) from the edge. This is the limiting step as in the P configuration; the edges of the free layer are stabilized by the dipole field from the closest layer of the reference synthetic antiferomagnet. The $\mathrm{AP}\to \mathrm{P}$ proceeds by a complex dynamics involving fluctuations at the junction perimeter, followed by the shrinking of the unreversed domain until the bubble reaches a diameter in the 60 nm range, and finally collapses. The curves are taken from a 250 nm junction.