Subnanosecond spin-transfer switching: Comparing the benefits of free-layer or pinned-layer biasing

We analyze the statistical distribution of switching durations in spin-transfer switching induced by current steps and discuss biasing strategies to enhance the reproducibility of switching durations in spin valves. We use a macrospin approximation to describe a thin nanomagnet having easy-plane shape anisotropy and uniaxial magnetocrystalline anisotropy. We model the effect of finite temperature as a Boltzmann distribution of initial magnetization states (adiabatic limit). We compare three model spin valves: a spin valve with a free layer whose easy axis is parallel to the pinned-layer magnetization (standard geometry), a pinned layer with magnetization tilted with respect to the free-layer easy axis (pinned-layer biasing), and a free layer whose magnetization is pulled away from the easy axis by a hard-axis bias (free-layer biasing). In the conventional geometry, the switching durations follow a broad regular distribution, with an extended long tail comprising very long switching events. For the two biasing strategies, the switching durations follow a multiply stepped distribution, reflecting the precessional nature of the switching and the statistical number of precession cycles needed for reversal. We derive analytical criteria to avoid switching events lasting much longer than the average switching duration in order to achieve the highest reproducibilities. Depending on the current amplitude and the biasing strength, the width of the switching time distribution can be substantially reduced, the best reproducibility being achieved for free-layer biasing at overdrive current of a few times unity. An even smaller distribution of switching time is expected if the field is applied abruptly and synchronously with the current, following a so-called dynamic free-layer biasing configuration.