Resolution-dependent mechanisms for bimodal switching-time distributions in simulated Fe nanopillars

S. H. Thompson, G. Brown, Alan Kuhnle, Per Arne Rikvold, M. A. NovotnyPhysical 2009 · DOI · arXiv

Abstract

We study the magnetization-switching statistics following reversal of the applied field for three separate computational models representing the same physical system: an iron nanopillar. The primary difference between the models is the resolution of the computational lattice and, consequently, the intrinsic parameters that must be rescaled to retain similarity to the physical system. Considering the first-passage time to zero for the magnetization component in the longitudinal (easy-axis) direction, we look for applied fields that result in bimodal distributions of this time for each computational system and compare the results to the experimental system. We observe that the relevant fluctuations leading to bimodal distributions are different for each lattice resolution and result in magnetization-switching behavior that is unique to each computational model. Correct model resolution is thus essential for obtaining reliable numerical results for the system dynamics.

Introduction

As magnetic storage devices push toward terabit-per-square-centimeter densities, the stability and speed of individual magnetic bits become critical. Iron (Fe) nanopillars are prime candidates for this technology, but their switching behavior—how they flip from one magnetic state to another—is surprisingly complex. Experimental data often shows "bimodal" switching-time distributions, meaning the system doesn't just have one average switching speed; instead, it sometimes flips very fast and sometimes much slower. This research investigates whether simplified, lower-resolution computational models can accurately capture the physics behind this bimodality.

Methods

The researchers compared three different computational models of the same physical system (a 10 x 10 x 150 nm³ iron pillar), as illustrated in Figure 1:

  1. High-Resolution (6 x 6 x 90 cells): The most realistic model, where the lattice size is smaller than the "exchange length."
  2. Medium-Resolution (1 x 1 x 15 cells): A one-dimensional stack of spins that allows for non-uniformity along the length but not across the width.
  3. Low-Resolution (1 x 1 x 1): A single "macrospin" representing the entire pillar.
Computational models of an iron nanopillar at three resolutions: (a) high-resolution 3D lattice, (b) medium-resolution 1D chain, and (c) low-resolution macrospin.

All models utilized the stochastic Landau-Lifshitz-Gilbert (LLG) equation to simulate the time evolution of magnetization under thermal fluctuations. The primary metric was the First-Passage Time (FPT)—the time it takes for the magnetization to cross zero after the external magnetic field is reversed.

Key Insights and Resolution-Dependent Mechanisms

The most striking finding is that "bimodality" arises from three distinct physical processes depending on the model resolution:

  • The High-Res "Entropy" Path: Bimodality is driven by a high-dimensional free-energy landscape. The system can get "trapped" in a long-lived metastable configuration (the slow mode) or follow a path with constant free energy (the fast mode).
  • The Medium-Res "Coordination" Path: Bimodality depends on whether one or both ends of the pillar start switching at the same time. Simultaneous nucleation at both ends results in switching that is twice as fast.
  • The Low-Res "Precession" Artifact: Bimodality depends on whether the first precessional "swing" of the magnet is "knocked" over the threshold by thermal noise. If it misses, it must wait for the next precession cycle, creating a distinct gap in switching times.

Results and Discussion

Only the high-resolution model adequately captures the fluctuations that result in multiple switching paths in the free energy. However, the medium-resolution model showed the best agreement with experimental data regarding the minimum switching field, suggesting the actual metallic iron core of experimental pillars may be thinner than reported.

Cumulative probability distribution of switching times, showing the characteristic steps of a bimodal distribution.

As shown in Figure 2, the cumulative probability of switching times exhibits distinct "steps," which are characteristic of bimodal distributions. This visualization highlights how the system clusters into fast and slow switching events. The study warns that a model might produce the "right" statistics (like those in Figure 2) for the "wrong" physical reasons if the resolution is not carefully chosen.

Conclusion

Resolution matters fundamentally in magnetic simulations. While low-resolution models are computationally efficient, they may rely on artifacts rather than physical fluctuations to produce bimodal behavior. Future work will involve a more detailed analysis of high-resolution internal magnetization dynamics to reconcile discrepancies between theoretical models and experimental observations.

Summary updated

Abstract

We study the magnetization-switching statistics following reversal of the applied field for three separate computational models representing the same physical system: an iron nanopillar. The primary difference between the models is the resolution of the computational lattice and, consequently, the intrinsic parameters that must be rescaled to retain similarity to the physical system. Considering the first-passage time to zero for the magnetization component in the longitudinal (easy-axis) direction, we look for applied fields that result in bimodal distributions of this time for each computational system and compare the results to the experimental system. We observe that the relevant fluctuations leading to bimodal distributions are different for each lattice resolution and result in magnetization-switching behavior that is unique to each computational model. Correct model resolution is thus essential for obtaining reliable numerical results for the system dynamics.
BibTeX
@inproceedings{thompson2009resolution,
  author = {S. H. Thompson and G. Brown and Alan Kuhnle and Per Arne Rikvold and M. A. Novotny},
  title = {{Resolution-dependent mechanisms for bimodal switching-time distributions in simulated Fe nanopillars}},
  booktitle = {Physical},
  year = {2009},
  doi = {10.1103/physrevb.79.024429},
}
Figure