(ESICMM-G8 Symposium on Next Generation Permanent Magnets, Tsukuba, 2015)
Large scale parallel micromagnetic simulations of magnetization reversals in Nd-Fe-B magnets using transient and steady analyses
T. Oikawa1,2, T. Ohkubo1, and K. Hono1
1. National Institute for Materials Science (NIMS), Tsukuba, 305-0047, Japan
2. TDK Corporation, Ichikawa, 272-8558, Japan
Abstract:
1. Introduction
Micromagnetic simulation is a very effective tool to visualize magnetization reversal process in permanent magnets [1]. However, to simulate realistic magnetization reversal processes in polycrystalline magnets, large scale parallel computation is necessary and the current computational power is insufficient for this purpose. Therefore a computation method with high parallel efficiency is required. Transient and steady analyses have been employed for solving LLG (Landau-Lifshits-Gilbert) equations. In the transient analysis, external magnetic field sweeps by time step, even if the precession of magnetization vector is not converged. In the steady analysis, external magnetic field is changed after confirming the convergence of precession. Recently energy minimization method, which uses only relaxation term of LLG equation, was reported to be effective to accelerate the computation [2]. In the present work, we investigate technical problems for parallel computation by comparing these three methods.
2. Simulation model
Polycrystal1ine grains were created by using Voronoi tessellation of a cube. The model size was 150 or 300 nm containing 125 crystal grains. It consists of 3 to 20 million tetragonal meshes, respectively. As microstructure, core-shell grain structure was incorporated. Grain boundary was assumed to be soft magnet with a thickness of 5 nm.
3. Simulation results
We performed the parallel computation using super computer (SGI ICEX). Steady analysis and energy minimization method show almost the same magnetization reversal patterns and demagnetization curves. Results of transient analysis were inconsistent with the other methods. Relative calculation speed normalized by 16 parallel cores is shown in Fig.1. Calculation speed of transient and steady analysis saturates at about 500 cores. On the other hand, saturation goes up to about 1500 cores for the energy minimization method. We also found that the computation time decreases to 1/20 by using the energy minimization method, indicating that the energy minimization method is promising for large scale parallel computation. Based on the present results, we estimate that realistic magnetization reversal simulation will be possible using post- K computer.
[1] H. Sepehri-Amin et al., Scripta Mater. 89, 29 (2014).
[2] L. Exl et al., J. Appl. Phys. 115, 17D118 (2014).
