In this thesis some categories of magnetization phenomena,
connected with high-end technological applications have been
investigated. The followed approach has its foundations in
micromagnetics, which is capable of properly describing magnetic
phenomena on sub-micron scale. A first step has been done by
studying magnetization dynamics under the hypothesis that the
particles were uniformly magnetized. In this framework, it has
been recalled that magnetization dynamical processes of
technological interest, like damping switching and precessional
switching, can be treated with analytical approaches present in
literature, which provide critical design parameters like critical
fields, as well as switching time and, for precessional switching,
the time tolerance allowed in order to have successful switching.
A slightly different context in the framework of uniform mode
theory, has been touched, regarding some aspects of the LLG
dynamics under circularly polarized fields, which arises in
typical ferromagnetic resonance experiments. In particular, a
special perturbative technique, based on mathematical background
in the framework of dynamical systems theory, has been developed
for the study of quasiperiodic solutions of LLG equation under
circularly polarized field. In this respect, the problem of
finding quasiperiodic solutions has been turned into the
determination of limit cycles of a suitable modification of LLG
equation, obtained by introducing the appropriate rotating
reference frame. Basically, the result is that the study of the
existence, the number, the stability of limit cycles of the
dissipative LLG dynamics can be performed, at first order, on the
conservative dynamics. The analytical results have been confirmed
by numerical simulations that indeed show the accuracy of the
predictions.
In addition, this technique, which permits also to study the
(global) bifurcations of the limit cycles, has been applied to the
study of LLG dynamics driven by spin-transfer torque. This subject
is currently the focus of considerable research for its
applications to current-driven switching of MRAMs cells. In
particular, by using the above perturbative technique, and on the
other hand the analytical treatment of the conservative dynamics,
it is possible to predict analytically the critical values of the
electric current and magnetic fields that rule complicated
behaviors, like the onset of self-oscillations and the
current-driven switching, observed in recent experiments on
spin-injection. Moreover, with this technique characteristics of
the self-oscillations like frequency and amplitude can be
analytically computed from the knowledge of the conservative
dynamics.
As soon as the hypothesis of uniformly magnetized particles has
been abandoned, the problem of the numerical computation of the
magnetostatic field, which has been recognized to be the
bottleneck of micromagnetic computations, has been analyzed. The
two most used methods, respectively for finite differences and
finite elements discretizations, have been described. Afterwards
it has been shown with micromagnetic simulations of damping and
precessional switching, that the former is an intrinsically
non-uniform process, whereas the latter can be reasonably
considered quasi-uniform also for dimensions of hundreds of
nanometers (half micron) and moderately low anisotropy. In this
respect, by computing the switching time with the uniform mode
analysis, reliable switching can be obtained. It has been
demonstrated that the accuracy of the prediction increases for
increasing anisotropy of the material. Moreover, for moderately
soft materials, with in the order of
J/m,
a tolerance of on the pulse amplitude is allowed. The
predicted time window agree with micromagnetic simulations with a
precision of few picoseconds.
Then the fast switching of tilted granular media has been analyzed
by means of a uniform mode approximation. The medium has been
considered as a collection of noninteracting grains with
dispersion of easy axes and initial conditions. In this framework,
the necessary condition for the switching of the whole granular
medium, i.e. the applied field amplitude range which makes fast
switching possible, has been derived by analyzing the single grain
dynamics, first conservative and then dissipative. A set of
micromagnetic simulations have confirmed the predictions made with
the uniform mode analysis.
Thus, the main result of this study is the fact that the uniform
mode theory can be applied to study processes of technological
interest, and in some cases it provides critical design
parameters.
Beside the above analysis, the problem of geometrical integration
of LLG equation has been addressed. In fact, the mostly used
numerical time-stepping techniques do not preserve the fundamental
properties of LLG dynamics, namely magnetization magnitude
conservation, Lyapunov structure for constant in time applied
field and hamiltonian structure in the conservative case. There is
an interesting example in literature of how the missed fulfillment
of magnitude conservation leads to inaccurate computation of
magnetostatic field. Moreover, the quantitative discordance
arising in the solution of micromagnetic standard problems
suggests that the particular choice of numerical methods may
affect the results of the computations. Therefore, we are
convinced that a numerical model has to qualitatively reproduce
the properties of the continuous model as best as possible, but,
nevertheless it must have a feasible computational cost. In this
respect, the proposed implicit mid-point rule technique has
revealed very efficient from both points of view. First of all, it
can be applied to any spatial discretization, like finite
differences and finite elements, which preserves the formal
structure of the effective field. Then, we have shown that the
mid-point discretized LLG equation exactly fulfills magnetization
magnitude conservation regardless of the time step. In addition,
in case of constant applied field, the discrete dynamics has
itself a Lyapunov structure regardless of the time step, and in
the case of conservative dynamics () the discretized
free energy is preserved regardless of the time step and the
hamiltonian structure is preserved up to third order in the time
step.
The implicit nature of the mid-point time-stepping leads to the
solution of a nonlinear system of equations at each time step.
Therefore, special and reasonably fast quasi-Newton iterative
procedure has been developed to solve this system. Since the
solution is approximate depending on the tolerance of the
quasi-Newton procedure, the precision in the fulfillment of the
LLG dynamics properties has been checked a posteriori. In
particular, for finite differences spatial discretization, we have
solved the micromagnetic standard problem no. 4. As far as the
accuracy tests are concerned, the magnitude has been preserved
within machine precision for each cell and the Lyapunov structure
is preserved with a relative error in the order of .
Conservative simulations of the same problem show that the free
energy is preserved with relative error also in the order of
. As far as computational effort is concerned, the use of
quasi-Newton technique which approximates the full jacobian matrix
of the nonlinear system of equations as a sparse matrix, allows
the use of fast iterative methods (GMRES) for the inversion of the
linear systems arising in the single quasi-Newton iteration. The
moderately low measured simulation times, together with the fact
that the time step can be chosen much larger than explicit methods
due to mid-point rule unconditional stability, make this method a
good candidate for accurate micromagnetic simulations.
Future work in this framework could be made by developing a finite
element code with mid-point rule time-stepping, which would permit
to treat magnetic bodies with in principle arbitrary shape.
Moreover the computational cost could be lowered by implementing
suitable preconditioning for the GMRES method. In addition, the
inclusion of the spin-transfer torque term in the code would
permit to investigate non-uniform spin-injection phenomena in
multi-layers structures.
Finally, the inclusion of thermal effects in magnetization
dynamics model and, consequently in micromagnetic simulations,
would be a considerable improvement of the investigation. This
direction will be pursued in future activities.
Next:A. Appendix A Up:main Previous:4.8.3 Discussion about computationalContents
Massimiliano d'Aquino
2005-11-26