optimization of magnetic tunneling junction

[zhangguangping]

Dear all,

I searched all the forum of how to do the opimization of a magnetic tunneling junction. For the magnetic electrodes, there are usually two configurations to considered: (1) parallel configuration--P; (2)antiparallel configuration (AP).

So, as I guess, one should do the geometric optimization for P and AP configuration, respectively.

At the same time, the electronic convergence of a magnetic tunneling junction usually harder than a non-magnetic one.

Therefore, is there a feasible way to solve this? Or how does one usually do for this?

With best regards,

Guangping

[Jess Wellendorff]

I would go for a mix of the procedures described in these two tutorials:
http://docs.quantumwise.com/tutorials/fe_mgo_fe.html
http://docs.quantumwise.com/tutorials/device_relaxation.html

The first tutorial does currently not use the more rigorous device relaxation method described in the second tutorial.

[zhangguangping]

Dear Jeff,

Thanks very much for your reply.

I have read the second tutorial before, and this time I read the first tutorial. This tutorial is of great help. As I understand, to speed up the electronic convergence in Magnetic Tunnel Junctions, one usually use a large electron temperature (indeed, this temperature is not the real temperature but a smearing and a trick for the convergence) up to ~1200 K.

I have two questions yet:

(1) Can the effect of parallel configuration or  antiparallel configuration of the electrodes be neglected to the geometry of the device? I saw the geometry optimization is only carried out for the parallel configuration. And this optimized geometry is used to do the electron transport calculation for antiparallel configuration.

(2) What is the idea of using a converged electron structure of the parallel configuration to initialized the antiparallel configuration? The tutorial metioned that this could improve the convergence behavior of the antiparallel configuration. is this a general rule? It is hard to believe.

With best regards,

Guangping

[Jess Wellendorff]

1) I do not think the AFM forces will be much different from the FM forces. This is of course an approximation, and you can certainly relax the AFM state also. But I'm guessing the difference will be small.
2) Using the FM state as an initial guess for the AFM state works like this:
a) use the converged  FM state as an initial state for the AFM calculation.
b) use the InitialSpin method to flip the spins on one side of the device in order to create an AFM state with the same (but partly flipped) density matrix as found in the converged FM calculation.

[zhangguangping]

Dear Jess, Thanks very much for your kind reply. It is of great help.

Code

initial_spin = InipialSpin(scaled_spins=[1.0, 1.0, 0.0, 0.0, -1.0, -1.0])

the scale factor specified by scaled_spins is used to scale the initial guess. If there is no initial guess, then it represents the scale factor of the maximum magnetic moment of the atom. Now I understand why the converged FM can be used as a good initial of AFM configuration. Thanks very much. Guangping