1
General Questions and Answers / Re: Spintronics and MRAM in ATK
« on: June 2, 2011, 03:30 »
Hi Anders Blom,
You mentioned: "ATK actually computes the Gibbs free energy for device configurations".
Does it mean that we should use a device configuration to calculate the intralayer exchange coupling, not a bulk configuration?
For example, in the Fe/MgO/Fe system, we need calculate the total energy with P and AP configuration of both electrode Fe and surface Fe.
Can we use a Fe/MgO/Fe bulk to calculate the intralayer exchange coupling by setting P and AP configuration of Fe at two sides?
If we don't consider the bias, these two methods look the same.
thanks.
You mentioned: "ATK actually computes the Gibbs free energy for device configurations".
Does it mean that we should use a device configuration to calculate the intralayer exchange coupling, not a bulk configuration?
For example, in the Fe/MgO/Fe system, we need calculate the total energy with P and AP configuration of both electrode Fe and surface Fe.
Can we use a Fe/MgO/Fe bulk to calculate the intralayer exchange coupling by setting P and AP configuration of Fe at two sides?
If we don't consider the bias, these two methods look the same.
thanks.
Both of these quantities are, at least in their simplest version, rather trivial to compute once you have the currents and total energies for the parallel and anti-parallel electrode configurations.
Following Phys. Rev. Lett. 97, 237205 (2006), in the collinear approximation you can compute the spin torque parallel to the interface planes as
T||(theta) = hbar/2e*sin(theta)*(I_AP-I_P)/2
where I_AP/I_P is the anti-parallel/parallel SPIN current (i.e. spin up minus spin down current).
The interlayer exchange coupling is the difference in total energy between the two configurations; see http://cnst.nist.gov/epg/Pubs/pdf/epg734.pdf, Eq. (3). Now, ATK actually computes the Gibbs free energy for device configurations, rather than the total energy, and I don't know for sure how that influences the results, esp. given the notes in the references article that it's often difficult - and crucial - to determine the energy difference accurately.
I'm sure the general computed trends wouldn't be much affected, but it is clear one must take care to converge the calculations to a low tolerance, and make sure to have proper k-point sampling etc (as we know this influences at the transmission very much, and hence the current, and the difference between the two configurations).