阴水
Atomistix ToolKit Tutorial Part 1: Two-probe geometry & convergence parameters
Outline
Part 1:
» Fundamental concepts of a two-probe system » Solving convergence problems » Based on a specific example which is hard to converge (FeMgO MTJ)
Fundamental but complex topics in ATK
Understanding the fundamental concepts of a two-probe system is crucial to establish confidence in the simulations and their results Most calculations in ATK converge with default parameters
蓝衣魔鬼» But some (for a variety of reasons) Geometrical aspects
» Related to certain “hidden” approximations and/or simplifications in ATK
Numerical parameters
» Accuracy » Solving convergence problems
Both aspects are related to convergence, but may also compromise quality of results if not chosen appropriately (even if the calculation converges!)
Three ways to improve performance Total calculation time = time/iteration × number of iterations Reduce time/iteration
» Parallelization (scaling depends on system & parameters) » Code and algorithm improvements (ATK 2008.10)
Reduce number of iterations
» Algorithm improvements (new convergence criterion in ATK 2008.02) » Parameter tuning
Third way?
让朋友圈清清爽爽» Avoid running end-less calculations which do not converge! » Avoid re-running calculations because of poor quality results
Our guinea pig Fe/MgO/Fe magnetic tunnel junction (MTJ) Anti-parallel electrode spin polarization
» Among the most difficult systems to converge » Strong peak in the minority DOS at the Fermi level
Fe MgO Fe
Many other interface systems exhibit similar convergence issues
铁橡栎Au/Si/Au two-probe