Dopant diffusion simulation in crystal Si

[pshinyeong]

Hello,

I am trying to simulate dopant diffusion (B,P,Ga,Te) in crystal Si for a start.
I am trying different methods such as NEB, AKMC and it seems using MTP is much less time consuming than other potentials. On top of that, I couldn't find any suitable force field potential.

1. To create MTP for a single dopant in crystalline Si, do I use the Defect Training template in the workflow?
If so, my reference configuration will be just pristine Si?
2. Would batch learning be enough or would you suggest active learning MTP for this case?
3. Is there a tutorial or a description/manual for defect training template?
4. Any ways to reduce the memory consumption while training MTP? (most of the MTP training attempts gave memory errors)

Thanks!

[Anders Blom]

1. Yes
2. Definitely use Active Learning because you dopant needs to move to many different locations in the crystal not captured by the simple "shake and twist" protocol in Batch (move atoms around a bit, change unit cell szie and shape a bit). You need to explore regions of the PES far from equilibrium for the NEBs to be accurate. Notably, in the newest version you can actually do active learning on the NEB itself though, that might save a lot of time!
3. Eh, not yet...
4. Really? Maybe share some data with us, so we can check. You could try to use a smaller basis_size of  course. But don't try to include all dopants at once, that will be too many elements.

[pshinyeong]

Dear Anders,

Thank you very much for your reply! Regarding the active learning on the NEB itself, I have the newest version and I prepared my NEB configuration. Could you possibly guide me on how I can apply active learning on NEB itself?

Also I have attached the python script and log file regarding the memory errors that appeared during MTP training.

Thanks alot!