Author Topic: Fermi energy misalignment in graphene-based models  (Read 4164 times)

0 Members and 1 Guest are viewing this topic.

Offline Quantamania

  • Heavy QuantumATK user
  • ***
  • Posts: 38
  • Reputation: 0
  • Profoundly deaf and legally blind since birth.
    • View Profile
Fellows,
     I have been investigating layered materials based on graphene and hexagonal boron nitride.  Recently, when using Virtual NanoLab 2008.10, I examined effects of conformation and composition on the band structures of these layered materials.  I created a lattice-matched model for both layer materials to be used in the same unit cell, assuming that they relax to form the interface.  These materials differ by less than 1.8% in their lattice constants, so little strain is expected for the interface formation.  One consistent tendency I found was misalignment of the Fermi energy level of the graphene or carbon band structure.  The individual graphene bands rarely line up with the Fermi energy level at the K and H points of the Brillouin zone (where the Dirac intersection point should be).  Since band structure is calculated relative to the Fermi energy level, the misalignment is quite visible in band structure diagrams, because it is not placing the Dirac intersection point right at zero energy (should be the Fermi energy) of the diagram.  What factors can cause misalignment of the Fermi energy level, as in these examples?

I had found that boron nitride layers were causing the graphene layers to experience loss of degeneracy at the K and H points, turning graphene sheets into indirect band gap conductors.  The misalignment is not making it difficult to understand the K-H line behavior, but is making it tricky to look at carrier concentration in the line.  We do know that a band gap opens in graphene when it is covered with boron nitride layers, regardless of orientation.  This finding seems to be very similar to the discovery that electric fields cause degeneracy lifting in graphene or bilayer graphene (Nature, June 11, 2009), as the boron nitride layers have intrinsic electric fields of their own.  I have included four examples of the misaligned K-H lines, computed at DZDP basis level with LDA for exchange correlation.

Offline Quantamania

  • Heavy QuantumATK user
  • ***
  • Posts: 38
  • Reputation: 0
  • Profoundly deaf and legally blind since birth.
    • View Profile
Re: Fermi energy misalignment in graphene-based models
« Reply #1 on: June 22, 2009, 18:59 »
Seems that no one had an answer...

So I went ahead and had an idea that something in the parameters had to do with this quandary.  I noticed that the Brillouin zone integration grid has a very dramatic effect on where the Fermi energy level is for these materials.  I first tried a 10x10x1 grid then another experiment with 15x15x1 (different from the default 5x5x5 grid) on the same graphene model.

I got the K-H line to lie very close to the Fermi energy level with the 15x15x1 grid, so it is the integration grid that was producing the misalignment.  As I was using so few sample points in the grid, overestimation ran rampant in the calculations.  This led to the misalignment, so I plan to repeat the layered material experiments with a different grid.

So if you notice something different with your band structure calculations with the default 5x5x5 grid (even if the k*p behavior is great), try altering your integration grid by adding more points to it.  That will definitely improve accuracy of the calculations significantly, sometimes at little cost of computational time.

Offline Nordland

  • QuantumATK Staff
  • Supreme QuantumATK Wizard
  • *****
  • Posts: 812
  • Reputation: 18
    • View Profile
Re: Fermi energy misalignment in graphene-based models
« Reply #2 on: June 22, 2009, 20:28 »
I think you have point on the number of k-points - for transport calculation it can not be outlined enough that the number of the k-points in the transverse direction has a huge impact on the results of the convergence of the two-probe calculation.

Offline Quantamania

  • Heavy QuantumATK user
  • ***
  • Posts: 38
  • Reputation: 0
  • Profoundly deaf and legally blind since birth.
    • View Profile
Re: Fermi energy misalignment in graphene-based models
« Reply #3 on: June 22, 2009, 21:22 »
That is correct.  I am actually planning on repeating the layer calculations with a grid of 40x40x10 and using a different k-point pathway to expose the Dirac intersection points at K for graphene-containing models.  My original 5x5x5 grid results were good enough to reveal the splitting patterns and is being used towards a grant proposal in the future.

This means if you have a feeling about where the Fermi energy level should be but default grids are not treating it properly, use a finer grid to see if it fixes the energy placement of the individual bands in the calculation.

Offline Nordland

  • QuantumATK Staff
  • Supreme QuantumATK Wizard
  • *****
  • Posts: 812
  • Reputation: 18
    • View Profile
Re: Fermi energy misalignment in graphene-based models
« Reply #4 on: June 22, 2009, 21:50 »
However I also think it is important to note that the unitcell you are calculating on, is small ( 2 atoms ? )
For unitcells with perhap 500-1000 atoms, a 5x5x5 grid would properly be perfect.

Offline Anders Blom

  • QuantumATK Staff
  • Supreme QuantumATK Wizard
  • *****
  • Posts: 5405
  • Country: dk
  • Reputation: 89
    • View Profile
    • QuantumATK at Synopsys
Re: Fermi energy misalignment in graphene-based models
« Reply #5 on: June 22, 2009, 22:15 »
Also note that ATK is parallelized over the transverse k-points, so if you're running on a cluster, and you're in a situation where you are using a lot of k-points, increasing the number of MPI nodes, if possible, can have a substantial influence on the performance.

Offline Quantamania

  • Heavy QuantumATK user
  • ***
  • Posts: 38
  • Reputation: 0
  • Profoundly deaf and legally blind since birth.
    • View Profile
Re: Fermi energy misalignment in graphene-based models
« Reply #6 on: June 23, 2009, 00:37 »
Yes, I have unit cells with only two to six atoms in them.  I took notice of the grid size effect on the Fermi energy level after testing whether that strategy would help.  I am actually running this program on two dual-core computers and one single-core computer, but they are not parallel with each other.  This allows me to run three independent experiment trials.

The strategy here is to prepare high-quality runs for the same layered models as before, using the finer grid over the coarse grid that I started out with.  That way, I will have publishable band structure pictures that reproduce other works using VASP (theoretical plane-wave) and ARPES (experimental method).  So I basically had run 'crude' tests with the 5x5x5 grid and now am in a position to do the bigger grid runs.