Hello, laborious quantumatk staffs and users, thanks for your attention with this topic.
I met a problem with the simulation of graphene-nanoribbon-based MOSFET. The device diagrams are attached. This is a typical double-gate device model. The thickness and relative dielectric constant of top and bottom oxide layers are 1 nm and 3.9, respectively. (I have assumed that the graphene layer is of no thickness. For more precise simulation, one must confirm the dielectric constant of graphene nanoribbon. As far as I know, it is still an open question) And the electrodes are n-doped. For simplification, a very small device demo is shown, but it does not influence the final result. I have simulated this structure using both LCAO and DFTB calculators. The Parallel Conjugate Gradient Poisson solver is adopted with Neumann boundary in the AB direction and Dirichlet boundary in the C direction. (Multi-grid always reports a residual warning and is really computation-expensive) Distinct results are attached. It is amazing that the gate almost has no effect on the potential profile computed by LCAO calculator. I know the electrostatic control of the gate is dramatically reduced due to the short channel effect. But the same configuration computed by DFTB calculator show an evidently better electrostatic control of the gate. And the difference becomes more obvious in a long MOSFET model according to my previous calculations.
The LCAO-DFT calculation has considered more electronic states compared to the DFTB calculation. May this account for this? But I think that there should be the same (or similar) number of net charge in two kinds of calculations and thus the electrostatic control of the gate should not be distinct. Can anyone give me some tips?
