Problems and Limitations¶
How much memory do I need?¶
We recommend 16 GB or larger RAM. 8 GB may be sufficient for some input files, but the calculation becomes incredibly slow if there is not sufficient memory available.
You should run heavier simulations on a computer having 32 GB of RAM. 16 GB RAM might require you to reduce the number of parallel threads used for the gain calculation. We recommend a Windows PC with 64 GB RAM and a recent CPU. The whole PC is then around USD 1,000.
My results do not agree with experiment.¶
A discrepancy in the used effective masses could be an explanation. In addition, the Interface roughness scattering plays an important role. The values given in the paper of A. Wacker seem to be taken in order to fit the experimental data. However, their model for elastic scattering is simplified (with no in-plane dependence for the scattering processes), so it might be that the actual interface roughness is different in reality.
Why does the current density vary at different places? Current does not seem to be conserved.¶
The NEGF calculation is done within the mode space basis that depends on Axial energy cut-off. Mathematically speaking, the Hilbert space of the Green’s functions has lower dimension than the initial Hamiltonian which had the full real space resolution (SpatialGridSpacing). As a consequence, the current calculated from the solution of the Dyson-Keldysh equations is not exactly conserved in real space, especially through high barriers. The current should be better conserved when more subbands are selected.
In the gain spectra, the energy grid is not as expected from the input dEPhoto/dEPhotoSelfConsistent.¶
The energy interval for the gain calculation will be set to at least the EnergyGridSpacing.
The simulation results contain the Gain folder but it is empty.¶
Note that the gain is output for the voltages specified in the input file by Vmin and Vmax. This is to reduce the gain calculation time.
Last update: 31/10/2024