Sorry for the delay! Point 1 explains it partly.
1. The electrodes had better be conducting, but it is also possible to use doped semiconductors. In ATK 12.2.2 (release expected early next week) it will be possible to assign a charge to an electrode. This shifts the Fermi level just like doping would, but instead of an explicit dopant atom, you get a much more reasonable effective doping concentration.
You can choose to have the channel extend into the electrode or not, it really depends on what you wish to simulate, but note that the electrode is repeated semi-infinitely to either side, so if that is not the physical situation, then you don't want that.
2. It's hard to define vertical or horizontal without a direction of gravity
So, in order to speak the same language, we had better be more explicit. The electrode is contacted in the C direction, which should be parallel to Z. The AB plane (and hence the XY plane) must be perpendicular to C.
3. It's really hard to say. The time increases roughly linear with the length in Z but perhaps more like N^2 with the extension in XY. Under any circumstance at 1000 atoms you are looking at a serious simulation. If you only have a single machine, MPI is not going to help much, unfortunately. MPI provides a very good scaling advantage when you run over several machines, however, so if you have 2-5 machines you can get a very nice speed-up (or if there is a cluster available, even better). But don't run MPI on a single machine, it will actually be slower.
4. Not really, but we can help you with details for each step as needed.
