gate scan can not be finished

[yang su]

Hello, I am running a gate scan of a MOSFET to plot the Id vs Vg curve of it. First I calculated the device configuration and transmission spectrum at equilibrium. And then I use the analysis from file to do a gate scan from -2V to 2V. the equilibrium transmission spectrum take like 5 hours to finish, but the gate scan has been running for 24 hours and just complete 2 points. Can you take a look at my script and help me to figure it out? Thank you very much.

[yang su]

And by the way, when I try to visualize the results like transmission spectrum and device density of states in the vnl window when I clicked the show button, nothing popped out even a blank window...I am wondering what is going on.

[Anders Blom]

It is much harder to converge the system at finite gate voltage, and this is why you use the zero-bias as starting point. However, the way you set up the scan over the gate voltage, it starts at the hardest case, -2 V. What you need is to split the positive and negative voltages, and start with the lowest voltage and climb up. So, one run with

Code: python

gate_voltage_list=numpy.linspace(0,2.0,8)*Volt

and another with

Code: python

gate_voltage_list=numpy.linspace(0,-2.0,8)*Volt

I would also save the converged calculation of each gate voltage, in case you want to compute any other quantity (or increase for instance the k-point sampling) later. Otherwise you will have to recompute the whole thing again. So, add

Code: python

nlsave('/home/it2/suxxx172/ATKSIMS/nc_files/gatescan_%s.nc' % voltage, device_configuration)

Note how this splits it up by saving in separate NC files, the calculation files can be quite large. But the transmission can go into one file so it's easier to plot the i-v curve. You should also always have the ".nc" extension on the nlsave file. The other problem is unclear to me...

[yang su]

Thank you.. So where should I save the device configuration after each scan point? Should I put that right before the saving of transmission spectrum at each point?
I also got a question that do I need to load the updated device configuration at the beginning of every step or just like right now. I am thinking if I load the original configuration at each step, the saving would not help right?
One more problem is that I calculated the energy dependent lddos at equilibrium and plotted lddos vs z and energy using the script just as in the tutorial. I got the figure attached and it is just from this device configuration. In the tutorial the lddos was calculated by adding all the dos at every location of z axis. It comes up that the figure make no sense for analysis. I can not extract information from that. Can I have some band diagram thing like what we usually do in the device analysis? Or how does this lddos figure can tell the information of the band structure information of the device? Thank you.

[Anders Blom]

Well, before the transmission yes, but you should also add an "update" statement, instead of relying on the transmission to do that for you. So, add

Code: python

    device_configuration.update()
    nlsave('/home/it2/suxxx172/ATKSIMS/nc_files/gatescan_%s.nc' % voltage, device_configuration)

(indented correctly) between the setCalculator() and the transmission. That's all, you don't need to worry about reading the configuration. The LDOS for zero bias is probably not very informative since most likely the current is zero (well, it is by definition). The point of the LDOS plot is to analyze the current, but you need a finite current for that. It looks from the picture that you need a bit of bias before there will be much current.

[yang su]

Thank you. So if the lddos calculation is just for analyzing the current flow in a device, what calculation I should do to plot the band information along the device like in other simulation tools? I think the band diagram is important to device guys like me for the information like band alignment..

[Anders Blom]

There is a very fundamental question in here: what is the "band edge" for a very narrow film? Can we really define it unambiguously? The entire concept is originally founded in two (related) assumptions: 1) periodicity - but the structure is not periodic in Z any more, and 2) macroscopic averaging such that the fluctuations on the atomic scale are not important.

Neither of these holds any more for an ultrascaled device: the central region is not periodic, and it's not large enough to average over. Or, well, you can average actually, and that does then to give a bit smoother plots, and looking at the LDOS (or PDDOS) does give a reasonable picture - within some limitations - of how the band edges look.