calculation not converge at low bias

[qddavid]

Dear sir, I  built a MoS2 nanoribbon device which has been optimized. However, when I calculated the device at 0.6V, calculation cannot converge. And the calculations were not easy to converge when the biases were 0.4V and 0.5V, the converge steps were more than 120.   

I have read the Converge Tips & Tricks tutorial and changed some parameters, but it was not very useful.

Could you help me  converge the calculation as well as save time?  Thank you very much.

[zh]

From the geometry structure of your device configuration, it looks like a MoS2 nanowire rather than a MoS2 nanoribbon. Have you checked the electronic structure of the bulk electrode? Is it semiconductor or metallic?

To accelerate the convergence of scf calculation, you may increase a bit the value of electron temperature in the "electron_temperature=300.0*Kelvin,".

[qddavid]

From the geometry structure of your device configuration, it looks like a MoS2 nanowire rather than a MoS2 nanoribbon. Have you checked the electronic structure of the bulk electrode? Is it semiconductor or metallic?

To accelerate the convergence of scf calculation, you may increase a bit the value of electron temperature in the "electron_temperature=300.0*Kelvin,".

Hi zh,
Thanks for your advice, it is a semiconductor.
Can you tell me what the reason the device cannot converge, it is due to the complex electronic structure of MoS2, or the device I built and the parameters I set are wrong.

ps: maybe it is more accurate to call this decice nanowire

[Umberto Martinez]

You have 100 k-points along C in left and right electrodes but only 50 in the central region.
Anyway, you can try to increase all of them to 200.

Also, increasing the central region size may be a good idea.

[Anders Blom]

Similarly to a couple of other recent posts here on the Forum, your central region is just a perfect extension of the bulk electrodes, and you can't expect sensible results to come from a finite bias calculation of that. Since the structure is perfectly periodic from z=-infinity to z=+infinity, you cannot apply a finite bias across it in the coherent, elastic approximation used in ATK, because there is nowhere for the voltage to drop naturally, i.e. there is no (potential) scattering and hence no resistance - and the slope of the voltage drop is a measure of the "local resistance". Thus, even if you can converge this at lower finite bias, and perhaps with some tricks at 0.6 V, the results are rather meaningless, so I would not spend time on it. You need some inhomogeneity in the middle for things to make sense, like a gate, a defect, an adsorbant, or similarly.

PS: Actually the central region kC-points are not used, so you can put any number for that. But 200 kC-points for the electrode can be a good idea in many cases. It's just not really the problem here.

[qddavid]

You have 100 k-points along C in left and right electrodes but only 50 in the central region.
Anyway, you can try to increase all of them to 200.

Also, increasing the central region size may be a good idea.

Thanks. I will try these recommendations.

[qddavid]

Similarly to a couple of other recent posts here on the Forum, your central region is just a perfect extension of the bulk electrodes, and you can't expect sensible results to come from a finite bias calculation of that. Since the structure is perfectly periodic from z=-infinity to z=+infinity, you cannot apply a finite bias across it in the coherent, elastic approximation used in ATK, because there is nowhere for the voltage to drop naturally, i.e. there is no (potential) scattering and hence no resistance - and the slope of the voltage drop is a measure of the "local resistance". Thus, even if you can converge this at lower finite bias, and perhaps with some tricks at 0.6 V, the results are rather meaningless, so I would not spend time on it. You need some inhomogeneity in the middle for things to make sense, like a gate, a defect, an adsorbant, or similarly.

PS: Actually the central region kC-points are not used, so you can put any number for that. But 200 kC-points for the electrode can be a good idea in many cases. It's just not really the problem here.

Thanks for your explanation and recommendation.

I donot fully understand your explanation, it means that the central region is not long enough or it is meaningless to research the pure MoS2 nanoribbon device.

 According to your advice, I doped some W atoms in the middle and set the kC-points to 200 , but the calculation also didnot converge at 0.6V.

[Anders Blom]

A pure MoS2 nanoribbon is not a device. If you added an impurity it is more a device, but still not so much... A p-doped piece of silicon is not a device. An n-doped piece of silicon is not a device. But a pn-junction, made from a combination of the two - that's a diode!

I think your central region is too short and you probably also need to use the usual trick of first converging the zero-bias calculation, and then climb up in bias. Also try more kC points.

[qddavid]

A pure MoS2 nanoribbon is not a device. If you added an impurity it is more a device, but still not so much... A p-doped piece of silicon is not a device. An n-doped piece of silicon is not a device. But a pn-junction, made from a combination of the two - that's a diode!

I think your central region is too short and you probably also need to use the usual trick of first converging the zero-bias calculation, and then climb up in bias. Also try more kC points.

Thanks a lot.

I fellowed your advice and it worked a little when kC points increased from 200 to 300, but as the bias is 0.7V, it didnot converge again under 200 steps. 

I have no idea that how long the central region is not too short and reasonable.

Awaiting for your kind reply.

[Anders Blom]

I have no idea that how long the central region is not too short and reasonable.

Then you should make it longer. The point being - if you don't know if it's long enough, it means you haven't tried to increase it to see if that helps.

[qddavid]

I have no idea that how long the central region is not too short and reasonable.

Then you should make it longer. The point being - if you don't know if it's long enough, it means you haven't tried to increase it to see if that helps.

Hi,Dr. Blom,

I had increased the central region a little before I asked that question,  but it was no use. This time I doubled the central region, and then the calculation didnot converge even at 0.4V when it converged with a shorter central region.

Awaiting your kind reply.



 

[Anders Blom]

Are you still running the infinitely perfect system? Then it probably will not help much whatever you do. See my detailed explanation above. If not, you will need to provide more details.