VTH vs. W

[claudiotalarico]

The change in VTH when sweeping W seems to be excessive (especially going from 5um to 10um).
Please see the attached plots (the device used for this analysis is the SG13 LV nmos)

The xschem's file and the matlab's file I used to generate the analysis are provided at this link

[bmurmann]

Below is a plot from the measured data:

It is expected that Vt is a strong function of W near the minimum allowed value (due to narrow width effects). The above data also shows this. But I would not expect is that Vt changes dramatically as the width changes from ~1-10um. The Vt value usually stabilizes for large W as shown here: https://www.semanticscholar.org/paper/Implications-of-Small-Geometry-Effects-on-%24g_m-I_D%24-Ou-Ferreira/7982fbeb3a43c594a244923438360aa5ebd1b478/figure/4

Looking at the measured data, I see identical Vt for W=0.6 and W=2um, which makes sense to me (but the current models do not reflect this). Perhaps the measured data (with potential outliers) was overfitted to generate a non-physical trend in the spice models?

[dwarning]

I grubbed out few old ngspice scripts to extract Vth over W and L and rebuild them with the help of Holger.
To extract the threshold voltage a simple constant current criteria (W/L*1e-6) was used and follows in deviations compared with the IHP documented behaviour. But it confirms clear the discussed characteristic and that the models reflect the measured and documented values.
lv_nmos Vth over W with L as parameter 0.13...0.16um in linear range:

For completeness also lv_nmos Vth over L with W as parameter 2...8um in linear range:

[KrzysztofHerman]

@dwarning could you please share your testcase? I would like to get the exact difference between model and the measurements.

[dwarning]

No problem, but as said there are some margin in accuracy of the vth determination.

Vth_W.txt
Vth_L.txt

Good luck

[KrzysztofHerman]

@claudiotalarico first of all we are sorry to let you wait such a long time for response. Reviewing your simulation you use W in a range up to 50u. Our models are valid up to 10u and to obtain W=50u one should use multi-finger approach. Could you please regenerate the graphics to be aligned with the one we have shared in this thread ? I mean semilogx axis for width and vertical grid of 50 mV.

[hpretl]

Maybe this issue is related to #164? Could be if the number of fingers is also changed in the investigations above.