Do you have any specific references?  I've seen some papers where they state the Quasi-Static models are accurate to some fraction of the cutoff frequency (e.g. 1/12).   Some NQS models work by trying to solve an additional internal continuity equation or by segmenting the transistor into smaller transistors.

Is this a digital or analog application?  Analog  applications may require more accuracy if the nonlinear mixing behavior is important.

I am not really a compact modelling kind of guy.  Tsividis has a whole chapter on large-signal modelling and discusses the QS and NQS effects.  He has another chapter on high-frequency small-signal models where he also discusses NQS effects, noise, and MOSFET modeling for RF applications.  Overall a great MOS book.

Yannis Tsividis, Operation and modeling of the MOS transistor, 2ed, 1999.

Hi Paul,

You can see a noticeable difference between QS and NQS mode at much lower frequencies if you try to simulate charge injection effects in switched-cap circuits. I tried this but I'm still not sure if I can trust the accuracy of chrge injection simulations even with NQS model.

Eugene.

My recollection is that BSIM3 has a parameter, the Elmore delay, that is used only for NQS.  If you don't have that in your model card, then you probably can't trust the NQS results.

BSIM4 uses one of the gate resistances plus the oxide capacitance to set the time constant; the gate resistance of your model may or may not have been partitioned correctly to get the right NQS result.

I happen to study something about NQS recently. Hope this will help.

NQS effect happens when input frequency is pretty high at the order of transit frequency. A resistor, which is series connected with the Cgs, is used to model the time delay between the stimulus at the gate and the response at the channel by Elmore, and the resistance value can be expressed by 1/(5*gm). Then later on, Prof. Chenming Hu published some papers, saying that the NQS effect could be best modeled by an external resistor, whose value is extracted from the y parameters. So even there are no ELM or NQSMOD parameters in your model, which is QS model. You still can use an external resistor to simulate the NQS effect. Moreover, the inclusion of an external resistor at the gate termal can give us another benefit, i.e. the gate-current noise, which is correlated to MOSFET thermal noise, through Cgs capacitive-coupling. (Why? Please go to see Prof. Thomas Lee's paper about 1.5GHz LNA.)  I believe that this kind of noise is not included in the model, neither.

---Eric

The NQS effects are important when  transit delay is comparable to the device rise/fall time. So, this become important either of these two conditions:
1.)Long channel device
2.)High frequency

For analog, either of these conditions usually apply, it is advisable to include NQS effect in the simulation.