The Designer's Guide Community Forum

The Designer's Guide Community Forum
https://designers-guide.org/forum/YaBB.pl
Modeling >> Semiconductor Devices >> Subthreshold thermal noise model in bsim3?
https://designers-guide.org/forum/YaBB.pl?num=1109959152

Message started by 101questions on Mar 4^th, 2005, 9:59am

Title: Subthreshold thermal noise model in bsim3?
Post by 101questions on Mar 4^th, 2005, 9:59am

Did anybody see that deep subthreshold noise model in bsim3 shows about 3 times lower noise than it should be?
I compare a weak inverted MOSFET in linear mode with equivalent size resistor. I'd expect their noises to be equal, as measured at drain. When noimod=3, MOSFET is 3 times less noisy. Interesting enough, noimod=2 gives correct result.
Is this a known deficiency of bsim3 noise model? To see it the inversion should be really weak, like rds=1 GOhm or more.

Title: Re: Subthreshold thermal noise model in bsim3?
Post by Min-Chie Jeng on Mar 4^th, 2005, 12:14pm

:)
The difference between noimod=2 and noimod=3 in bsim3
is in the channel thermal noise. noimod=3 uses regular SPICE-like
thermal noise [8kt/3*(gm+gds+gmb)]. noimod=2 uses
the inversion charge (Qinv) to calculate the channel thermal noise. I think the SPICE-like channel thermal noise model (noimod=3) was derived for strong-inversion region.
In the deep-subthreshold region, gm, gds, and gmb are extremely small and may under-estimate the noise.

noimod=2 (the bsim3 noise model) is based on the inversion charge, which is more applicable to all operation regions.

The SPICE-like channel thermal noise model is included in bsim3 (and other bsim models) are for backward-compatibility with older MOSFET models. You can consider this is a defficiency of noise model from previous model generation.

Title: Re: Subthreshold thermal noise model in bsim3?
Post by 101questions on Mar 4^th, 2005, 12:34pm

Sorry, I messed things up. Its just the opposite to what I tell in my post.
What I see is that noimod=2 (bsim3 inversion charge based model) under-estimates noise.
Old spice model (noimod=3) gives correct results.
So the newer is not necessarily better in that case.
Any explanation?

Title: Re: Subthreshold thermal noise model in bsim3?
Post by Min-Chie Jeng on Mar 4^th, 2005, 1:07pm

If possible, please send me the model file you used for the
simulation. I'll see if I can find some explanation.
In many cases, the simulation results have to do with the
model parameters used. You are also right that new enhancement in the model is not equivalent to more
accurate results. All models have valid ranges based on the parameter values given.

Title: Re: Subthreshold thermal noise model in bsim3?
Post by 101questions on Mar 4^th, 2005, 1:24pm

It's a TSMC 0.18um process. Most probably you guys at Cadence have its model in your custody. I'm binded by NDA and can not send it.
As for the test circuit, it has nmos_3v transistor in weak inversion, with source at gnd, drain floating. The drain noise is compared with a noise of equivalent size resistor. Gate bias is chosen to have rds~10 GOhm.
Let me know if you need my test netlist.

To make sure I also tested it in another fab process with exactly same results. To me it looks more like a inherent model bug, rather TSMC-specific one.

Title: Re: Subthreshold thermal noise model in bsim3?
Post by Paul on Mar 4^th, 2005, 1:55pm

Hi,

I'm not sure about what I am going to say, but I think the noise parameters are extracted for a given type of model (defined by noimod) and you probably cannot simply change the noimod parameter and expect the noise parameters to remain consistent. Maybe you could state on what was the initial noimod value in your model. Maybe Min-Chie can also confirm or negate the above statement.

Paul

Title: Re: Subthreshold thermal noise model in bsim3?
Post by 101questions on Mar 4^th, 2005, 2:17pm

Thermal noise is not extracted. It comes from a bsim3 model equations directly, based on other extracted parameters, like gm, gds and some internals. So you can change noimod freely.

Well, not exactly freely when it comes to 1/f noise. This one is extracted and its parameters are noimod-dependent.

Title: Re: Subthreshold thermal noise model in bsim3?
Post by Paul on Mar 16^th, 2005, 6:04am

101questions,

Of course you are perfectly right, I messed up with flicker noise. But nonetheless, I had a look at the Spectre Device Model Equations manual and I noticed that for noimod=2 or4 (inversion charge based model), the exact version of the model is important and also the capmod parameter. I did not go through the equations, maybe you can have a look at them yourself. I would suggest to have a look at how the inversion charge is calculated in subthreshold regime.

I hope Min-Chie could provide more helpful information.

Paul