The Designer's Guide Community Forum

The Designer's Guide Community Forum
https://designers-guide.org/forum/YaBB.pl
Simulators >> Circuit Simulators >> Switch gds vs. PSD
https://designers-guide.org/forum/YaBB.pl?num=1220626273

Message started by Adam Ring on Sep 5^th, 2008, 7:51am

Title: Switch gds vs. PSD
Post by Adam Ring on Sep 5^th, 2008, 7:51am

I am running an AC Noise simulation in Spectre using bsim3v3 models of a switched capacitor integrator. The amplifier is ideal with infinite bandwith, no noise, but finite gain. Integrating the noise at the output during the integrate phase does give me the expected kT/C, however, I am trying to understand the noise spectral density of the switches. I have four switches (standard parasitic insensitive integrator) that are straight NMOS. Spectre is reporting that the gds of each switch is 800umhos which should give me 4kT/gds PSD of 2.17e-17 V^2/Hz. When I plot the total noise (I turned off all flicker noise in the models) SD of one of this switch, the flatband region of the noise is 3.58e-18 V^2/Hz.

I was reviewing the spectremod.pdf for bsim3v3 (subversion 3.24), and from what I could remember, this model uses one function to model all three regions of operation. They list the thermal noise SD as id^2= 4/3*k*T*gamma*(gm+gds+gmb) with gamma defaulting to 2/3. Using this, I would expect a flatband somewhere around 4.6e-18 V^2/Hz, which is closer but not close enough.

Finally, when I look at the 3dB point of the SD roll off and the cap size I'm using, it confirms a conductance of 800umhos. So my question is, how is Spectre calculating the SD and RMS noise? For kT/C to work, the SD of the switch has to be 4kTR, but I'm getting nothing of the sort through the Results Browser. Am I missing something here? Thanks.

Adam

Title: Re: Switch gds vs. PSD
Post by Geoffrey_Coram on Sep 8^th, 2008, 5:28am

Did you check the NOIMOD parameter for your model? The equation you cited is for NOIMOD=1 or 3. There's a correction for NOIMOD=5,6 that multiplies that SD by 1.5 - vds/(2 vdsat) when vds < vdsat; some simulators added this term before it was in official Berkeley BSIM3, so you might be getting it even for NOIMOD=1 or 3.

If NOIMOD=2 or 4, there's a different formula involving qinv.

Title: Re: Switch gds vs. PSD
Post by Tlaloc on Sep 8^th, 2008, 7:54am

Yes, I had checked that noimod=3. With that, I zeroed out the flicker noise by setting noia, noib, and noic to 0. There are still non-zero values for ef and em, but it doesn't seem like that's important. Gamma was not defined, so it should be the 2/3 default.

The spectre version that I am running is 6.2.0.420, and the model subversion is 3.24.

Title: Re: Switch gds vs. PSD
Post by Geoffrey_Coram on Sep 8^th, 2008, 10:01am

Try changing NOIMOD to 5 or 6 and see if you get the same answer. What's vdsat (and vds) for the operating point?

Title: Re: Switch gds vs. PSD
Post by Tlaloc on Sep 8^th, 2008, 1:22pm

It looked like noimod=5 and noimod=6 were identical. They gave the flatband PSD of the switch as 3.495e-18 with noimod=3 reporting 3.555e-18. So, both 5 and 6 were very, very close. There was also some small movement in the RMS noise but it was in the noise (no pun intended). :)

vdsat = 273mV
vds = 4.8pV

Title: Re: Switch gds vs. PSD
Post by Geoffrey_Coram on Sep 9^th, 2008, 5:40am

Noimod 5 and 6 differ only in the flicker noise; sorry I wasn't clear about that.

As I review what we've written, it occurs to me that we haven't mentioned the gain from the switch to the output node. Have you taken that into account somewhere?

Title: Re: Switch gds vs. PSD
Post by Tlaloc on Sep 9^th, 2008, 8:20am

I skipped over the detail in my first post. The integrated root noise at the output from simulation is 5.44uV with 25pF input cap and 60pF integrating cap, and the temp is 27C. Straight kT/C at the output would give 5.36uV or 12.87uV input referred.

Title: Re: Switch gds vs. PSD
Post by Geoffrey_Coram on Sep 9^th, 2008, 10:44am

I thought you probably had covered that, but figured I'd check.

I don't have any more suggestions on that circuit; you might have to take it back to a single-transistor circuit. Eg, try putting in a single transistor with the vgs, vds conditions of your switch set by voltage sources, then measure the current noise in the vds source and check that it matches the id^2 value.