how to simulate the offset of fully differential amplifier

[taeyui]

Hi guys,

I tried to simulate the offset of fully differential opamp but the result looks something wrong.
When I simulated Monte Carlo for "Mismatch only", that result is reasonable standard variation value, about 8mV,
but in case of "Process only", that result about zero.(~ 70e-23)

I guess there are something I missed.

here are simulation environment



I induce VDD=2.7 and VCM=1.35, simulated difference btw INN and INP.

Is there something wrong for unity gain feedback? or simulation schematic?
(actually I also simulated in same way with ideal Diffamp which has a offset(setting parameter 5mV) but it has 5mV offset



One more question,

I appreciate if you explain easily what's difference between systematic offset and random offset.

 

[erikl]

... in case of "Process only", that result about zero.(~ 70e-23)

I guess there are something I missed.

Yes. Perhaps the process mismatch variance parameters are missing in your model file? Compare with those from single-wafer mismatch!

 

[taeyui]

Hi erikl,

Thanks for your reply first.

I also doubt that I miss the Monte Carlo "process model" but I checked another simulation for Vth variation.
I attached about this below and I confirmed it works properly in same model.

thanks.

 

[erikl]

I also doubt that I miss the Monte Carlo "process model" but I checked another simulation for Vth variation.
I attached about this below and I confirmed it works properly in same model.

Vth variation alone doesn't guarantee that your process MC model file also contains offset variance parameters.
I'd suggest to check the process model file itself for such parameters.

Here's an example: View attachment Cadence_Monte_Carlo_simulation_tutorial_p33.pdf
From nomenclature I can't recognize offset variance parameters, either.

 

[dick_freebird]

A well designed amplifier -should- have trivial Vio variation
with PVT and one which has any sort of autozero function
may display no measurable Vio even with gross VT scatter.
You need mismatch statistics enabled (by you) and modeled
(by foundry, or roll your own) to see realism in Vio. The PDK
docs ought to tell you the series of motions required.