offset of a fully differential amplifier

[jiaguoce]

Hi,

I need to design a preamp for the comparator..
the input range is 1V, for a 12bits ADC, the offset should < 120uV.
assume the offset of the preamp is 10mV, to make it smaller than 120uV, now i plan to do the autozero, and in this case, a DC gain of the preamp should > 40dB, which means ios is the proper auto-zero method.
anyone here can help me to comfirm that whether it's right?

now i first need to meansure the offset of the preamp in cadence. i use the architecture in the figure(i find it from this forum), however there is no voltage difference between the two outputs.
then what's the method to meansure it?
thanks in advance

 

[narasimhan4u]

You have to model mismatch between the typically matched transistors (like the input transistors of input differential pair) to see an offset at the output. I would do a monte-carlo simulation with Mismatch as the parameter and ensure that the mean+(3*standard deviation) value of offset is less than 120uV.

 

[jiaguoce]

when i do the monte carlo simulation, i can't see the gauss distribution, the following graph is the offset distribution.
I think maybe something wrong with the model library. TSMC0.18 is used, stat, stat_mis section are added in the library model.
anyone would help? thanks

 

[checkmate]

You have to look into the model to find out what went wrong, or maybe you forgot to activate some settings which are DK specific.
If you are lazy, just dump the worst case 10mV offset, and ensure that your auto zero can handle that.

 

[jiaguoce]

I checked the library model file......
besides stat_mis and stat, section mc could also be useful, however there will be lots of errors during simulation if i add it to the library. like "parameter "noiseflagn" was previously defined", etc.
btw...every time i login the cadence, it shows:"There is no PDK installation", this is the cause?

 

[narasimhan4u]

Are you using the "mis" models of the transistors? for instance "nmos2v_mis" and "pmos2V_mis" from the tsmc18rf library? These transistors have the mismatch variables modeled in them which will be used when running a monte-carlo analysis with mismatch.

I am not sure if "There is no PDK installation." is what is causing your problem or not. If you were already using the mismatch models for the transistors, and your monte-carlo simulation still wouldn't give a Gaussian distribution, then the PDK issue could be your problem.