Well, Monte Carlo simulation gives you a statistical result, and in some cases you are going to get better performing, more ideal circuits and the opposite is true as well.

Langvelt,

  Other questions to consider:
1) How reliable are the Monte Carlo models you are using?
2) What are the bias current variations?
3) Are the bias current variations strongly correlated with open-loop gain?
4) Is 60dB consistent with hand calculated values of open-loop gain?
   Or, is it possible that there is a saturated gain stage in the current
   design and under process conditions the stage leaves saturation so there
   is a big increase in gain?

Hi sheldon, croaker, bharat
I'm using models which have the mismatch information at particular
process corner. These results i am getting at typical conditions.
Range of the open loop gain with all process variation, +/-20% bias current
variation, +/-20% supply variation and -40 to 125 temp variation(without montecarlo), is 60-80db.
for ur another question about saturated gain stage, every transistor has a good
satuartion margin.
This 60db is appx. similar with the hand calculated gain.

What i am worrying is that wherever i bias this opamp i cannot get 150db kind of gain with this structure. it's
a simple single stage gain boosted folded cascode amplifier, means every transistor has a good saturation margin.

What the results generally you people get, when you do montecarlo analysis in terms of gain of the amp.

I am throwing the question:
Whatever the spread I am seeing in MC simulation, the designer owe a part of it and major part goes to process inaccuracy. If all the transistors are in saturation, what else a designer can do. I am also seeing similar results in my op-amp while calculating voltage offset. The ratio of mean value to 3 sigma value is close to 150. Designer's task is to meet the specs for mean value and some percantage in the neighborhood of it.
If the model is in a way that variation in L, W, Vth, overlapping cap, doping concentration is so much, how can really design can contain these variations and go for 'robust design'. If the expectation is that all those variations should be taken as design challenge, I am afraid if I am agree.
Also, for example, if the model considers that there may be mismatch in the device dimension, shouldn't they account for that all the devices are laid out in common centroid fashion for perfect matching with dummy devices to avoid any distortion in the devices while polishing in the process. (because I ran pre-layout MC sims)
In addition, is it worth to burn the night oil for a designer to design for such a device which has worst L, W, Vth, overlap cap, doping conc mismtach in one device which may be one in millions. Isn't it overly pessimistic and far from real life scenario.
Also, what do you say for die to die variation. How can one contain it?
Please comment if I shouldn't think this way as a Designer.