Monte Carlo analysis on schematic Vs post layout

There are three reasons why it may be more accurate, but I suspect these are not the effects you are thinking of:

1. The physical dimensions of the devices will be more accurately captured, including the parasitic effects and layout dependent effects (things like stress effects, well proximity effects and so on)
2. Any m-factors on the schematic will be extracted as individual devices. With an m-factored device, the mismatch is applied to the instance rather than the individual devices; it is possible to model the statistical impact of the m-factor but I don't see that commonly done. However, post-layout, each device will (probably) be extracted separately and so mismatch would be applied to each finger separately.
3. You may have parasitics in the netlist as well, which should add to the accuracy of the simulation.

I suspect you are wondering whether the effects of good layout matching will be taken into account. Well, the random monte carlo variation is generally capturing the remaining random variation, and not systematic variation due to layout proximity and alignment. If there was some element of that included in the variation of the devices, there might be some benefit in providing some information about the correlation between the devices, but:

1. The models usually don't have this effect included
2. The tools don't extract anything that could be converted into a correlation coefficient
3. Even if they did, it would be hard because the random variables are often rather abstract and are combined using equations in the models to model the overall statistical variation (I sometimes see them called random1, random2 etc - with little resemblance to the physical parameter that may be statistically varying).

That said, using capabilities such as mismatch contribution in ADE Explorer/Assembler (a Virtuoso Variation Option feature, or as an ADE GXL feature when using ADE XL) can help identify the variation of which devices contributes most to the variation of the output, and typically these will be the devices for which you need to take greater care on for matching to ensure that all you're getting is the remaining local random variation and not systematic variation due to poor layout.

So post-layout statistical simulation will be useful (although it will often be slower of course, especially if you have parasitics). 

Hope that helps!

Regards,

Andrew

THANK YOU very much Andrew for that detailed and very insightful reply. It surely helps