Hello all,
I am writing my first post in this useful forum starting with a question. Here is the background:

I need to design a simple continuous time comparator, so I started with the basic two stage comparator described in the Allen/Holberg book. I use Spectre Monte Carlo analysis to simulate the offset. Here are the 1 sigma results with different aspect ratio of the input pair, but same area:

W/L           1 sigma
-----------------------
4/5      ->  1.59mV
10/2    ->  1.39mV
20/1    ->  1.35mV
40/0.5 ->  1.34mV

My intuition tells me that longer transistors or those biased in a strong inversion should exhibit less random / delta Vt mismatch, as it is in a current mirror for example. The question is why I get just the opposite effect here?
Any input is greatly appreciated.

Thanks!

In my understanding, the operating point should also influence the spread, not only the transistor area. For example, larger Vgs overdrive basically has a positive effect on the Ids spread, while it seems to be the opposite for a differential pair.

Thanks Aaron, I found the chapter in Razavi’s book. I follow the equations, but honestly I am still having hard times to grasp what is behind. Why lowering the Vgs-Vt would help us to reduce the offset. In my understanding low overdrive voltage would make the input pair more sensitive to Vt variations, which in turn would reflect in larger current variations. Obviously I am missing something very basic. Can someone help me with somewhat more intuitive explanation?