Dear all,
   I use an differential-to-single open-loop amplifer with an inverter chain as a comparator. The reason why I don't use a dynamic comparator is that this comparator needs to work continuously.
   However, I found that even if the input is changing slowly (you can see from the pic, with the inverter chain delay being at most 500ps), there would be some "overshoot". I mean, vinn must be smaller than vinp by 21mV for the output to change. Moreover, this 21mV is process-dependant, which introduces significant error in my design. I want to reduce that 21mV to less than 2mV.
   Is there anybody familiar with this kind of comparator helping me? I also want to know if there is some discussion/paper/thesis on this comparator or some better structure.
   Your comment is appeciated.

ray_wang,
I expect that some of your "error" is simply comparator delay.
BOE

Hi, Mayank
   This comparator is used in an feedback loop to form an relaxation oscillator and I use the inverter chain here just to avoid metastability. And I don't think its contribution to the overall delay is significant. Here I only want to increase the speed of the open-loop opamp.
   I searched "cross coupled Level-Shifter", but I did not get what I desired. Could you provide more details?    
   After posting, I randomly swept the transistors' size and got a much better result. And I found that the result is even worse if I use PMOS input diff-pair. But to be frank, I still don't know exactly the mechanism about that "overshooting".  :-/

To BOE,
   You are right, but I believe there is systematic delay for the comparator, not just common intrinsic delay, since 2ns is too large for 0.18um devices.

To loose-electron,
   I don't know what "conventional approaches" are, can you provide me with information on continously-working fast comparator?

Thank you for your comment.

Thank you for your help!
loose-electron wrote on Jan 31^st, 2010, 9:47am: