Interesting comments, but I don't see very well how to apply them to this circuit...
1. Yes, using a NMOS diff input pair will yield greater gm for the same biasing current.
All other things being equal, the gain-bandwidth will increase. But, all things being
equal, the stability will worsen... in Miller compensation the outpust stage gm has to go up in order to push the non-dominant pole above the unity-gain frequency.
Given more input gm, everything else will need to be adjusted, I think.
2. These devices are rather wide (30um), so M=1 or M=2 does not matter, in my opinion, since each 30um device will be drawn as a multiple finger device, which can then be interdigitated with the others to improve matching.
3. It's a single ended output amplifier w/ high-gain, so , yes, it should always be used in a feedback configuration, as any other high gain amplifier. Since it's single-ended, any feedback will stabilize both differential and common mode levels. I don't see how/why a CMFB would help.
4. I meant essentially the same thing in my previous post. Yes, he/she is right to
say that there are 4 stages. but M25 is diode-connected, so only the 3 high-impedance nodes count for stability. Actually only the dominant and the first non-dominant pole matter really, and all the other non-dominant poles (such as the one in the M25 gate) will just try to make your life a bit more difficult. It may be that M25 and M18 will both appear as doublets in this configuration, in this case you'll have no stability problems from them, but your settling time wil be ruined...
In any case, follow the directions of your teacher, I'd say.
Thanks for the file, but I don't have enough points to dload it. Could you put it on rapidshare ? Is there any pdf describing this somewhere btw ?