I think you should use a folded cascode in order to avoid having your nmos input pair going into triode. I don't know how you are going to get this circuit to operate over temperature though, since Vt generally varies between 1 and 2 mV per degree celsius. If you have low-vt devices or natural nmos devices you should use them.

Hi Bharat,

if your diff pair or the load go out of saturation, this will directly affect your gain due to the increase of output conductance in the triode region. In that sense I can only confirm Icarus' statement to accept operating the tail current source in linear region. However in this case, your bias current will be supply, temperature and input common-mode dependent, which affects the gain-bandwidth product through the variation of transconductance and is not ideal either.

You may want to modify your bias current mirror so that it can work in linear regime without being subject to channel length modulation effects. For this, you need to monitor the current source's drain voltage and apply the same to the drain of the "diode-connected" mirror transistor through a cascode device, as shown in the following paper:
A high compliance CMOS current source for low voltage applications
Quarantelli, M.; Poles, M.; Pasotti, M.; Rolandi, P.;
ISCAS '03

Paul