Hi walker
When u did simulation at VDD=2.5V, how's your margin for VDS compared to VDS(sat), will there be chances some of your transistors are touching linear/saturation region that once there's a mis-match in process, in real silicon, some transistors are not saturated. I think you should work out for a plot for PSRR vs VDD from 2.0V to 5.5V to check out for the corner it happens, then go check the simulation for corner case. U can do a simple input offset measurement to check out the mis-match from input pairs to extrapolate if this can be happened to other transistors. Connect both inputs to Vcm and measure the VOUT to check the offset, you should know how mis-match varies
Added after 10 minutes:
Hi walker,
Something arouses me, why you think the degradation of PSRR comes form mis-match?
I think the mis-match should be a fairly constant value which does not depend on VDD. So, your guess from mis-match may not be correct.
For the change of PSRR, w.r.t VDD, in AC wise, PSRR = VOUT (change) / VDD (change), so it does not depend on absolute VDD. In real case, it does as the circuit performs differently across different VDD, so why your circuit performance changes with different VDD? it must be sth come from your circut that cannot be operable in such low VDD. So Pls do check your circuit if it really can work under such low voltage. i bet you use thick oxide transistors to do your design, since you work at 5V and now you work at 2.5V, which is normally for thin device, so your transistors' VT has already been greater, and thus you need more margin for VDS to be saturated, if your circuit has cascoded stucture, it may happen that your transistors are likely under marginal saturation region. If your circuit works fine, check for your biasing circuit then if it can really work under such low VDD.
Hope this can help you