bandgap PSRR analysis

for basic BANDGAP core (VBE+deta(VBE) structure using two bjt, a resistor, two PMOS, an opamp to force Vip=Vin), how to analyze the PSRR+ (vref/vdd)? in my simulation, in 300KHz or so, two poles appear (close to each other), and the PSRR+ becomes positive, that is, vo/vdd>1 in some frequency range (300K ~ 600K), why? what is the possible reasons? why can PSRR+ be positive?

besides, in some process, cascode opamp can not be used in the above bandgap (especially NMOS cascode due to body effect), because leakage can decrease the output resistance of opamp, so gain decreases also. i want to ask, what is the leakage path of NMOS cascode transistor due to body effect? and why leakage is related to NMOS body effect? is the leakage current caused by parasitic diode or bjt(npn or pnp)?

thanks.

I doubt the body effect (Vbs influence on VT) has to do with
leakage. You can sometimes see leakage current turn up
at higher Vsb (because Vdb is higher still). Elevated D-B
currents can be amplified by parasitic BJT action.

You can indeed have overunity vdd-out gain, you have to kill
it. Particularly troublesome are PMOS final stages and vdd-
referred current mirrors where any capacitance from gate to
ground is gained up by device gm. Stiff bypassing to the rail
can help, ideally you'd make this part of your compensation
network so that it doesn't cost you double in the forward
gain path. Miller-only comp schemes have nothing for PSRR
outside the feedback loop; shunt comp in places that also
benefit your HF PSRR is better, though much more area
consuming.

Properly cascoding current mirrors can cut a lot of injection
from the low side and stray capacitances.

thanks for reply.