start-up problem of bandgap

[chaojixin]

look at the figure. there is not only one 'degenerate' bias point in this bandgap and the output would settling down to a strange value in some case (the normal start-up circuit can only eliminate the zero current bias point). I couldn't find a paper discussed the start-up and 'degenerate' bias point of this type of bandgap in detail.
any one has effective start-up circuit for this bandgap?

thanks in advance!

 

[mike.zhang]

u need a switch to pull down the gate of pmoses.
the bandgap voltage or current is a good trigger.

 

[chaojixin]

thanks. but I don't know if you know what i mean: the fault 'degenerate' bias points is not only the one which is descript in many books. there are additional bias point that pmos may works in subthreshold region. the vgs of pmoses may be as high as several hundred mV which is only a little smaller that when bandgap functions well. so it is hard to discriminate this fault bias point from the correct bias point by a simple startup circuit which contain only some simple inverters and common source amp.
whatever I adjust the W/L of moses in startup circuit, there is still some corners in which the bandgap malfunctions.