Please refer to the attached circuit.
It's obvious that MP1-3, R1&R2, Q1&Q2 work as a bandgap reference circuit itself.
With appropriate combination of the PTAT current and the CTAT Veb voltage drop,
a desire bandgap voltage reference can be produced.
However, what about the middle part?
It seems like there's an amplifier over there, but I can't understand how it works.
could anyone elaborate it? :shock:
Not exactly.
In your schematic, output node of the opamp is simply the VBG.
besides, I guess this opamp's power will be connected to VDD as usual.
While in my case, it's kind of complicated to point out how the opamp works.
As shown in my circuit, the opamp's power is coming from VREG node, which
might be a variable potential.
It's a 2-stage Miller-compensated opAmp (MN2-MP5 and MN3-MP6), its output (MN3-MP6) provides a relatively well regulated supply voltage VREG (s. below) to parts of itself and to the bandgap circuit, like in my image schematic above.
VREG is relatively well regulated, because the whole circuit is defined via (multiplied) currents: W/L(MN4) > (or >>) W/L(MN2), via MP7 & MP6 the whole amplifier and bandgap circuit is supplied by this multiplied current. In steady state condition, in 1st approximation this translates to a voltage division between VDD and VREG, because both amplifier and bandgap circuit are supplied by constant currents, too, i.e. VDD changes will be attenuated by the corresponding ac division factor, where the amplifier's output impedance might even be lower than VDD's (?).
update: 3rd May 2015 at 17:17:
Probably my last sentence (with the ?) is not correct, but the amplifier itself will show some PSRR (not a very great deal, but anyhow), which helps to reduce power supply variations.