Lowest mismatch variation bandgap?

[endru]

bandgap design npn

Hi,

Can anyone suggest me, which bandgap architecture that gives the lowest variation due to mismatch? Any paper? Thanks.

 

[mpig09]

bandgap mismatch

Dear endru:

I think design and layout plan are critical points to reduce mismatch of circuit , the architecture is not a very important.

mpig

 

[PaloAlto]

First of all. NPN bandgaps are always better than PNPs. If your process is bipolar or BiCMOS go for it. Now, there are several mismatch sources in a BG:

* The bipolars
* The amp
* The PMOS current mirrors
* The array of resistors

You can do now several things:

* Of course, a good layout. In general do not use minimum size devices
* Large ratio in the bipolars
* Big PMOS current mirrors. Specially large L.
* Large input pair of the amplifier. Specially large L.
* Use a unit resitor to build of your resitor values.

 

[Fom]

Why NPN bandgap is always better than PNP?

 

[PaloAlto]

Because of the amplifier used to create the PTAT voltage.

The NPN bandgap loop is closed w/o amplifier. You have gnd->vbe->n*vbe->Rptat->gnd, so, the only error is the NPNs mismatch.

The PNP bandgap loop is closed with the amp. Then gnd->vbe->Voff_amp->n*vbe->Rptat->gnd. You also have the amplifier error.

Those errors will always be there, no matter what you do later. In one case the bipolar mismatch, in the other the bipolar mismatch + amplifier offset

 

[Fom]

I don't understand that:

gnd->vbe->n*vbe->Rptat->gnd

without schematic.

May be you meant process with NPN having separated collector that can be connected anywhere unlike PNP those collector always connected to P-substrate.
But it is not standard CMOS process.

 

[PaloAlto]

If it is not standard CMOS process and you can create the circuit with the PNPs, the Rptat and no amplifier, then it'd be the same, just the bipolars mismatch

 

[endru]

Thanks a lot guys! I agree with you.
I also found that if we stack the BJTs, the mismatch effect can be reduced. But of course with the trade-off of area..