Raj - the main reason people use the 8:1 ratio is that the person before them used an 8:1 ratio  ;D

A bigger ratio will help noise. In fact it will help just about everything. You can also get a bigger ratio by scaling the currents in addition to the area. Also, at some point the current density in the "big" bipolar becomes low enough that it starts to behave non-ideally. But up to that point, bigger ratio is better...

A more efficient way to accomplish this is to put the bipolars in series to generate the PTAT voltage. If you put two 8:1 diodes in series the voltage would be 2*Ut*ln(8) which is the same as if you had used a 64:1 ratio. The bandgap in fig 3 of my paper <click here> uses Darlington connected devices to get the increased PTAT voltage.

raja.cedt wrote on Feb 13^th, 2012, 2:50am:

Thanks for the compliments.

ppm will depend on your temperature range. You can expect about 4mV drop from 30C to 125C (or from 30C to -55C) in a perfectly tuned bandgap, plus whatever PTAT error you get from mismatches and process corners.

Quote:

I'd say the biggest mistake people make is underestimating the requirements of the opamp. Your opamp will be configured in a gain of 8 or so, so all of your parameters that depend on loop gain will reduced by 18 dB. Power supply rejection, load rejection, and systematic offset are probably the ones that will hurt the most. And for the same reason, your offset and noise will be gained up by that factor of 8. (Using a high ratio of current densities like we talked about will reduce the amount of gain needed, which is why it is so important.)

There are two schools of thought on bandwidth - way lower than the load bandwidth so the output won't move with transient loads, or way higher so that it will settle quickly. To get a low bandwidth you'll need a huge compensation capacitance, or you'll have to lower your loop gain which will reduce your PSRR, load rejection, etc. I can't think of a case where I've opted for the "low bandwidth" solution unless I had an external capacitance.

Good luck!
Rob

raja.cedt wrote on Feb 15^th, 2012, 2:20am:


I was talking about loop gain. For example, if your opamp loop gain is 80 dB, when you use it in a bandgap where the delta-vbe term is gained up by a factor of 8 (18 dB), then the loop gain will be 62 dB.

Quote:

You can trim it if that is what you mean. It is easier to explain with a schematic, but if the offset of the opamp is zero then the bandgap can be trimmed to the same voltage at room temperature to achieve about 0 ppm. It is a good exercise to show that all the mismatches and corners (except offset) result in a PTAT error and can be trimmed out by adding a PTAT term. Trimming can be done by changing the resistor ratio or the current in one of the bipolars. You can't just put an adjustable voltage divider on the output because that won't add a PTAT term.

CMOS opamp offset is not PTAT so it can make it difficult to get 0 ppm at room if it is large.

You can also add curvature compensation circuitry, but I'm not sure that you can do a single temperature trim to get 0 ppm. (The technique I've been using allows this, but it is patented.)

There might be a tutorial out there that explains this better than I can. Maybe Rincon-Mora?

Best,
Rob Gregoire

mists wrote on Feb 20^th, 2012, 7:51am:


First you figure out what the voltage is at the output that gives you zero TC. Usually you have to do this with actual measurements. It will be about 1.23 V. This is often called the "magic voltage" because no matter what your bipolar or resistor corner is, or even what yout bipolar or resistor mismatch is, you can trim the bandgap to this point and it will bring it back to 0 TC. Note that your trim circuit must add/subbtract a PTAT voltage to accomplish this. This can be done by trimming the resistor ratio.

Google "Bandgap Magic Voltage" for more information on this.

mists wrote on Feb 21^st, 2012, 3:51am:

yes