The use of PN junctions to define a bandgap reference is done because the consistence of the voltage difference between two PN junctions of different geometric size and common current is extremely accurate.

Notice that I say "difference between two PN junctions".

If you go do the math on it you will see that all the process variance falls out of the equation and thus makes for a very accurate reference. That is with both PN junctions on the same piece of silicon.

Not the case with MOS threshold voltages. MOS threshold variance is much much worse.

Bandgaps on CMOS processes are generally done with the PN junctions that are available on the silicon.

Ask good questions and you will get good answers.

Eliot wrote on Jan 3^rd, 2015, 4:34am:


That device totally ignores threshold variance, which is very significant.
Also, no silicon was evaluated just simulations.

Novaris wrote on Jan 5^th, 2015, 5:46am:


That is not the thing I have a problem with. Vth has a huge amount of variance in it, and also a large amount of variance from device to device on the same piece of silicon. In the real world that will kill the design.

That is the reason that the majority of bandgaps are done using PN junction ratios.

In the real world.

In academia? It depends on the professor, if they got no real world experience you might be able to pull it off, but if you hear the phrase "have you considered...." then you are dead.

loose-electron wrote on Jan 3^rd, 2015, 2:49pm:


Good info. I will add that the temperature variation due to process variations in the bipolar and the resistors and also some mismatches can be eliminated by trimming the output to a known voltage at a pre-determined temperature of moderate accuracy. In other words, bipolar references can handle process variations easier than other types.

With a CMOS reference the VT variation messes things up and you require measurements at multiple temperatures to trim out the temp-co. I will leave it as a research project for the interested parties to figure out why this is .

The classic paper on bipolar BGRs was written by Brokaw. It provides a good understanding of the fundamentals. Rincon has also published more state-of-the-art papers. Vittoz published an all mos reference quite a long time ago. These are essential readings.

Also research Makinwa's work. He uses an alternative to all cmos and bipolar, but I don't know much about it.

When you find things out please update us. I have never been a fan of the CMOS references on the assumption that process variations are too hard to deal with, but I have never done an in-depth investigation.