Bandgap Reference BJT Questions

[ytliang]

npn is 5umx5um

Does it matter what kind of BJTs one would use in a bandgap reference design? Does beta and other parameters matter? What about the size of the BJT? (The size of a single BJT, not the relative size between the 2 BJTs)

Thanks,

 

[haff99]

Depending on exactly what type of Bandgap architecture you use beta can be issue. With a low Beta you have more base current. This base current error can contribute to the overall accuracy of your bandgap. THere are ways to cancel this out. Depending on exactly what you need it may or may not be an issue. If you are making a classic bandgap in a CMOS process using the parasitic PNP's beta it isn't an issue. If you are making a bandgap using NPN's it may be an issue in high accuracy cases.

The size of the bandgap, depending on your process may be an issue. It really depends on the matching of your process. A 5umX5um emitter is typically going to match better then a 1umX1um emitter. So the matching affects your accuracy.

Also if you plug your NPN devices or not. That affects your bandgap as well in terms of performance.

 

[palmeiras]

Hi haff99,

Please, could you explain, why low-beta BJT can be a problem in the accuracy of the Bandgap Voltage Reference?

for device Q1: VBE1 = VT*ln(IC/IS); => VBE1 = VT*ln(alpha*IE/IS)
for device Q2: VBE2 = VT*ln(IC/n*IS), ==> VBE2 = VT*ln(alpha*IE/n*IS)

Then cancelling alpha of these two devices...

deltaVBE = VT*ln;

So, this make me to conclude that low-beta is not a problem.
Please, could you explain where is my error?

Thanks very much,

Martini

 

[jc2]

When you use PNP transistors in a CMOS process, you are not going to have problems with the BETA. But, when you use the widlar source using NPN transistors, you are going to have problems with the BETA, because the collector current is not equal in spite of the mirror: part of the current goes to the short between the collector and the base.

 

[dick_freebird]

Diode-connected BJTs in my experience have a better (more ideal)
log-linear characteristic across wider range of current. This to me
makes them better than plain diodes for reference elements. You
want mainly to operate the mismatched-current-density pair in the
region where neither low- nor high-current nonideality is showing.

Active BJTs do introduce base current error and curvature, and these
are best run where beta is decently high but base current remains
low enough to not push the curvature up.