the bandgap reference problem

[hslsc311]

dear all
i am designing a bandgap reference circuit using the classical topology,and there are two kind of bipolar transistor(A and B) provided.i get different Vref vs temp curve by using different bipolar transistor,they are both parabola ,but the voltage difference between the top point and the bottom point are different very much. the deltaV of the circuit with A is 17mV,and the deltaV of the circuit with B is 3mV. what confuse me is that the deltaV of circuit with A changes to be 3mV after i substitute the parameter tnf1 and tnf2 of A with these of B, why is that???

---------- Post added at 05:00 ---------- Previous post was at 04:59 ----------

could anyone help me ?

 

[varunkant2k]

hat confuse me is that the deltaV of circuit with A changes to be 3mV after i substitute the parameter tnf1 and tnf2 of A with these of B, why is that???

Would you tell us what is men by "parameter tnf1 and tnf2". It will be better if you can give more details about your schematic.

 

[dick_freebird]

I don't know what these params are but jacking around the
model is bad behavior for designers.

Base resistance is a significant element of curvature, the
pinched base wanders quite a bit with temperature and it
may not be the best modeled aspect, but this also has a
direct dependence on device geometry and base doping.
You might look more at the device design differences for
ideas about that.

 

[RobG]

A quick googling tells me that TNF1 and TNF2 are the temp cos of the emission coefficient (NF).

You need to tell us what your particular "classical" topology is, as well as what sort of bipolar it is (vertical, lateral, substrate, etc).

A bandgap reference with a well behaved bipolar typically has about 4 mV of curvature across -55 to 125C. The 17mV given by transistor A would set off alarm bells, especially since you can bring it back by changing tnf1 and tnf2 to be the same as Transisor B (suggesting it isn't caused by too high a current density). It could be that you are running the transistor at too high a current density so the base resistance starts affecting things, but I'd suspect a poorly behaved transistor -- perhaps it is a lateral and transistor B is a substrate?

I'd stay away from A since there is something funny about the transistor or the model.

dear all
i am designing a bandgap reference circuit using the classical topology,and there are two kind of bipolar transistor(A and B) provided.i get different Vref vs temp curve by using different bipolar transistor,they are both parabola ,but the voltage difference between the top point and the bottom point are different very much. the deltaV of the circuit with A is 17mV,and the deltaV of the circuit with B is 3mV. what confuse me is that the deltaV of circuit with A changes to be 3mV after i substitute the parameter tnf1 and tnf2 of A with these of B, why is that???

---------- Post added at 05:00 ---------- Previous post was at 04:59 ----------

could anyone help me ?

 

[hslsc311]

hello,everyone
i am so sorry for replying so late and thank you for your replyings
the image below is my circuit ,and some more details are these:
both A and B bipolar transistor are vertical (N+/P-base/HVN-well);the doping desity of p- base maybe different( for the resistance of Pbase corresponding to A is lower than that corresponding to B);beta of A is smaller than that of B( for A is 60 ,90 for B)

the startup circuit is deleted

 

[RobG]

Is your "flat spot" at the same temperature for both transistors and the "delta V" that you speak of is just the curvature? or are you adding a linear temp co to the circuit?

 

[hslsc311]

Is your "flat spot" at the same temperature for both transistors and the "delta V" that you speak of is just the curvature? or are you adding a linear temp co to the circuit?

1 ,yes,the "flat spot" is at almost the same temperature(25 and 30 degree) for both transistors,the delta V is the voltage difference between the top and the bottom of the curvature.
2," are you adding a linear temp co to the circuit?" sorry,can i beg your pardon ?

---------- Post added at 04:22 ---------- Previous post was at 04:09 ----------

hello ,dick_freebird
thank you for replying,why is that "jacking around the model is bad behavior for designers"? could you share your opinions?
and you said "you might look more at the device design differences for ideas about that",but as for me,i am always made to look into the model by the device difference,if i should avoid reading model file,what can i do facing the device difference?

 

[RobG]

hslsc311 - 1) About 4 mV is "normal" for a bandgap over the -55C to 125C range. I'd steer clear of the one with 17 mV curvature as there is something funny about that transistor, plus it gives worse performance. 2) I was asking if the extra 13 mV was coming from a PTAT (proportional to absolute temperature) voltage being added to the output by not perfectly balancing the PTAT and Vbe terms, which would be the case if the flat spot moved. In that case, the extra delta V would not be coming from the added curvature.

I disagree with Mr Freebird about changing the models - I think you did the exact right thing in figuring out which terms are causing the non-ideal behavior. In fact, I think the best way to learn about bandgaps it to zero all the non-ideal terms and then make your design match hand calculations. Then start adding the non-ideal terms (like base resistance) and understand how they affect the results. In a properly designed reference the non-ideal parameters will have minimal effect on the performance.

HOWEVER, MAKE A COPY OF THE MODEL FILE and MAKE SURE YOU DON'T TOUCH THE MODEL FILE THAT EVERYONE USES! Also, make sure you point back to the original model file.

 

[hslsc311]

hello RobG
thank you for your replying,and thank you for reminding me of making a copy of the model file.
i dont quite understand your second point "I was asking if the extra 13 mV was coming from a PTAT (proportional to absolute temperature) voltage being added to the output by not perfectly balancing the PTAT and Vbe terms, which would be the case if the flat spot moved. In that case, the extra delta V would not be coming from the added curvature."
i am wondering how the PTAT and Vbe terms are balanced?

 

[RobG]

A bandgap works by balancing the Vbe term (negative temp co) with a PTAT term (positive temp co). When you weight the terms properly the linear portion of the temp co disappears and all you are left with is the second order temp co (curvature) and a small amount of higher order terms.

In the circuit you show you probably adjusted the PTAT term by changing the ratio of the two resistors until it was balanced with the Vbe term.

 

[hslsc311]

hello,RobG
the Vref vs temp for both circuit are parabola curvatures,does that mean the Vbe terms are properly banlenced?

 

[RobG]

hello,RobG
the Vref vs temp for both circuit are parabola curvatures,does that mean the Vbe terms are properly banlenced?

Yes, that sounds correct to me.