The Designer's Guide Community Forum

The Designer's Guide Community Forum
https://designers-guide.org/forum/YaBB.pl
Design >> Analog Design >> bandgap question
https://designers-guide.org/forum/YaBB.pl?num=1119394198

Message started by driveforce on Jun 21^st, 2005, 3:49pm

Title: bandgap question
Post by driveforce on Jun 21^st, 2005, 3:49pm

Hi,

I designed a bandgap circuit, and ran temperature sweep at different conrers. The output voltages were distributed from 1. 18 (at one corner) to 1.22V (at anther corner). How to minimize this range to 1.2V +/- 1%, say, 1.188 to 1.212. Any guidelines and suggestions?

I noticed the output voltage was more dependant on the resistor's corner than other factors.

Thanks advance!!!
-driveforce

Title: Re: bandgap question
Post by ericjohnson on Jun 21^st, 2005, 5:25pm

Did you use an op amp to regulate the biasing? Also, here are some of my thoughts.
Basically the reference voltage can be expressed as:
Vref = Vbe + (R2/R1 ).delta(Vbe). You really couldn't do much about Vbe term except for choosing the appropriate current level. If you have some kind of regulation, then you have better control on the PTAT current. And a good layout will significantly improve the matching for the BJT pairs and the resistor ration. By the way, what kind of resistor did you use?

Just my 2cents.

Eric

Title: Re: bandgap question
Post by driveforce on Jun 21^st, 2005, 7:47pm

Hi, Eric,

I used an opamp.
In the typical corner, the variation is small (less than 1mV). The problem is the difference between differnt corners. Vbe changes a lot corner by corner, but delta (Vbe) doesn't change.

Any other comments?

Thanks,
-driveforce

Title: Re: bandgap question
Post by sheldon on Jun 21^st, 2005, 8:54pm

Driveforce,

Is the current result unreasonable? From your description, you are
using a typical bandgap core, ratioed devices with a resistor in the
series with the emitter of the larger device stacked on top of a
resistor. Is this the structure you are using? In that case, you are
matching a transistor to a resistor.

Rptat = Vt * ln C / [ Iptat] , where C is the device ratio 8(?)

So you are achieving about +/-2% untrimmed with +/-10 % [or more]
variations in your components. Seems like you are doing pretty good
already. Also from "Voltage References" by Gabriel Rincon-Mora,
page 122, the expected untrimmed accuracy of a PTAT source is
2%-3.5%.

Seems like you need to consider trimming to meet your target.

Best Regards,

Sheldon

Title: Re: bandgap question
Post by ericjohnson on Jun 21^st, 2005, 9:41pm

It's nice to hear from you again, Sheldon.

As Driveforce mentioned, the large variation of Vbe causes the Vref variation, but Vbe itself is determined by the process of which the designers don't really have control.

Could you provide more information on how the trimming can be performed? Is this done at wafer-level? How do the manufacturers (such as ADI or TI) perform the measurement and adjustment? Thank you.

Warm regards,

Eric

Title: Re: bandgap question
Post by driveforce on Jun 21^st, 2005, 10:14pm

Sheldon and Eric,

Thanks for your reply.

The current is reasonable (with +/- 20%) at different corners, and all the transistors are in saturation.

The structure is almost like what you said, but with 2 additional resistors are connected at the input of the opamp respectively.

by the way, what is trimming?

-driveforce

Title: Re: bandgap question
Post by ericjohnson on Jun 21^st, 2005, 10:49pm

Trimming is a way to add or reduce resistance after fabrication. Usually this is done with laser. I think this technology has existed since the 60's...

Title: Re: bandgap question
Post by driveforce on Jun 22^nd, 2005, 9:14am

Eric,

Will this be done for each commercial chip? won't this be expensive?

-driveforce

Title: Re: bandgap question
Post by Paul on Jun 22^nd, 2005, 1:07pm

Hello everybody,

trimming is indeed relatively expensive in production and that's what makes the cost of very precise voltage references. What is important when considering trimming is that you must be able to perform the compensation by measuring the voltage at a single temperature (ideally room temperature of course). If you need to measure the voltage at more than one temperature, you need to perform interpolation and the trimming algorithm becomes complex, thus even more expensive.

By the way, you should be aware that corner simulation does not take into account device mismatch, which will directly affect the stability of the bandgap voltage over temperature. The BJT matching is intrinsically good, but you should have a look at the influence of resistor and MOSFET matching on your Vbg(T) characteristic. Some people use mismatch compensation schemes, like chopping or auto-zeroing, to solve this problem, but it needs some area-expensive low-pass filtering.

Paul

Title: Re: bandgap question
Post by ericjohnson on Jun 22^nd, 2005, 6:49pm

Yeah, trimming will certainly increase the cost. I'm just curious how the IC manufacturers measure and perform trimming for a large volume production... Anyone could provide some details if you have experience?