bandgap

Apr 18, 2010 8 Replies

Vbe has kind of mystified me for a long time. I finally dove into a good textbook and picked apart the equations.



This is actually the math for a bjt (diode-connected), and it has a term related to base doping level, so it may not apply to a diode that consists of a single pn junction.



But that's just as well; it makes more sense in most applications to use a diode-connected transistor anyway.



The expression for Vbe is a linear function of the log of absolute temperature.



Vbe = Vg + (kT/q)( ln(E*Ic) - (4-n) ln T)



where Vg = band gap voltage for silicon



k/q = 86.17 microvolts per degree Kelvin (Boltzmann's constant divided by electron charge)



E is a bunch of device parameters bundled together that doesn't depend on temperature, according to the text. I got an expression for E in terms of the fundamental device parameters but it's nasty.



Ic = collector current


-n is the exponent in the power law that relates average electron mobility in the base to absolute temperature. A possible value for n, according to the text, is 0.8



Taking the derivative, dVbe/dT = (k/q) (ln(E*Ic) - (4-n) (1+ ln T))



Ok, so the temperature coefficient is still a linear function of ln T. Now the question that always interested me was, what happens to the temperature coefficient as the temperature varies across a range, say -55C to 125C? According to the equation, tempco changes 166uV. If the Vbe is 2mV per degree Kelvin, then diode tempco changes about 460 ppm per degree centigrade. It changes about 8 percent from -55C to 125C.



So for example, say you are using a diode-determined voltage or current to cancel a thermal-voltage determined source in a bandgap circuit.



The thermal voltage is perfectly linear over temperature, in fact it is directly proportional to absolute temperature.



The fact that diode voltage is not quite perfectly linear points up a systemic error in the bandgap circuit architecture, and explains why it may not be possible to get a perfectly flat voltage or current reference with respect to temperature from a bandgap circuit, even in theory. But it's still pretty darn good. And there are ways of compensating even the small curvature in a bandgap reference, apparently.


It never ends.


Use a micro to take care of all that! :P

Jon

Most bandgap voltage references have an "upside-down" voltage VS temperature coefficent, apparently the peak is near 25C. Read "What's all this VBE stuff, anyhow" bi His Nibs Bob Pease. Be advised somewhere there is some talk abut compensation of that curve and furthermore that the compensation is not all that swift - so the curve has a curve which seems to have yet another(??).

A micro would be a lot more than is needed for the job but the idea of using more and more transistors to correct for the curve can bring it down to smaller and smaller errors.

If you really-really want a stable reference, put it in an oven. You can hold the temperature to within a few degrees and thus only end up going over a very small part of the curve. To get this to work very well, the temperature control needs to be fairly good.

Yeah, it's kind of like the iteration process you get into solving a transcendental equation I guess. Or calculating the infinite series for some value. You gan go through n terms and get n-th order accuracy, but you never quite get to the theoretical finish line.

Back on the subject: on further reflection, the argument of a log function should be dimensionless, so there's something suspect about those equations. I think I'll try to rework it. I could use Boltzmann's constant and the bandgap voltage to cancel degrees Kelvin. Vg*q/k equals about 13 thousand degrees Kelvin. I can use that as the new base unit for temperature. Call it Ta, for apocalypse temperature. Then temperature expressed in terms of that base unit would be dimensionless, and ln (T/Ta) would make sense.

I'll see if I can find the Pease book in the local University library, or maybe look it up on Amazon.

It's a freebie:

formatting link

John

Wonder why it is you get (almost universally), when measuring Vbe at a constant operating current, -2mV/°C ?:-)

[snip naive student blah-blah-blah]

A better equation is...

Vbe = kT/q * ln(I/IS) + IR

Where 'IS' is an _extraordinarily_ complex function of temperature. ...Jim Thompson

| James E.Thompson, CTO | mens | | Analog Innovations, Inc. | et | | Analog/Mixed-Signal ASIC's and Discrete Systems | manus | | Phoenix, Arizona 85048 Skype: Contacts Only | | | Voice:(480)460-2350 Fax: Available upon request | Brass Rat | | E-mail Icon at http://www.analog-innovations.com | 1962 | The only thing bipartisan in this country is hypocrisy

Yes! a 2-terminal voltage reference with a micro inside...

But, a relatively simple function of emitter area... I'd argue that the temperature dependence of almost ANY system is pretty complex; that 'IS' temperature dependence isn't any more complicated than, for instance, the voltage on a battery cell.

The difference between a $3 mercury cell and a $60 reference cell is a design change that reduces atmospheric pressure dependence...

Join the Discussion

Have something to add? Share your thoughts — no account required.

Didn't find your answer?

Ask the community — no account required