A low drift voltage regulator design (question)

Thread Starter


Joined Sep 13, 2011
I have a need for a +12Vdc, 20mA power source that is stable over temperature (it is a lab instrument, so it will likely see between 25 and 40C). It doesn't need to be precise or trimmed, it just needs to be the same voltage as temperature varies. I don't have a drift spec, I'm just trying to come up with a circuit that has as low of a drift as is feasible. I've Looked at the LM317 and it looks to be in the 500+ppm/C range (estimating the T/C of the voltage ref from the temp curve times the effective loop gain), the LM723 is 150ppm/C, but I wonder what could be done to get a lower drift.

The attached schematic is something I ginned up to see if I could improve on the COTS parts (the circuit is not built yet, its a strawman at the moment -- I *think* I got the opamp polarity right). The divider resistors look to contribute about 12ppm/C of drift if their T/Cs are at opposite extremes. The op-amp T/C would be about 0.72ppm/C (0.15ppm/C input offset drift * the effective loop gain).

The voltage ref T/C gets multiplied by the effective loop gain, which is about 4.8 here. Adding up the drift results gets me at about 109ppm/C (assuming that superposition holds).

So, the questions become: are my calculations at least in the ball park, and what improvements could be made to reduce the T/C of the circuit? If the basic circuit is valid, what compensation would be needed to clean it up?

Note that part of this effort is to come up with a circuit, while the other part is to learn something about T/C (so just buying something off the shelf only solves part of the problem).





Joined Nov 30, 2010
There'a always the EZ-Bake oven method...Put the whole circuit in a box and thermostat the box at 50C. No delta t = no delta t per centigrade.

It's a very small circuit. Wouldn't require kilowatts to bake it.


Joined Mar 6, 2009
You might check out the LM136 datasheet from National Semiconductor or Linear Technology. The data sheet shows a claimed low drift circuit configuration using the LM136 in conjunction with a 3-terminal regulator.



Joined Aug 21, 2008
You have the opamp connected correctly.

The base resistor of the 2N2222 doesn't have to be precise, just a couple of hundred ohms to keep it from turning into an RF oscillator, though you are probably also safe by omitting it all together.

Putting the whole thing in an oven, as suggested by #12, is certainly the best way to improve stability. Just be careful of the design of the oven, thermostat, and most importantly the selection of the insulation, lest you find your voltage reference going up in flames (ask me how I know ;-)

Thread Starter


Joined Sep 13, 2011
While the oven approach is not without merrits, it is rather brute force and I would prefer to avoid its consideration as that would limit the utility of the result. At the moment, I am designing for a piece of bench equipment, so an oven isn't out of the question. However, this is not always the case, and a T/C circuit can be used in places where an oven is bad ju-ju.

To that end, I ordered and received my LM336 devices and spent the day nursing the temperaute chamber to get data over a range of temps. With only a rudimentary trim of the circuit offered by tnk (I am currently only testing the LM336 portion), I was able to get a T/C on the order of a few 10's of ppm (as expected, the V(T) curve is not linear) compared to something like 70+ppm for the TI LM285's I have available (which are supposed to be 20ppm or better). I'm going to try an actual trim pot tomorrow (I have been hoping for a non-pot solution, but I realize that isn't necessarily an optimistic outcome) to see what results I get. After that, I will assemble the whole enchilada and see what happens with the LM317. Of course, with that, I'll need to look at a range of load currents too, just for completeness.



Joined Mar 8, 2007
LM399 - 6.95V Temperature stabilised Precision Refference, from National / LTC comes with a heater that requires a supply of 9-40V, with about 0.0001% per deg C, or 1ppm, with a warm up time of 5 secs @ 30V heater supply. No, No, it is TO-46 package, and, some variants are available too.
BTW, will the ppm get multiplied by Loop Gain?