LM399 based 10 V reference

[David Hess]

Did you use the offset voltage adjustment terminals of the LT1001?  They will produce an offset voltage drift of 1uV/C for every 300uV of offset away from zero.

From where do you get those numbers?

They on page 7 of the application section of the LT1001 datasheet:

Trimming to a value other than zero creates a drift of (Vos/300)µV/°C, e.g., if Vos is adjusted to 300 µV, the change in drift will be 1 µV/°C.

At one point when I was doing precision circuits, I had a test jig for measuring things like this including the ratio of current into a trim pin to input offset generated.  Unfortunately I left those notes with that job but they would be easy enough to recreate if I had a reason to.

The ratio of 300 µV of offset producing 1 µV/K of drift is a good approximation, but not 100% accurate.

Oh, of course it is not exact, but it is design dependent.  Not all operational amplifiers have their lowest offset voltage drift at their lowest trimmed offset voltage.  See below.

The main reason for offsets in BJT based amplifiers is usually a mismatch in current or effective size of a transistor pair. This results in an offset voltage proportional to absolute temperature (This is one way to make a PAT source) and thus the approximate ratio.

With most BJT based OPs adjusting the offset to zero will also reduce the drift to a low value. However there can be small extra contributions to the drift. The relation between offset and drift is usually better at higher offset.

The min / max values in the datasheet are usually what is tested - so there ratio may not reflect the physics behind it.

I remember an old precision operational amplifier datasheet or application note discussing this and thought it was the OP-07 but inspection shows that that was not it.  Maybe it was the Fairchild 725 but I do not see anything there either.  At some point this became a design feature in precision operational amplifiers which distinguished them from "741" type designs and is consistent with them using different offset trimming circuits.

I only brought it up on the outside chance that branadic had used the offset adjustment terminals to trim the output creating a large offset voltage drift.

[Cerebus]

One additional contributor to drift in some bipolar OPAs is going to be internal input bias current compensation. The LT1001 has a bias current reduction circuit, so does the OP27 and so does its predecessor the OP07.

In most of these bias current compensation circuits there are two arms to the compensation circuit, one which measures the bias current on one side of the main differential pair (usually from a cascode bias) and one that supplies the 'make up' current to the inputs. They tend to look a awful lot like PTAT current sources.

For low drift applications it might be wise to avoid OPAs with input bias current compensation/reduction circuitry.

[David Hess]

One additional contributor to drift in some bipolar OPAs is going to be internal input bias current compensation. The LT1001 has a bias current reduction circuit, so does the OP27 and so does its predecessor the OP07.

In most of these bias current compensation circuits there are two arms to the compensation circuit, one which measures the bias current on one side of the main differential pair (usually from a cascode bias) and one that supplies the 'make up' current to the inputs. They tend to look a awful lot like PTAT current sources.

For low drift applications it might be wise to avoid OPAs with input bias current compensation/reduction circuitry.

This is a deliberate design decision and not a factor.  I doubt there is any precision operational amplifier produced which does not operate this way.  The internal current biasing of the operational amplifier drives the differential input stage with a PTAT current so that the adjusted input bias current does not vary as much over temperature; it is a deliberate effort at temperature compensation.  The input bias current compensation circuit just goes along for the ride.

For those who might be interested in the history of this, the improvement made to the LM301A over the LM301 was to use a PTAT current source to bias the input differential stage instead of a constant current.

[Cerebus]

One additional contributor to drift in some bipolar OPAs is going to be internal input bias current compensation. The LT1001 has a bias current reduction circuit, so does the OP27 and so does its predecessor the OP07.

In most of these bias current compensation circuits there are two arms to the compensation circuit, one which measures the bias current on one side of the main differential pair (usually from a cascode bias) and one that supplies the 'make up' current to the inputs. They tend to look a awful lot like PTAT current sources.

For low drift applications it might be wise to avoid OPAs with input bias current compensation/reduction circuitry.

This is a deliberate design decision and not a factor.  I doubt there is any precision operational amplifier produced which does not operate this way.  The internal current biasing of the operational amplifier drives the differential input stage with a PTAT current so that the adjusted input bias current does not vary as much over temperature; it is a deliberate effort at temperature compensation.  The input bias current compensation circuit just goes along for the ride.

For those who might be interested in the history of this, the improvement made to the LM301A over the LM301 was to use a PTAT current source to bias the input differential stage instead of a constant current.

I think you misunderstand precisely because you quote the LM301A, which doesn't have input bias current reduction circuitry - the bases of the input transistors have no current sources connected to them.

We're not talking about the main current source that sits in the tail of the differential pair, but something that actively reduces the input bias current (i.e. base bias), typically by mirroring the base current of a cascode transistor and injecting that into the input. This is a completely different thing to trying to even out the fall of input bias current with temperature rise by compensating out the rise in beta of the main differential pair with rising temperature by sourcing/sinking a tail current that is PTAT, as the LM301A scheme tries to do. I believe that the OP-07 was the very first OPA to use active input bias current reduction.

There are plenty of precision OPAs that don't do active input bias reduction, the LT1013 for example.

The giveaway that you're looking at an OPA with active input bias reduction is that the input bias current on the datasheet will be (a) low for a bipolar OPA, (b) specified as +/- some current and be very close in value to the input offset current figure and (c) not a datasheet item, but they will work, albeit not at their best, without an external DC bias path to the inputs because they're already providing most of the base bias current internally. Conversely, the way to spot one without input bias current reduction is that the input offset current and input bias current differ by a factor of 10 or more, whereas with an active input current they are of the same order.

Again, I'm not saying that an input bias current reduction scheme is definitely a source of offset drift, but is something to consider as a possible source of drift.

[montemcguire]

I doubt there is any precision operational amplifier produced which does not operate this way.  The internal current biasing of the operational amplifier drives the differential input stage with a PTAT current so that the adjusted input bias current does not vary as much over temperature; it is a deliberate effort at temperature compensation.  The input bias current compensation circuit just goes along for the ride.

This makes good sense, and points to the benefit of a non-bipolar input stage amplifier. There are modern chopper amplifiers such as the ADA4522, which I believe is a CMOS amplifier, so it lacks traditional input bias current. Since it's a chopper, the normally huge 1/f noise of any sort of FET input is eliminated. You do get chopper noise from the input chopper though, and one can think of that as "input bias", but it doesn't have to be all that large, and the temperature coefficient associated with it seems to be very low. So, it's worth considering, especially since its overall performance is pretty good, and its design eliminates a few of these PTAT factors that you mention.

[Alex Nikitin]

Again, I'm not saying that an input bias current reduction scheme is definitely a source of offset drift, but is something to consider as a possible source of drift.

As the input bias current itself. A 15nA input current of the LT1013 would create a 150uV drop on a 10K source impedance and it is temperature dependent (about 25pA/C or 0.25uV/C on 10K). An opamp with a bias current cancellation would have a considerably lower current to start with and a lower bias current v temperature drift. For example the LT1097 has <100pA input current with about 1pA/C drift. One thing to watch for on the opamps with a compensated input bias current is the current noise, as it is much higher than that on opamps with a non-compensated input current of the same magnitude. Plus the polarity of the input current can be either positive or negative and can easily change over temperature.

Cheers

Alex

[Cerebus]

One thing to watch for on the opamps with a compensated input bias current is the current noise, as it is much higher than that on opamps with a non-compensated input current of the same magnitude.

Added to that is that some datasheets get this wrong, and show input current noise specifications that are just the calculated shot noise in the quoted input current (which is less than the real total input current including the internal bias compensation). They can be one or more orders of magnitude out.

[branadic]

So what do you suggest as a replacement for LT1001?

BTW: Updated the graphs of the last measurement.

-branadic-

[David Hess]

So what do you suggest as a replacement for LT1001?

I would need to see the schematic again to make a recommendation but the LT1001 is one of the better options.  Unless there is a problem with high impedance, the input bias current and input noise current will not be a problem.

The LT1012 has 1/20th of the input bias current and twice the low frequency noise of the LT1001.  The LT1007 has 20 times the input bias current and 1/5th the low frequency noise of the LT1001.  The chopper stabilized LTC1150 will reduce all drift errors to essentially zero but has 6 times the low frequency noise of the LT1001 unless the measurement bandwidth is severely limited which may be the case anyway.

Before changing the amplifier, I would double check the impedances at the inputs to find the effect of the bias current and current noise.  I might also trim the offset voltage just to reduce the offset voltage drift.  Something else to consider is if the output drives a heavy load which includes the feedback network, then self heating will create other error terms.  This can be prevented by buffering the output with an emitter follower or other unity gain buffer.  The self heating from driving a heavy load results in lower open loop gain and longer settling time.

[branadic]

I was able to reduce the temperature dependency of the gain stage down to 1.41µV/K. How?

A long story short: During some industrial investigation we found, that printed resistors showed quite big humidity dependency. After covering the printed resistors with some self-adhesive aluminium foil this dependency vanished almost completely.

So this was the initial impuls for me to wrap the BMF resistors together with some copper foil for better temperature equality. Afterwards I wrapped around a few layers of self-adhesive aluminium foil as aluminium has a bigger thermal capacity. One side effect is, that humidity has less surface to get into the molding compound of the resistors, thus a much bigger time constant. Here are the current results:

Temperature coefficient: ~ 1.41 µV/K --> including 0.15 ppm of the reading plus 0.01 ppm of the range per degree c = 562nV/K of 3458A
Humidity coefficient: ~ -329 nV/%rH
Dew Point coefficient: ~ -371 nV/K
Pressure coefficent: ~ -1 nV/hPa

I think I'm now at a point with the need to measure inside a climate chamber.

-branadic-

[babysitter]

That means awesome14 was guided almost right by God and the ieee (imaginary electronic engel..angels), just misunderstood how to use the copper tape.

[branadic]

Damn, no. This has nothing to do with kisses by some mysterious omnipresent creature. The resistors have been alomst close together, but I wanted to avoid thermal glue to tie them together, as this can give unpredictable mechanical stress. So I decided for copper tap to tie them together and bring them as good as possible to same temperature.

There is no free hanging copper tape bridge of special length between the opamp and the resistors to compensate for temperature 

-branadic-

[2N3055]

There is no free hanging copper tape bridge of special length between the opamp and the resistors to compensate for temperature 

-branadic-

Not only magical length.. Don't forget special twists and wrinkles according to divine rule of chaotic wrinklity....

Sinisa

[babysitter]

Hehe, great.

At work I was recently trying to "clone" our best Thermometer 6 times, so I hooked 6 BC546 to a HP3488 switch and a 34401A.
When trying to get them at the same temperature, i first went to taping my sensors a block of cooper and later submerged it in a stirred Novec 7100 bath to adjust my setup to the reference temperature. My colleague who used it later re-verified with a similar setup using FC-40, setup was still stable and OK. So, to be really AWESOME, use a liquid filled chamber.

[branadic]

I have something similar on my to do list as you might remember, I do have some Vishay foil resistors on my desk, that want to be packaged in some golden copper package filled with ZT150/ZT180. Something I might manage after christmas or maybe next year. My wish to Santa Claus? More time for hobby

-branadic-

[branadic]

Unfortunately the measurement was interrupted during free xmas time by a reboot of the computer. 
However three days of the "copper tape modified" reference were aquired to get a raw idea about the improved 10V output of my LM399. It seems the reference is now stable enough for my purpose  Connecting it to AD5971. 

-branadic-

[branadic]

Hello,

again a ageing chart of 2 LM399 references.
now after 2 years (more than 730 days) 24/7 operation. 

And probably the last time. Since I may need the multiplexer
channels for two brand new LTZ1000A references soon.

Hi Andreas,

another year has past. Do you have new data or have you skipped monitoring both LM399's? After your last update in 2016 it seemed, that the slottet LM399 needed longer, but had better performance afterwards.

-branadic-

[Andreas]

Hello,

I monitor my new LTZs per hand (only once per week).
So no change for the LM399 setup.

and yes. the CH6 seems to have some seasonal drifts. (or maybe popcorn noise),

CH6 with short legs and no slots
Ch7 with short legs and slots
on your "slot or not" PCB.

with best regards

Andreas

[branadic]

Thanks Andreas, nice to see latest data. It seem there is some lag between LM399 CH6 and LM399 CH7 or maybe this is just a miscorrelation or just an accident? We should have used a significant largner number of references for this experiment.

-branadic-

[Insatman]

Earlier I posted about a new LM399 reference that seemed to be drifting a bit too much.  I got a lot of great feedback and set about using it.   Eventually I modified the circuit to replace the LM1001CN8 with a OPA177FP.  Some experiments while building another circuit showed this op-amp superior to the LM1001.  The feedback resistors were also replaced and a smaller value trimpot used.   The unit's metal enclosure was also lined with thin foam and the LM399 enclosed inside the box by small hollow Styrofoam cube.   The new schematic and test results are attached.   Overall the unit has improved a lot.    In the meantime I've also built a LTZ1000 based reference but that's for another post.  I'd like to say Thank You for the advice.

[David Hess]

Eventually I modified the circuit to replace the LM1001CN8 with a OPA177FP.  Some experiments while building another circuit showed this op-amp superior to the LM1001.

That is not surprising; the OPA177F is a premium grade and the LT1001C is not.  This is the kind of application where grading the operational amplifiers yourself will make a different in performance if only from weeding out the outliers.

The non-A (TI changed and reversed the suffixes, of course) OPA277 might be even better.

[zhtoor]

Hello Mickle,

your experiment orientations are that what I call orientation "0" and "5" where I also have only little differences (up to around 0.5ppm) in my setups.

The pin 1 marker directs to the right side on the photo.
So the critical directions would be the putting them on the left and on the right (connector) sides.
I guess that in this case there would be also about 3-4 ppm difference.

When looking at photos from older HP34401a or Keithley 2000 units the critical directions should be putting them on the left and on the right side. The pin 1 marker shows either to the left or the right side in this case.

With best regards

Andreas

hello andreas,

you might want to have a look at this.



@ 1:03:50

best regards.

-zia

[Andreas]

Hello,

what has the bad thermal design of the LM399 to do with that weird theory?

with best regards

Andreas

[The Soulman]

Hello,

what has the bad thermal design of the LM399 to do with that weird theory?

with best regards

Andreas

Well, 

ZENER DIODES QUANTUM DETECTORS DETECTING SPACE DIRECTLY

obviously. 
He haven't even touched on the indirect detecting of space, go figure. 

[chickenHeadKnob]

Hello,

what has the bad thermal design of the LM399 to do with that weird theory?

with best regards

Andreas

Well, 

ZENER DIODES QUANTUM DETECTORS DETECTING SPACE DIRECTLY

obviously. 
He haven't even touched on the indirect detecting of space, go figure. 

I think the point was the video claims the orientation of the zeners  within the detector and the orientation of the multiple detectors generating correlated noise are what measure the anisotropic "ether" in that presenters theory.