Author Topic: LM399 based 10 V reference  (Read 536674 times)

argintviu, DH7DN and 2 Guests are viewing this topic.

Offline CaptnYellowShirt

  • Frequent Contributor
  • **
  • Posts: 447
  • Country: us
  • Scooty Puff Jr.
Re: LM399 based 10 V reference
« Reply #200 on: May 25, 2014, 04:15:09 am »
Does anyone have experience with using current regulating "diodes" for this purpose?

http://www.mouser.com/Semiconductors/Discrete-Semiconductors/Diodes-Rectifiers/Current-Regulator-Diodes/
 

Offline Andreas

  • Super Contributor
  • ***
  • Posts: 3221
  • Country: de
Re: LM399 based 10 V reference
« Reply #201 on: May 25, 2014, 05:23:04 am »
Read the first post of this thread. It's a better way to create a constant current as stable as the LM399 it is powering.

That is still the best circuit for creating a stable reference.
If you look at the HP34401 service guide you will find exactly this ciruit. (Additionally with some startup pull up resistor).

If you do not want to use the 10V output, but create only a stable current source for the LM399 you could even use somewhat cheaper resistors (RC55Y with 15ppm/K) instead of S102 or Z201.

Does anyone have experience with using current regulating "diodes" for this purpose?
These are simple J-FETs where Gate + Source is connected together. -> you will never satisfy the requirements of a precision circuit.

If you have very much time you could try to use a LM334 current source and use the ciruit with TC compensation of the datasheet. With individual adjustment of the resistors you can get a final TC for the current source of below 50 ppm/K. But it will take several temperature cycling to get you there.

With the above cirquit of fmaimon you will get instant stable output depending on your resistors.
But where is more fun?

With best regards

Andreas
 

Offline Galaxyrise

  • Frequent Contributor
  • **
  • Posts: 531
  • Country: us
Re: LM399 based 10 V reference
« Reply #202 on: June 25, 2014, 03:11:32 am »
After not having much luck with LTZ1000, I thought maybe parallel LM399 was more my style.  Having something I could leave powered off most of the time appealed to me, too.

I ordered 10 LM399A from digikey, hoping I would get several that behaved like Andreas's star performers.  Turns out the Linear parts aren't quite like the National Semiconductor units, my lowest reference is 7.03V and highest is 7.1  (this seriously foiled my resistor plan, too.)

I then built a crude aging oven (an electrical box with nightlights under it, housed in a box made from a ceiling tile), wired my 10 references up in serial with a constant current driver, attached a keithley 2000 scan card, and fired it up.  There's also a RTD inside hooked up to another meter (a GDM 8261), and I wrote a simple program to talk to both meters and take readings.

 I can now confirm
There is a large possibility that the tempco is not linear but either a parabolic or s-shaped curve (due to compensation of the zener with a Vbe).

The x axis is temperature of the RTD, y axis is ppm change from a row I choose based on what the operating temperature would be in normal use.  The actual die temperature will be lagging the RTD temperature by a variable amount, accounting for some of the 'S'-ness of the curve (the experiment was only intended to compare TC of the different units and not compute an accurate TC curve.)  Imagine my surprise when one of them turned downwards at around 80C! Too bad for me I only have one like that.  Just like with the other reports in this thread, my lowest voltage unit was the best.  Absolute voltage isn't represented on this chart, but lower voltages corresponded to lower tempco.

My range of unheated tempco appears to match Andreas exactly.  Were those units also LM399A?

Has anyone tested if the LM399 also has some kind of zero tempco current
(near the heater temperature) in the valid range of 0.5 .. 10mA for the zener?
I have tested 0.6, 1.0, 1.1, and 1.5mA.  The resulting graphs were effectively identical.
« Last Edit: June 25, 2014, 04:12:32 am by Galaxyrise »
I am but an egg
 

Offline Edwin G. Pettis

  • Frequent Contributor
  • **
  • Posts: 451
  • Country: us
  • The plural of anecdote is not data.
Re: LM399 based 10 V reference
« Reply #203 on: June 25, 2014, 04:15:35 am »
Hi Galaxyrise,

I've got some data on the LMx99/A series from Bob Pease, as soon as I can dig it out of my archives I will try to post it, may be difficult in the form it is.
 

Offline Andreas

  • Super Contributor
  • ***
  • Posts: 3221
  • Country: de
Re: LM399 based 10 V reference
« Reply #204 on: June 25, 2014, 06:52:32 pm »
My range of unheated tempco appears to match Andreas exactly.  Were those units also LM399A?

I have tested 0.6, 1.0, 1.1, and 1.5mA.  The resulting graphs were effectively identical.

Hello,

my LM399H are all non "A" versions from National Semiconductor.
Good to know that the A-Versions currently stocked on DigiKey have the upper limit.
So I´ll probably better try the non A version.
The typical ageing spec of LT is somewhat lower than the NS spec was.

With LT1027 parts I have also made the experience that the C-grade devices (LT1027CCN8-5) in average had lower temperature gradient around room temperature than the 5 B-grade devices (LT1027BCN8-5) that I ordered later.
On non heated devices I can imagine that they tweak the (average) tempco at the temperature extremes giving more temperature gradient around room temperature (S-shaped curve). But with heated devices ....?

1N829A Zeners have their minimum T.C. around 7.5 mA +/- 3 mA

p.s.: do you have a photo from your measurement?

With best regards

Andreas
« Last Edit: June 25, 2014, 06:54:21 pm by Andreas »
 

Offline Galaxyrise

  • Frequent Contributor
  • **
  • Posts: 531
  • Country: us
Re: LM399 based 10 V reference
« Reply #205 on: June 25, 2014, 08:10:25 pm »
1N829A Zeners have their minimum T.C. around 7.5 mA +/- 3 mA

p.s.: do you have a photo from your measurement?

I'll let it cool off tonight and then try 7.5mA tomorrow to see if it's any different.  I think that would be awfully high from a long-term stability standpoint, though. 

Since I'm letting it cool down anyway, I can easily photo anything you'd like, but I'm not sure what you're asking for...  Could you explain what you'd like to see?

I've got some data on the LMx99/A series from Bob Pease, as soon as I can dig it out of my archives I will try to post it, may be difficult in the form it is.
Neat! I look forward to it.  I did get the impression he played around with this part a lot.
I am but an egg
 

Offline Andreas

  • Super Contributor
  • ***
  • Posts: 3221
  • Country: de
Re: LM399 based 10 V reference
« Reply #206 on: June 25, 2014, 09:08:21 pm »
Hello,

I just can´t imagine how your "ageing oven" from night lamps and tiles looks like.
sounds interesting.

With best regards

Andreas
 

Offline Galaxyrise

  • Frequent Contributor
  • **
  • Posts: 531
  • Country: us
Re: LM399 based 10 V reference
« Reply #207 on: June 26, 2014, 05:06:04 am »
I just can´t imagine how your "ageing oven" from night lamps and tiles looks like.
sounds interesting.
My pleasure to do a bit of show and tell :)

Things not clear from the pictures: The inside is roughly 6" square; the tile was originally 24" square.  The pieces of tile are tacked together with finishing nails.  The lid is another section of ceiling tile inside a cardboard box cut and taped to be an exact fit. I made the lid pretty tall on a whim, but that extra layer of cardboard looks to do a fair amount of good!

The bulbs are 15W each, and they're plugged in to a dimmer (the white box with the slider.)  As you can see from where the slider is, the bulbs aren't quite run at full power (the oven gets to around 140C at full power.) Heating 100C above ambient has a time constant around 30 min.

The metal platform is a steel plate from an old keyboard with three 2" bolts for legs.  I burned a paper towel and the smeared the ashes all over the bottom of that plate, blacking it.  (That was a huge help.) On top of the plate is a piece of perforated aluminum serving as my oven grill.  The holes in it were also handy for tying things down.  Then sitting on the plate but surrounding the grill is the 4x4" electrical box.  The wires going into the oven are a combination of white wire-wrapping wire and red enameled wire, both 30 gauge.
I am but an egg
 

Offline Vgkid

  • Super Contributor
  • ***
  • Posts: 2710
  • Country: us
Re: LM399 based 10 V reference
« Reply #208 on: June 26, 2014, 01:31:47 pm »
Certainly an interesting build, how stable can you get it?
If you own any North Hills Electronics gear, message me. L&N Fan
 

Offline Galaxyrise

  • Frequent Contributor
  • **
  • Posts: 531
  • Country: us
Re: LM399 based 10 V reference
« Reply #209 on: June 26, 2014, 02:00:07 pm »
Certainly an interesting build, how stable can you get it?
It's not regulated at all; it varies over a 5C range as ambient conditions change.  Good for aging parts, not so ideal for accurate TC characterizing.

I may update it some day with a peltier module+fan instead of the lamps and make controlled environment out of it.  (But peltier modules typically don't like going to 125C, hehe)
« Last Edit: June 26, 2014, 02:56:43 pm by Galaxyrise »
I am but an egg
 

Offline jlmoon

  • Supporter
  • ****
  • Posts: 609
  • Country: us
  • If you fail the first time, keep trying!
Re: LM399 based 10 V reference
« Reply #210 on: June 26, 2014, 02:10:05 pm »
That's an awesome looking ez-bake oven..  ;D
Recharged Volt-Nut
 

Offline Vgkid

  • Super Contributor
  • ***
  • Posts: 2710
  • Country: us
Re: LM399 based 10 V reference
« Reply #211 on: June 26, 2014, 05:40:27 pm »
Certainly an interesting build, how stable can you get it?
It's not regulated at all; it varies over a 5C range as ambient conditions change.  Good for aging parts, not so ideal for accurate TC characterizing.

I may update it some day with a peltier module+fan instead of the lamps and make controlled environment out of it.  (But peltier modules typically don't like going to 125C, hehe)
good enough, i might try it in the future.
If you own any North Hills Electronics gear, message me. L&N Fan
 

Offline jlmoon

  • Supporter
  • ****
  • Posts: 609
  • Country: us
  • If you fail the first time, keep trying!
Re: LM399 based 10 V reference
« Reply #212 on: June 26, 2014, 06:11:54 pm »
I was wondering how I was going to age some parts I have.. without spinning the electric meter off the wall.  Great idea Galaxy
Recharged Volt-Nut
 

Offline Andreas

  • Super Contributor
  • ***
  • Posts: 3221
  • Country: de
Re: LM399 based 10 V reference
« Reply #213 on: June 26, 2014, 08:36:04 pm »

My pleasure to do a bit of show and tell :)


This is how I like it: simple + effective.  :-+

With best regards

Andreas

 

Offline Galaxyrise

  • Frequent Contributor
  • **
  • Posts: 531
  • Country: us
Re: LM399 based 10 V reference
« Reply #214 on: June 29, 2014, 12:45:27 am »
Thank you all for the kind words!  And yes, it is like a low-power, old-style Easy-Bake Oven, hehe.  That collection of LM399s even vaguely resembles a cookie...

I had some technical difficulties when I first reconfigured the experiment, and then I opted to measure more currents--thus the delay.  I must say I'm very impressed with the robustness of these parts: given the errors that I've made, I'm surprised they all still function properly. 

Attached are voltage (as ppm difference from 90C) vs temperature at 0.6mA, 7.5mA, and 9.5mA.  Also is voltage vs temperature for #8 at different currents.  I corrected the graphs for temperature lag as best I could, which gets rid of that erroneous S bend at the beginning.  You can see the difference in the shape of the 0.6mA graph, which is the same data as the graph in my earlier post.

As Andreas expected, there is a visible difference at higher currents; a very non-linear shift.  It looks like most of my units can not be "zeroed" out at 90C within the 0.5 - 10mA range.  Following my curve-fit guess for how the parabola moves with current, #8 would "zero" at around 8.5mA.

Also not surprising, at 9.5mA there is a substantial amount of self-heating; raising the RTD by 12C or so when the oven lamps are off.

Edit: Removed charts that showed a decrease in the temperature of peak voltage with increase in current.  I had only accounted for half of the self heating.
« Last Edit: July 05, 2014, 09:19:54 am by Galaxyrise »
I am but an egg
 

Offline Andreas

  • Super Contributor
  • ***
  • Posts: 3221
  • Country: de
Re: LM399 based 10 V reference
« Reply #215 on: June 29, 2014, 09:11:50 am »
Hello,

amazing result.

Thanks for sharing

With best regards

Andreas
 

Offline Edwin G. Pettis

  • Frequent Contributor
  • **
  • Posts: 451
  • Country: us
  • The plural of anecdote is not data.
Re: LM399 based 10 V reference
« Reply #216 on: June 29, 2014, 05:16:34 pm »
Hello everyone,

Galaxyrise, nice set of curves, good repeatability within your limits.  I have found one note that Bob Pease sent me so far, you may have a hard time reading his hen scratch, this was part of a letter he sent me back in 2003, he did not mention just what meters were involved in the test nor the exact date of the test.  The 'graph's (I'll use that term loosely) indicated a repeatable drift in the measurements of the LM399 (he did not mention what the suffix was, an H or AH).  The test was over some months long but Bob didn't mention just how long, usually a long term drift measurement was about 10-12 months.

The maximum TCV of a LM399H is 2 PPM/°C, if you are seeing anything higher than this, there is a problem with the test setup.  The LMx99 is very linear, regardless of operating current, the internal temperature is regulated at ~90°C.  The operating ambient temperature range for the LM399/A is 0°C to +70°C.  The internal temperature of the LMx99/A is self regulating within the specified ambient temperature range.  Your measurements should indicate a linear TCV with change in ambient temperature.  All those curves seem to indicate problems with the test setup, even with repeatability of the curves, it indicates the error sources are relatively stable themselves.  Yes, the LMx99/A run at 90°C so you are going to see heat radiating off of them, the zener generates much less self-heating as is the design intention.  The heater runs at about 300mW at 30V, the zener would run at ~70mW at 10mA, at 1mA only ~7mW.  The heater compensates for the zener's heat as well, mostly leaving just the TCV of the zener itself.  The drift spec indicates a zener current of 1mA +/- 10%.  I have used the LMx99/As and I have never seen anything beyond a consistent TCV with them.  With varying ambient temperature, I have measured nothing more than the drift of the zener, usually under 1 PPM/°C for a LM399/A version.  Thermal EMFs can be really difficult to control and they tend to be nonlinear in nature.

I have attached Bob's note to me plus a National data sheet from 1999 which gives some interesting specs.  I also attached the LM199 data sheet from Linear Tech from 1990.  Linear made some improvement in the TCV over the older National part, but we're talking a few tenths of a PPM/°C here.

I remember someone mentioning the old 1N827/A voltage reference zener which dates back to the 1950s from Motorola, this special zener could exhibit a TCV of 5 PPM/°C but this wasn't the only Motorola zener that could do that; the 1N4569/A, 1N4574/A, 1N4579/A, 1N4584/A, 1N4779/A, 1N4784/A, 1N939/A/B, 1N4769/A, 1N4774/A, 1N2624/A/B, 1N2169/A, 1N2170/A, 1N2171A and the 1N945/A/B.  These zeners all had 5 PPM/°C TCV and voltages between 6.2V and 11.7 V nominally.  The zener currents varied between 0.5mA and 10mA for best performance and the temperature extremes could be as wide as -55°C and +150°C, pretty impressive I'd say.
 

Offline jpb

  • Super Contributor
  • ***
  • Posts: 1771
  • Country: gb
Re: LM399 based 10 V reference
« Reply #217 on: June 29, 2014, 05:32:01 pm »
The maximum TCV of a LM399H is 2 PPM/°C, if you are seeing anything higher than this, there is a problem with the test setup.  The LMx99 is very linear, regardless of operating current, the internal temperature is regulated at ~90°C.  The operating ambient temperature range for the LM399/A is 0°C to +70°C.  The internal temperature of the LMx99/A is self regulating within the specified ambient temperature range.  Your measurements should indicate a linear TCV with change in ambient temperature.  All those curves seem to indicate problems with the test setup, even with repeatability of the curves, it indicates the error sources are relatively stable themselves.  Yes, the LMx99/A run at 90°C so you are going to see heat radiating off of them, the zener generates much less self-heating as is the design intention.  The heater runs at about 300mW at 30V, the zener would run at ~70mW at 10mA, at 1mA only ~7mW.  The heater compensates for the zener's heat as well, mostly leaving just the TCV of the zener itself.  The drift spec indicates a zener current of 1mA +/- 10%.  I have used the LMx99/As and I have never seen anything beyond a consistent TCV with them.  With varying ambient temperature, I have measured nothing more than the drift of the zener, usually under 1 PPM/°C for a LM399/A version.  Thermal EMFs can be really difficult to control and they tend to be nonlinear in nature.

I was under the impression that for this experiment the heaters weren't wired in - but looking back through the posts I can't see where I got that impression from now (i.e. the zeners just connected as zeners).

Perhaps Galxyrise can clarify this.
 

Offline Edwin G. Pettis

  • Frequent Contributor
  • **
  • Posts: 451
  • Country: us
  • The plural of anecdote is not data.
Re: LM399 based 10 V reference
« Reply #218 on: June 29, 2014, 05:47:19 pm »
Hello jpb,

The point of using an LMx99/A is to utilize the internal heater to stabilize the reference zener's operating temperature, hence the exceptionally low TCV.  Attempting to measure the LMx99/A's zener TCV without the heater energized is an invalid test of the part and only erroneous readings will result if that is the case.

In the graphs, I noticed that the readings were referenced to 90°C, it would appear to infer that the LM399s were not being operated with the heater energized and that while the LM399's storage temperature does extend to +150°C....unenergized.....then the conditions and data inferred by the graphs are invalid.  If the LM399 had been operated with the heater energized and the ambient was as high as 100°C, then the chips were being operated improperly under those conditions, that does not appear to be the case though.

It would be appreciated if Galaxyrise could clarify his test conditions for us.
 

Offline jpb

  • Super Contributor
  • ***
  • Posts: 1771
  • Country: gb
Re: LM399 based 10 V reference
« Reply #219 on: June 29, 2014, 06:31:12 pm »
Hello jpb,

The point of using an LMx99/A is to utilize the internal heater to stabilize the reference zener's operating temperature, hence the exceptionally low TCV.  Attempting to measure the LMx99/A's zener TCV without the heater energized is an invalid test of the part and only erroneous readings will result if that is the case.

Again, as I'm not Galaxyrise I can't claim any deep knowledge, but my understanding was that by finding which devices had flattest curves around 90C good devices could be selected without having to monitor the heated devices for a very long period - but this is just a guess on my part. I think the point was to find devices (and current levels) which would be optimal when the heaters were used in the final circuit.
 

Offline Galaxyrise

  • Frequent Contributor
  • **
  • Posts: 531
  • Country: us
Re: LM399 based 10 V reference
« Reply #220 on: June 29, 2014, 06:42:33 pm »
Attempting to measure the LMx99/A's zener TCV without the heater energized is an invalid test of the part and only erroneous readings will result if that is the case.

The heater is not energized for the above; jpb is correct.  And as such, those charts are not the TCV you'd get from a "fully operational" LM399.  Sorry if I confused anyone with that!  But I wouldn't call the readings "erroneous", either. I believe they represent the TCV of just the reference half.

I would expect that the TCV of a fully powered LM399 would essentially be a measure of the change in internal temperature as external temperature changes. It's hard to guess if the A units are 1ppm/C because the heater regulates better, the reference TC is better, or both, but 140ppm corresponds to a 10C shift on my "worst" unit. It would be very hard to notice any non-linearity in my graphs over that 10C range, I think. You'd need a reference with a zero TC current and a very good experiment to see that gentle curve--the curve would only be 7ppm high over the 70C operating range!

So I think my results are consistent with your experiences and expectations from the datasheet.

The story is that I was rigging these LM399 to age at 125C (as suggested by Dobkin himself), and I wanted to be able to pick the most stable references of my 10 over time when aging was done.  I logged the warmup period to help correct for temperature variations when evaluating time stability.  Then I saw that downward turn on one reference, and the rest is in the posts here :)

It wouldn't surprise me at all if there remain artifacts from my test setup as I'm no expert in measuring this kind of thing and the setup wasn't even intended to measure it in the first place. But everything has been behaving consistently and repeatably, so at this point I'm inclined to think the shape and movement of the curves are reasonably representative of truth. 
I am but an egg
 

Offline Galaxyrise

  • Frequent Contributor
  • **
  • Posts: 531
  • Country: us
Re: LM399 based 10 V reference
« Reply #221 on: June 29, 2014, 06:49:49 pm »
I have attached Bob's note to me

Thanks! An enjoyable read. I also struggle with how to judge long-term stability of LM399 when my measuring gear is also LM399-based, hehe. 

Quote
plus a National data sheet from 1999 which gives some interesting specs.  I also attached the LM199 data sheet from Linear Tech from 1990.  Linear made some improvement in the TCV over the older National part, but we're talking a few tenths of a PPM/°C here.
   The NS sheet does have interesting extras, like that 1000hr aging chart.  And a TO-92 part, teehee! 
I am but an egg
 

Offline Edwin G. Pettis

  • Frequent Contributor
  • **
  • Posts: 451
  • Country: us
  • The plural of anecdote is not data.
Re: LM399 based 10 V reference
« Reply #222 on: June 29, 2014, 07:54:22 pm »
Galaxyrise,

You are welcome, I think I have some more data from Bob, it may be buried in old e-mails from him or I just can't find the hard copy of it.  Bob usually liked hard copies instead of e-mailing.  His handwriting though was a stinker to decipher at times, particularly schematics which was helter-skelter drawn.  By the way, Jim Williams of Linear Tech was also schematic drawing challenged.

Sorry, if I had come across this thread earlier, I might have been able to save you some work.

Your assumption about the 'zener' diode in the LM399 is understandable but it is still an invalid assumption.  The buried zener circuit was designed to operate at the chosen temperature of 90°C and was optimized for that temperature only, deviating from that temperature even a few degrees will not 'show' the TCV of the zener accurately.  Its characterization depends on its operation at the specified 90°C only internally.  Outside of that temperature range, the buried zener will have a composite TCV curve made up of the entire voltage reference circuit not just the buried zener.  It cannot be characterized in the same manner that you would apply to a regular zener diode unfortunately.

If you have several LM399s available, you could parallel them using the appropriate summing circuit to get an improved composite characteristic, it doesn't just work for reducing noise.
« Last Edit: June 30, 2014, 04:03:41 pm by Edwin G. Pettis »
 

Offline Galaxyrise

  • Frequent Contributor
  • **
  • Posts: 531
  • Country: us
Re: LM399 based 10 V reference
« Reply #223 on: June 29, 2014, 10:10:10 pm »
Your assumption about the 'zener' diode in the LM399 is understandable but it is still an invalid assumption.  ... Outside of that temperature range, the buried zener will have a composite TCV curve made up of the entire voltage reference circuit not just the buried zener.
Characterizing the entire reference circuit is exactly what I thought I was doing (as it's what matters in the end), but I hadn't really thought about that in detail. I was just reading through National Semi's App Note 161, and I can really see what you mean.  The current changes, especially, basically don't affect the zener at all!  (Though it does affect the circuit as a whole.) I suspect I will run a similar experiment when aging is done, but with the heaters on (and a narrower temperature range, obviously.)  It will be interesting to see if it correlates to the unstabilized data at all, or if the charts I produced are just a curiosity. 

Quote
If you have several LM399s available, you could parallel them using the appropriate summing circuit to get an improved composite characteristic, it doesn't just work for reducing noise.
That's the plan!  I've seen where Pease talked about running them in groups of 4, and I was figuring 4-6 based on how they looked in the latter half of aging.
I am but an egg
 

Offline cellularmitosis

  • Supporter
  • ****
  • Posts: 1111
  • Country: us
Re: LM399 based 10 V reference
« Reply #224 on: June 30, 2014, 07:10:29 am »
I have found one note that Bob Pease sent me so far, you may have a hard time reading his hen scratch, this was part of a letter he sent me back in 2003, he did not mention just what meters were involved in the test nor the exact date of the test.  The 'graph's (I'll use that term loosely) indicated a repeatable drift in the measurements of the LM399 (he did not mention what the suffix was, an H or AH).  The test was over some months long but Bob didn't mention just how long, usually a long term drift measurement was about 10-12 months.

Wonderful!

I have tried my best to read Bob's hen scratch and have typed it up in a google doc:

https://docs.google.com/document/d/1fpuR_R9YvK-08QuEIy1pIZLdUrmNANFGbx4z22E5e68/edit?usp=sharing

That document is publicly editable, so let's have our community collectively come up the best interpretation of his hen scratch :)
LTZs: KX FX MX CX PX Frank A9 QX
 


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf