Ultra Precision Reference LTZ1000

[MisterDiodes]

I would expect to see a voltage reference standard or 8 1/2 high end multimeter based on LTZ1000A installed in a climatized lab room between 20°C ... 24°C [68°...74.2°F] and not being stacked on a pile of hot T&M equipment - at least that is how I would do ...
The 12k5:1k00 is a compromise between max. expected temp. in lab room - minimizing long term stability - I my case I opt to minimize the long term drift.
It is hard for me to imagine what the lab room conditions are if you would recommend the use of a 15k0:1k00 ratio - at least on planet Earth.

That's a lot easier to imagine than you think, and this is what the Pickering paper failed to take into account:  On a production line situation, you can't always control the environment.  Its not really the top-end heat setting over ambient, its how much its changing throughout the course of a day, or how much wind is blowing by / or not blowing by.  For instance, the modules we make had to fit behind and under a stack of test equipment, and it was either hotter/ colder than we wanted in the cabinet, OR the best place in the clean room that was convenient for the rest of the workflow meant the cabinet had to be next to the air-lock door, with people coming and going countless times per day.  Every time that door opens there is extra breeze blowing across that part of the room that we have no control over, and the cabinet cooling system couldn't keep a perfectly stable interior temp with the automatic door nearby.  Even though the clean room temp is relatively constant (within a degree or two), the air flow over the control cabinet is not.

The other problem is if the equipment gets hotter or cooler depending on what type of product is running through the machine (the motorized motion slides are running longer or shorter distances), and about a zillion other "gotcha's" that pop up.

If you place an LTZ1000a in this location, you'll find that the system Vref becomes a tracking thermometer in a hurry if the die temp is too low.  We did indeed have to build several versions with a 15k/1k ratio in order to keep the die temp from responding to local rapidly changing local atmosphere conditions that we have no control over.  Generally we found that in real life, about 13k ~ 14k/1k gave the best performance, but we would go to 15k or higher as required.  And yes, the Vref is covered, but you'd be surprised how much the entire enclosure heats or cools - that's a lot of surface area, and airflow has a pretty large heat conduction effect on the enclosure as a whole system.

You actually have to do this in real life to see what I'm talking about - if you use a 12.5k/1k ratio, your Vref will be more wobbly than you want in a real world production situation.  Granted, if you're in a lab you can get away with the lower ratios, but sometimes you have to use these Vref's in non-lab conditions.

I can guarantee you - if you have a rack- stack of 3 or 5or 10 HP3458a' (which I have seen and not uncommon in a semicon QC area) they will NOT be even close to the same temp. inside, and if you run the 12.5k/1k ratio, your boxes will NOT have great stability, even over an hour.  Now if if its a solo 3458a by itself in a cal lab, then you will probably have it work pretty well at the lower ratios. But from experience, when you have 10 high-end DMM's in a cluster rack, you have to have the higher die temp on the Vref - otherwise you have a very large, expensive, sensitive  thermometer.

And then of course there was the situation where you get the system installed and running, and then the room gets re-arranged, and now the cabinet is acting much different than before; you have a new hot piece of equipment covering what you thought was an air vent, etc.  Sometimes, no matter how hard we try, the customer always finds a way to circumvent the resistor selection.  That's why we design these to run at a little higher die temp than book-theory dictates.

Also, a lot of times these are used where you really don't care what the drift is over the course of a year.  Sometimes you want good drift performance over 30 ~ 90 days, and what you're after is an extremely low noise Vref that you know is good over the next -several weeks-.   On a semicon production line that is doing critical testing with wafer probers, the Vrefs can be swapped out as often as needed and calibrated against the master transfer standards back in the cal room.  When you're in a main assembly room where say 10,000 parts are produced per hour, 24/7, and each device is worth $250 ~ $500, then nobody is going to chance that the Vref they are using hasn't been calibrated within the last month or so.  The devices are just too valuable - so the Vrefs tend to get swapped out fairly often - and you aren't really after 365-day drift stability numbers.  Its something good to shoot for, but not really always cost effective in all applications.

So, yes, sometimes you really have to go for the higher resistance ratios on the heater circuit, depending on the application.

[eurofox]

If someone is interested on ebay France, the reference module for HP3458A

I'm not involve in it

http://www.benl.ebay.be/itm/161371505611?ssPageName=STRK:MESINDXX:IT&_trksid=p3984.m1436.l2649

[MisterDiodes]

Well, 15k/1k how 3458a's come supplied, last I checked.  Do they supply a different ratio on newest units?

"Modifying" the DMM  would be to drop the heater resistor ratio down to 12.5k/1k, and that works also, as long as the environment is compatible.  We've built Vrefs that way sometimes.  For a solo unit by itself in a temperature-controlled room , that can work.  When you run a cluster stack, or are running in a really fluctuating environment its been my long experience that you'll want the higher ratio.  For a large cluster of DMM's (don't care what brand) You will find that the boxes in the top / middle / bottom of the stack will have much different temps inside than what you thought.  In a clean room where floor real estate is at a premium, the DMM's had better run to spec when sitting in a tall stack.

Or in the case of what I design, the Vref sitting in machinery sharing space with a motion control system / processor board / probe board inside of a control cabinet.  You need to have the die temp -well above- whatever the crazy changes in ambient temp are happening around the Vref, even though it might be covered from direct airflow.  Especially the rate of change of temperature.  Calculating for a steady-state ambient temperature is one thing, but when you move the design into reality, then you'll see that sometimes you need a higher ratio for best overall performance - and that includes those Un-controllable environmental issues.   Also known as "Real World":  It'll get you every time <Grin!>   

Nobody on a production line that I know of will make any modification to their 8.5 digit DMM though.  They want them to run in spec as supplied from the manufacturer, and nobody I know if will be changing out heater resistors if its a company-owned DMM.  I have never seen that at least.

You just select the resistors for intended application, that's all.  This isn't a one-size-fits-all design.  You test it out and make adjustments as you go until you get the performance you need.  If the 12.5k/1k heater resistor combo works for your situation, then that is "perfect" for -your- application.  It is not perfect for -every- application though.  A for-profit business may not be at all interested in 365-day drift numbers, they just want their products shipped out and running, and they will calibrate their test systems as often as needed.

[MisterDiodes]

Gaz,
Most of the high-end DMMs on the production line will be 3458's, as far as I can see.  Sometimes a Fluke or Kiethly, but usually its mostly 3458's at least in my circles.  In fact, in more than one plant they will have a guy that works for Agilent/Keysight with his own office down the hallway.  These DMM's are the units in battle zone.  Down in the cal room where its quiet, really temperature controlled and you don't have 200 people stomping by every hour, that's where the expensive Fluke calibrator / transfer standards / precision resistors / etc will be found.

Re: Dr. Frank's observations - I agree mostly with those:  Yes, if you were to ask me to build you an LTZ1000a circuit, you would get good wirewound resistors on a copper-filled board; if I didn't know where the unit was going I'd give you 13k/1k heater resistors to start out, and you would never see any sort of expensive Vishay magical foil (and even more magical datasheet) resistor anywhere on the board.  If you're paying more than $5 or $7 per resistor, you're doing something wrong - I least from what my experience tells me.  The Op-amp would be sealed, and you could run the thing under water if you wanted to. No slots, goofy copper layouts or other gimmicks.

The designs I use have lots of hours on them, they run a long, long time - over decades time spans.  The older they get the better they run.  I figure it takes at least 2 or 5 years for the crystal lattice strain to level out on the die, and then you should be getting much better drift numbers after 5 or 6 years.

If I knew you were going to baby it in a very climate-controlled room, I'd give you a 12.5k/1k heater resistor setup.  If it has to work in the real world the ratio would be more like 13.5~14k/1k and if it had to work really hard  for a living in any hard hot/cold situation it would be 14.5k~15k/1k heater resistors.

I would probably not take the time to swap out an LTZ1000a for an LTZ1000 because for me, there is not a lot of long-term benefit in that.  The LTZ1000a's were invented to be less sensitive to surrounding environment changes, and nobody I know in a profitable business worries about 3ppm per year drift vs. 1ppm year drift.   I'm not saying its a bad thing to do, I'm saying that 5, 10 or 20 years down the road - something I have experience with - the '1000a and '1000 will probably drifting at just about the same rate, all else being equal.  If you did do this swap, down the road you would have wasted a good LTZ1000a part, spent money on a LTZ1000 and used up 4 to 8 weeks burn in time.  I dunno.  Maybe if the Vref board you had had a falied LTZ1000a, maybe it would be worth it - but that might be rare.

Outside of the Volt-Nuts crowd - very nice people by the way - I don't know of a lot of people really -needing- a 1ppm /yr  drift circuit.  I know the hobbyists enjoy chasing after low drift numbers for fun and entertainment - and I get it.   If you're a cal lab or on a production line, you'll be doing calibrations MUCH more often than once a year, and you'll probably be using a Fluke calibrator anyway, not some hacked-up board with a Dixie cup on top. <I say that tongue in cheek because I have one of those on my desk now>.

Its all fun, I understand - and there is certainly more than one way to build the circuit.  Use what works best for you! 

[Dr. Frank]

Dear Mister Diodes,

the discussion about the LTZ1000A circuitry began years ago, when the mediocre annual stability of the 3458A compared to other 8.5 (!!) DMMs, partly with the very same reference element, got obvious.

To remind you, besides some nice, harmless Volt-Nuts, the 3458A is also often used in metrology labs, and there, a 8ppm/a stability is really undesirable in many cases.
Yep, if this instrument features 8.5 digits AND 0.02ppm linearity (of input), then 8ppm/a really does not fit to this class of instruments.
No, 8ppm/a make 8.5 digits nearly worthless..


HP could have assembled equally an LM199 inside the 3458A, if they let the LTZ run on 90°C.

Maybe the discussion in this thread is quite academic, but for the design (from amateurs) of a separate LTZ1000 reference, the 3458A instrument and its reference implementation was very instructional. And that was the core of this thread, the idea to pimp the 3458A was more or less a by-product.

Also, the 1989 commercial flyer promoted the 3458A to be a metrological class instrument...
(reprint here: http://cp.literature.agilent.com/litweb/pdf/5965-4971E.pdf)
It would have been easy at that time to achieve a lot more of (metrological) stability by some simple means.
Instead, HP offered specially selected 4ppm/a references for a fortune.


By the way, for none of these 8.5 digits instruments it is specified, that/if their annual stability applies on intermittent or non-intermittent operation.

For the 3458A it's quite clear, from Service Note 18A, that the 90°C reference may show strong hysteresis, if it's NOT operated continuously.
That may be up to 15ppm!

And HP once again did not specify clearly, if that has been a temporary problem of certain S/N, like the SN18A implies by its title description, or if ALL LTZ references suffer from that problem.

Latter one I assume to be the truth, and that additionally would be a reason to reduce the LTZ temperature.. as such strong hysteresis obviously is only present on big excursions over room temperature.
Otherwise, if a spare instrument in a fab site is retrieved from stock, that would cause a big problem... also the high energy prices here in Germany may foil continuous operation.


Frank

[MisterDiodes]

DR Frank:

I agree with you more than you think, and if my previous message didn't covey that then I apologize.  Let me try again:

Your observations are correct, but I am talking about production line operation here, not hobby use - and a production line is always running 24/7, so that the power-up hysteresis effect is not as much as an issue.  Generally when a new DMM comes online, it will be running in warm-up mode for several hours - at least one shift - before it becomes operational.  Then it is running in parallel with another unit to double check that the go- no-go tests are working right, and then a few shifts later the oldest unit on the stack is powered down and sent to the cal room for testing.  Also what we're looking for at least some of the time is relative differences between two voltage measurements.  The absolute accuracy needs to be good, but relative measurements are more important sometimes.

I understand that in some countries, you can't afford the power cost (especially in a home lab), so of course you would be more aware of power-on hysteresis.  But even in Germany the production clean rooms are "on" 24/7, correct?  It is very expensive to power down a clean room / semicon production facility, so in most cases its better to leave the critical equipment on.  I know here in the States I don't know of any production facility that shuts down completely even for holidays. - the costs would be enormous to get everything turned on and warmed up again.

OF COURSE for a Volt-Nut hobbiest that making the resistor modifications would be great idea if you need those 8,5 digits, and your power cost is so high you can't afford to leave the critical equipment on all the time.  I agree!

How much does it cost to keep an '3458 powered up 24hrs per day where you are??  I know it really adds up over a year I think.

For my equipment, what we're after is more the low-noise and short-term stability characteristic (90 days) of the Vref, in which the LTZ1000(a) is better at than the 199/399. 

That being said, and LM199, or even 20~40 or so LM399's in parallel for lower noise makes a good Vref too, I will agree! 

At least here in the states, I've not seen a 3458 used in metcal lab as a master voltage - test reference for a long time; that job has gone to the Fluke calibrators anyway.  It might be used as a secondary test ref for sure.  But I know our local metcal just sold off its last 3458, I think they are just all 100% Fluke calibrators now.  But different labs have different setups I know.

But on the production lines, you see lots of 3458's more than any other 8.5 digit DMM running on the test probers.  At least with what I work with.  Almost all the probers I know of are designed to work with the 3458's as the DMM device, but you do see other vendors sometimes.  Not often but there are other DMM's at work.  Generally you will need several DMM's at a probing station if you want any throughput, and in general you want all the same model DMM at the prober.  Sometimes you don't need that though, it just depends on the product going through the prober.

HP ripping off people with the "better" Vref and charging people $$$$ for it: I 100% agree with you that's a dirty practice.

I'm still not clear on the Volt-Nuts crowd chasing after a 3ppm / yr drift with LTZ1000A vs maybe 1ppm/yr with LTZ1000, and why that is a net benefit, but I am all for people giving it a try if that's what they want.  Go for it!   I am looking at more the net benefit 10 or 20 years later, and from what I've seen - there just isn't a lot a difference between the two devices over long time spans.  If you have a different experience on devices running 20 years, then that's good to know.  I will stand by my advice: If you have a good, quiet stable LTZ1000a, and you've dialed the heater resistor down to 12.5k~13k over 1k, there may not be a huge benefit to swapping out the LTZ1000a with a LTZ1000.  You might be trading out a "good" LTZ1000a for a "bad" LTZ1000.

In any case years down the road they will be drifting at about the same rate, and I have seen that over and over again.  That's all.  If you see a good benefit over the first 4 or 5 years, then that's a good thing to know.

I keep telling my buddies that want to play with these Vrefs at home, and this will apply to 199/399/LTZ1000(a):  Get an older one that's been in use at least 4~5 years.  That will get you a head start on good stability.

I guess that's all part of the fun!

[TiN]

Generally you will need several DMM's at a probing station if you want any throughput, and in general you want all the same model DMM at the prober.

That's why K2002 have so called 3458A-compatibility mode for GPIB interfacing, I guess.

I'm still not clear on the Volt-Nuts crowd chasing after a 3ppm / yr drift with LTZ1000A vs maybe 1ppm/yr with LTZ1000, and why that is a net benefit, but I am all for people giving it a try if that's what they want.

I'm asking that question myself every day
Probably real answer is that one have some extra cash to spend on toys for fun, same as buying 8.5d DMM for home lab, and leaving it off most of time.

[ltz2000]

It seems that this has become a "fight" between the metrologists and the industry guy MisterDiodes. I have no experience of his playground, the semiconductor industry, except once visited a factory and spotted a number of 3458As, but I think he has a good point.

I believe too that the 3458A was not designed for the metrology labs. Of course they market and sell it to every possible application they can, but the emphasis must have been where the money comes from. They could have easily made a sister version, but I guess the market was not big enough.

We think the the 3458A is an expensive high end meter used only for the most demanding applications. But for example in military and aviation $10k is nothing. Hourly service rate can be higher as well as the simplest spare part. The 3458A is specified because it is known to be good, it is used everywhere else, the GPIB commands can be copy-pasted and believe or not because it is a full rack width unit (no need to play with flimsy half rack adaptation kits). It was quite common to see a 3458A in a "wrong company" with a function generator and a programmable power supply measuring something as simple as the supply voltages and their ripple.

For the 3458A, it was also designed for use in up to 55C environments [don't ask me why-- I have no clue]

Business? The ratio of instruments sold must be something like metrology labs 1 - automated test systems 100.

[Andreas]

Hello Andreas - long time ago you posted in this thread this schematic and I have a few questions:
T2 is an n-ch JFET boost stage that also acts as a pull up to force the start up of U2A in positive direction, well I am wondering why:
- you have added an extra pull up R13 (47k) ?
- you have added a min. load resistor R17 (22k) but you have already other resistive loads available (R4/R5) ?
- If you add 10k series gate-resistor to T2 (to limit output current of U2A in boundary situations), you could remove (short ?) D2, as T2 fulfills same task - even better to replace T2 with a BSS159 (depl. MOSFET) - did you consider this (removing D2) ?
- Why is R9 present ? Even LTC confirms in its ref. design that it should not be used with LTZ1000A-version.
Thanks and have a nice day !
Best regards,
gazelle

Hello gazelle,

Ok R13 is not necessary if the FET is already populated.
But I also wanted to have a fallback solution in case the FET would not work.
So R13 is populated.

R4 / R5 are not visible to the OP-Amp. The FET (or the diode) act only in one direction
this will make oscillations of the LT1013 rather likely.

I adopted the cirquit from a LM399 (together with a BF245A/B).
There I had (small) oscillations without the pull down.

D2 was also intended as fallback solution to the original cirquit.
Then I decided to let it in to avoid gate currents through the FET diode.

I did not consider a depletion MOS-Fet. (The BF245C were already in my drawer).
Question: how do you plan to limit the inrush current on power on
with a low ohmic device like the BSS159?

By the way the BF245C is no longer produced in TO-92.
So you would have to use the SMD part (= BF545C).

With R9 you have found a weak point in documentation.
It is not populated with my LTZ1000A, but I did not mention this on the plan.
(The cirquit could also be used with LTZ1000 as fall back solution).

With best regards

Andreas

[Andreas]

Isn't there a conductive path from output of U2A through D2 feeding R4 / R5 ?
I assume D2 is stuffed.

Hello,

D2 is populated.
D2 is normally not conducting since the gate of the FET is more negative than the source.
(and more negative than the zener voltage).

I have around 6V at the OP-Amp (U2 Pin 1) with 7.2V at the Zener.
So I gain about 2V of supply voltage headroom against a 1N4148 diode.

with best regards

Andreas

[MisterDiodes]

OK, I want to make one thing VERY Clear:  I -really- do appreciate all the work and generous sharing of information provided by DR Frank, Andreas, Digelent, and everyone else and this is a GREAT discussion, and a GREAT collection of very smart guys.  It is an honor to post some data once in a while.   My hats off to all of you because I know how much time and patience all of this takes.  Why do I tell you to wait a few years and re-test Vref when its settled down?  Because that's what we are dealing with here.   What I am trying to provide to the discussion is some feedback from the real world, and trust me - I understand the Volt-Nut vibe - its fun and addicting, for sure!  But I have to mix the desire for ulta-precision with pleasing customers (enough to pay their invoice) and to make a profit in a reasonable time frame: Truth be told, my main goal is to not get a phone call at 2am because some machine went down halfway around the world, so I have learned to really make the stuff last.  Other designers scoff because I tend to over-design, but then again most of the stuff we made in 1985 is still running - except for a brain-dead EPROM here and there.

I'm also old enough and lucky enough to have the stars align to have met the likes of Jim Williams and Bob Pease (working on other projects, not Vrefs) and they really were absolutely brilliant, and great to work with.

I didn't know I was in a "fight" with anyone, that is the very last thing I want to do -  I think what I failed to make clear is the difference between a carefully controlled lab at home or a very carefully temperature controlled metcal lab (I think this is Dr Franks expertise) vs. a production environment where you can't trust how abused these things will get, and how long does something last - and that is something I deal with on a daily basis.

As David Jones points out, if you are designing a Vref for a product, the "Ambient temperature" can be "a real trap for young players".  And what you will find missing on almost all high end DMM / Vref datatsheets is the RATE of CHANGE of ambient temperature that is allowed, and this will bite you in the butt every time when you deliver a product if you're not ready for it, and no SPICE simulator will handle this at all. This is one of the places where THEORY ends and REALITY begins.  For instance, the '3458 is rated for 55°C ambient, but you'll see in a moment why that is only part of the story.

Now:  Here are some real world numbers I jotted down from a facility I was at yesterday.  I tried to take pictures for you guys but it is not allowed.

The setup:
Test Prober station #6 has 4ea HP/Agilent 3458a's.  All are on 90-day recal schedule, and they are staggered so that there is only one new unit running at a time.  They range in age from 4yrs old to 14 yrs old.  None have had their Vref replaced or modified, but two of the older units have new VFD front panel displays replaced. There is one box with a pretty dim display that will need a repair soon, but as a computer-controlled DMM it is running fine.

 These are in a 19" rack, stacked near the middle.  Overhead are some test prober relay switcher boards and a couple blade PC's (one is running the prober & DMM's, the other one is dedicated to data logging and sending data to the inventory-control system, because at this station every working device is serial numbered and sorted into quality bin lots).  Under the DMM's are some power supplies, more relay boards and more power supplies.  Because the prober is known to create a small amount of dust, it is surrounded by a plexiglas glazing safety shield doors and the enclosure is under slight negative pressure to keep the dust inside. The doors are only opened up when a stack of wafers are installed or removed.   The whole thing is in a cabinet against a wall, on the other side of the wall is an air compressor and chiller and a bank of vacuum pumps.  There is also a high-pressure Osmosis - De-ionized water processing station. In other words the wall is fairly warm to touch.  To the right of the cabinet is another machine's motion control cabinet and power supply.  It is very warm also.

The room itself is running about 27~29°C on a very warm day outside.  You have to wear a Tyvek bunny suit, hood, mask, booties and gloves, and you wish they would have the temperature lower, but that's what we have on the menu today.

Inside the '3458a boxes, with the machine idle, bottom to top, the interior temps are 38, 41, 43 and 40°C.

Now here's the fun part:  When the prober starts, and then when the machine on the right starts, and then when you notice more vacuum pumps are online on the other side of the wall... The interior DMM temps are now reading 43, 46, 50 and 44°C when I check again, about 15 minutes after everything starts up.  Typically a 5 degree jump in about 15 minutes, or a rate of change of 20°C / hour.

NOW:  What makes this work in this situation is the data being collected is 6 or 7 digits, and the software compensation / ACAL routines running inside the '3458 are able to compensate well enough so that the testing operation is still working right.  AND the Vref Die temp is running at the higher temperature for stability during temp changes. The manager mentioned that they do get special support from Agilent in the form of customized firmware, and they do calibrations at more like 34 or 35°C.

So this is a different situation that you would encounter in a Volt-Nit or metrology lab situation.  BUT as a designer, I have to plan ahead for major abuse.

Bottom Line:

YES, for a cooler, temperature controlled lab, running the LTZ1000(a) at a lower temp makes perfect sense.  If your VRef is going into a very abusive environment where the temperature swings are sudden and relatively rapid - well then you really need to raise the Vref die temp to keep your Vref relatively stable, no matter what's going on around it.  That's one of the reasons '3458 have the die temp set as they are, and I've never really seen one that has "degraded performance" if the DMM has to work hard for a living, at least not over the 90-day calibration period (recommend by Agilent when these units are used like this - nobody ever expects these to run over a year without recal in this setup).  Even then you will need some compensation software to help smooth out the dips and glitches, and the '3458's do a pretty good job at that.

They have tried other DMM's on this rack, but the '3458's run the best while being abused.  At least in this setup.  I know there are other brands and I will not argue with you that they work well also in other situations.

Oh and by the way:  This is something else Vishay leaves off the data sheets of their very expensive magical resistors:  See how the tempco acts during high-rate-changes of temperature, and compare that to a wirewound - and then remind me why the Magical Vishays cost 10 times as much.  Lets just say I have never seen a 10 times performance increase, but that's just me.  I have never seen any sort of cost benefit ratio at all, at least for the stuff I design (like 20 / 24 bits DACs/ ADCS and high-precision current sources).  The main point I keep streesing:  After 10 or 20 years, these Vrefs tend to settle down and become very good performers even without any voodoo, magical datasheet gimmicks at all.

Your milage will vary, and to all Volt Nuts: I love reading about what you do and how you get there.   Honest!    The generous sharing of information here is fantastic!

[Dr. Frank]

Mister Diodes,

me also, I do not see any "fight" here, but a objective - excited discussion.
And I really appreciate your very interesting descriptions from those rough industrial conditions.
By the way, the high MTBF of the 3458A is another argument for these applications, I think.

Anyhow, what we have found out here in this thread, is that it would have been quite easy and cheap for HP to offer a metrological option also, let's say OPT 03 with 1ppm/a  @ Ta < 35°C, as it was done by the FLUKE 3458A/HFL modification, with 3ppm/a DCV stability, a more stable reference resistor,  and specified Ohm Transfer stability.

Reading again the HP Journal 4/1989 about the LTZ reference, the HP engineers obviously did not think about that at that time, 8ppm/a was top notch for them.


I was one of the early users of this instrument, at university in 1990 or so.
The laboratory was not air conditioned (+10..+15°C), and the instrument was placed in a rack with other instruments (+10°C), so it had up to 50°C internally.

Therefore it was good, that the LTZ was running on high temperatures.

On the other hand, a better annual stability would have been better for the calibration of the current sources and the thermometers over the years.

But it was ok, 100ppm was required.

And the 100k/16Bit sampling capability @low SNR was the key feature of my experiments.

Well, that's yet another use case besides industrial applications.

Frank

[ltz2000]

Andreas,

Slightly off topic, but could you (or other members) recommend an easy and economical way for recording humidity. You seem to have some sort of electronic sensor with voltage output (?) in your automated data collection system.

And totally off topic, what is the name and version of your schematic editor. The "style" looks very familiar but I can't remember the name of the software.

https://www.eevblog.com/forum/projects/ultra-precision-reference-ltz1000/?action=dlattach;attach=102804;image

[babysitter]

Andreas,

Slightly off topic, but could you (or other members) recommend an easy and economical way for recording humidity. You seem to have some sort of electronic sensor with voltage output (?) in your automated data collection system.

And totally off topic, what is the name and version of your schematic editor. The "style" looks very familiar but I can't remember the name of the software.

https://www.eevblog.com/forum/projects/ultra-precision-reference-ltz1000/?action=dlattach;attach=102804;image

Just introducing a Lascar EL-USB-2 at work, right now it runs a competition against a more expensive logger. ~80 EUR and with free Software.
Very easy, less economical depending on where you are.
If you dont mind to read out data by yourself, a DHT-11 or SHT-11 might be an option, too.
Very economical, but you need some readout.

[MisterDiodes]

Another intangible factor about what DMM is more popular (or not) on a real world situation is the customer service you get when you buy a product.

For instance, the particular facility I was at on Monday has an agreement with Agilent / Keysight that if they have trouble with a unit, it just "gets fixed".  There is a local tech less than an hour away they can call in anytime, and if a unit needs to go back to Colorado for repairs, there is a replacement unit online within an hour or two.  Anything shipped to / from Colorado travels via special courier van on an airbag mat in the vehicle; it doesn't travel by common freight carrier.  That way the chance of rough handling is minimized, and you know your instrument didn't sit broiling in the sun on some shipping dock somewhere.

I know other vendors offer these kinds of services, but in this case the production manager had a lot of praise for Agilent for keeping the products running.   They do have other DMM's, but in this case the other vendors aren't able to offer the same level of customer support at this location.

You have to pay for this and be a big customer, but it is an example of why a facility selects one instrument vendor over another, and this is a factor that doesn't show up on basic performance spec sheets.

My advice to anyone shipping a high-end product:  Keep your customers happy, no matter what.  In general they are happy to pay more for a product if that includes excellent customer service.  A lot of vendors have forgotten that good customer service is a major part of the sale.

[Galaxyrise]

4) It's been 20 years, so he didn't know the exact numbers, but Q1 is only rated to a few hundred micro-amps of collector current.  [This means my single-resistor idea will *not* work! -- back to the original circuit from the ap-note!]  Q2 will have a max rating of around 1mA, and the Zener can take quite a bit more.

I dug out my LTZ1000 with the dodgy heater isolation to do some more experimenting with it over the weekend, and I can report that it still functions after a 5mA Q1 collector current for about an hour (till I remembered this post and cut power.)  It was being run with a TEC+heatsink+fan (cycling the TEC on and off), which may have helped it survive.  It should be noted that driving the transistor that hard made the TC of the reference about 2.5x worse. 

[Galaxyrise]

In the various threads concerning the voltage divider required for the LTZ1000/A (actually two dividers would help....can anybody guess where the other divider goes?)

Are you referring to the technique (I first saw explained by Dr Frank  https://www.eevblog.com/forum/projects/ultra-precision-reference-ltz1000/msg240405/#msg240405) of using one divider to trim the zener output to exactly 7, then using 10 "identical" resistors to implement the gain divider to get to 10V?

You are in the right neighborhood, Dr. Frank's setup is quite interesting but a bit complex and all those Vishays are going to cost.  It does indeed involve the LTZ1000/A's zener diode and its TC... keep thinking you are close.  The 12.5K / 1K divider could also benefit as well.

Moving this conversation here since it's specific to the LTZ1000 circuit.  Has anyone else been noodling on this?  Now that I'm in the head space of the LTZ1000 circuit, I gave it another go using the hint provided. The zener has a positive TC, the two Vbe have a negative TC.  The "theory" is that the TC of Q1 and of the zener are matched such that Vbe going up (which increases Iz as well) compensates Vz dropping (and vice versa) so VRef doesn't change. But in practice, the result is usually a net TC of about +50ppm/C.

This means the temperature set point changes with reference temperature.  VbeQ1 could be used to counterweight the set point and remove that voltage shift.  Something like the attached image? 

Does this actually matter? There should still be a single temperature that the control loop targets even without this extra resistor. This has several side effects I an think of.  First and most obviously, this resulted in a final set point about 1mV lower if the other resistors aren't changed.  Second, I think this would effectively add to the proportional gain of the PI controller.  I don't know if that would be good or bad if using the datasheet provided gain values! I think 100ppm change of this added resistor would change the final output by like 0.002ppm, so that's not an issue.  Am I missing anything?  Are there going to be any noise issues with a largish resistor there?

So, on to the ~7->10 divider.  The same trick can be applied to the non-inverting input of the amplifier, though it drops the input voltage (and very slightly raises the zener current.)  Ah... this is why you were saying Dr Frank's approach is close: The divider he uses to bring the voltage from 7.x down to 7 would also serve to reduce the TC of the circuit if it was between VRef and VbeQ1 instead of VRef and ground!  With the values I'm simulating with, the voltage would have to be dropped all the way down to 6.22 to completely cancel the TC of the reference, and dividing it that low means that a 100pm shift in that divider contributes 13ppm error in the 10V output (instead of < 3ppm by trimming to 7.)
 Trimming to 7 appears to reduce the TC by about 10%, with very little other consequence.  Not bad!

Thanks for the puzzle!

[Mickle T.]

The most thorough analysis of LTZ1000: http://bbs.38hot.net/forum.php?mod=viewthread&tid=88278 (use google translate)

[Galaxyrise]

Edwin was referring to an LTZ1000A circuit that I shared with him.  Placing 20R to 22R of resistance in series with the Zener [at the cathode] with the anode resistor at 100R, and the Q2 collector resistor at 34K lowers the +50ppm/K TempCo of the LTZ to less than 5ppm/K [and that is without the heater].

Putting a resistor at the zener cathode is the datasheet solution to reduce TC, so I wasn't really considering that as an answer.  I'm surprised only 5ppm/C can be achieved this way, but I haven't tried it yet myself.  I hadn't thought of that resistor as part of a divider, though, that's an interesting perspective.  I'd considered that since Iz goes up when VbeQ1 goes up, putting a resistor in series is a way of adding more contribution from VbeQ1 and thus increasing the contribution of its TC.

Why lower the collector resistor to 34k?  (Btw, the datasheet calls the temp sensing transistor Q2, but I think you're talking about the TC compensation transistor.)  This would seem to work against the goal of reducing TC because a higher collector current will reduce TC of the transistor and require a larger cathode resistor.  I would think that the collector resistor should be increased to compensate for the increase in reference voltage and try to keep collector current at 100uA.

Mickle T: Thanks for the link! Will have to digest that this weekend.

[MK]

I made up a model of the LTZ1000 complete circuit in LTspice guessing a 5087 for Q 1 and Q2, as they are small signal transistors, anyway it seemed to me to be on the verge of oscillation without some help to tame its frequency response, so I can understand Mickles comment about noise sensitivity.

[Edwin G. Pettis]

Hello MK,

I think that the 2N5087 probably has a little too much gain for the circuit, a minimum  hFE of 250 up to 800, I've found a lot of the 2N5087s tend to have gain around 500-600.  I would think a gain of roughly 100-200 would be closer to the mark.

[Edwin G. Pettis]

Perhaps he meant to say a 2N5088 instead, similar specs.  I missed the PNP, should be a NPN of course.

[MK]

Hi,

my mistake, it was the 5089 I used as a small signal transistor to model the npn's on the die, but as Edwin says the beta is probably too high for some aspects of modelling what is happening.

Does anyone have any other suggestions for a small die transistor to act as Q1 and Q2? but now that I know there is a huge size difference, is there an even smaller transistor available to use for Q2?

MK

[TiN]

Hey folks.
How much time does properly designed and constructed LTZ1000(A) ref takes to get stable output, counting from power on?

Also what is usual current consumption of reference, given that temp set to 12K/1K resistors and powered from +12 or +15VDC?
Is 10.5mA sounds corrent?

[plesa]

Hey folks.
How much time does properly designed and constructed LTZ1000(A) ref takes to get stable output, counting from power on?

Also what is usual current consumption of reference, given that temp set to 12K/1K resistors and powered from +12 or +15VDC?
Is 10.5mA sounds corrent?

What a beautiful banana plugs in 2002 :-D It is some type of punk?