Ultra Precision Reference LTZ1000

[plesa]

I measured 3x LTZ1000 ( non A) with SMU 2636B
#1 0,595641 V
#2 0,601472 V
#3 0,604123 V
@ 100uA/23.2°C.
Date Code 1443.
Pulse measurement will follow.

[janaf]

Thanks Plesa. I'll sum them all up as if / when more data arrives.

That's one serious SMU you have! Me envy 

[plesa]

Measured Q2 Ube in 50uA->200uA range as Bob Dobkin proposed, pulse measurement (width 1ms with 100ms period).
This can be interesting for some VoltNuts http://etd.library.vanderbilt.edu/available/etd-08262013-170646/unrestricted/Atkinson_PhD.pdf
I'm thinking about to assemble the LTZ1000 ( non A) with R4/R5 12/1 divider and Q1/Q2 current set to minimal drift (75-50uA   - > R3,R2 100k-140k).
What do you think with this modification, did someone tried it already? Or rather stay with R3,R2 in 70k reqion?

[janaf]

Done it (edit post #1159 in this thread), if you go back a page or two in this thread. Actually did lots of measurements on lot's of things. If I may say so, many go far beyond what you find in the datasheet. Some day I'll collect and summarize them all. As you have first class equipment, I'd very much appreciate if you can repeat / expand some, especially those that differ from the LTZ1000 datasheet. I also did most of my measurements on the ACH.

R2:
If you increase R2, the temperature sensitivity of the LTZ1000 circuit goes down, and crosses zero(!) at R2 around 5M5. The sensitivity /demands on R2 itself remain unchanged. The bad news is that it's hard & expensive to get ultra-stable resistors at that Mega-ohm values. 

Edit: Changing R2 up to 150K hardly makes any difference at all as far as I can see. There is a trend but you need to go into the mega-ohms to have real benefits.

R3: I could not see any advantages or disadvantages or trends by changing R3 up/down at least between 40K and 150K

[plesa]

Done it (edit post #1159 in this thread), if you go back a page or two in this thread. Actually did lots of measurements on lot's of things. If I may say so, many go far beyond what you find in the datasheet. Some day I'll collect and summarize them all. As you have first class equipment, I'd very much appreciate if you can repeat / expand some, especially those that differ from the LTZ1000 datasheet. I also did most of my measurements on the ACH.

R2:
If you increase R2, the temperature sensitivity of the LTZ1000 circuit goes down, and crosses zero(!) at R2 around 5M5. The sensitivity /demands on R2 itself remain unchanged. The bad news is that it's hard & expensive to get ultra-stable resistors at that Mega-ohm values. 

Edit: Changing R2 up to 150K hardly makes any difference at all as far as I can see. There is a trend but you need to go into the mega-ohms to have real benefits.

R3: I could not see any advantages or disadvantages or trends by changing R3 up/down at least between 40K and 150K

Thanks! I needs to read whole thread once again. There are several candidates for new R2 from Caddock.
They have 2ppm/K USF240 series http://www.caddock.com/Online_catalog/Mrktg_Lit/TypeUSF.pdf
Or in the 5ppm/K region
TK634 http://www.caddock.com/Online_catalog/Mrktg_Lit/TK134.pdf
TK633 or TK639 http://www.caddock.com/Online_catalog/Mrktg_Lit/TK134.pdf

I'm going to repeat your measurement with 3x LTZ000, probably end of next week.

[janaf]

I have looked at the USF240, they seem only available on special order. Digikey stock 5M USF340, got one. The temperature coeffs are enough but the long term drift, 20ppm/per year maybe, depends on your demands. Will check out the others.

I forget to say there may be a noise issue too. Someone theoretically minded may be able to tell if there is a noise penalty from these resistors if the noise cancels somehow. Else, I can soon measure...

[Marco]

If you increase R2, the temperature sensitivity of the LTZ1000 circuit goes down, and crosses zero(!) at R2 around 5M5. The sensitivity /demands on R2 itself remain unchanged. The bad news is that it's hard & expensive to get ultra-stable resistors at that Mega-ohm values.

What's wrong with the datasheet way of doing zero tempco? (ie. a resistor on top of the zener)

[plesa]

I have looked at the USF240, they seem only available on special order. Digikey stock 5M USF340, got one. The temperature coeffs are enough but the long term drift, 20ppm/per year maybe, depends on your demands. Will check out the others.

I forget to say there may be a noise issue too. Someone theoretically minded may be able to tell if there is a noise penalty from these resistors if the noise cancels somehow. Else, I can soon measure...

The minimal order quantity for custom made USF240 like 5.6M is 250 pieces..
USF240-5.00M-0.1%-2ppm 32 USD. Few of them should be on stock and delivered withing week or two.
The long term drift is about 10 times higher than Z foil. Do we have another alternative?

[plesa]

Componets arrived. I hope that soon will be ready first prototype of LTZ1000.
Tube of  Bourns 3250 (50ppm/K) wirewound resistors recomended by Edwin for better long term stability, I was surprissed how big they are (see photo). Also some Vishay 1280G (15ppm/K). One box of each (all 100ppm/K) Copal CT20 trimer and Bourns 3299.
And buch of additional Vishay UXB resistors. For final circuit I would like to use PWW resistors. Only needs to define right values.
For burning process 4x LTZ1000 and also 16x LM399.

[macfly]

Hi butterfly,

I think, you are right. The 'block diagram' in the datasheet on page 4 shows the parasitic diodes in parallel to the heater.  The pins 1 and 2 have to be interchanged in the schematic of the 'Negative Reference'.

Regards,

Macfly

[Andreas]

I have only spotted this in the drawing of the negative voltage reference ... - am I overlooking something ?
I cannot understand that issue is still left in the datasheet - after 3 decades ...  ?
Any further ideas to improve this implementation of the negative version are welcome ?

Hello Butterfly,

obviously no one built ever the negative reference.
And finally you found the bug of the century.

If you look at AN42 of LT (figure 68) the pins 1 + 2 are exchanged against the datat sheet.
http://cds.linear.com/docs/en/application-note/an42.pdf

With best regards

Andreas

[plesa]

I tried to compare Bours 3250W 10k and Vishay 1280G and found that the wirewound 3250 pots has smaller hysteresis or deadband.
It is simple setup, does anybody have idea how to provide more empiric measurement with normal equipment?

[babysitter]

Side note which might be interesting for supporting parts like Opamps in epoxy case:
I recently started playing with oxygen sensors and I am measuring the time it takes for the oxygen sensor to respond to a step change of ambient air athmosphere to carbon dioxide and back, going thru different silicones as cover layers.
(These oxygen sensors are little pieces of foil with a dye, using the fluorescence quenching method. Basically oxygen reduces the fluorescence activity by physical means, no chemical exchange happens. The interrogator device stimulates with light modulated by a intensity sinewave and measures phase and amplitude of the light coming back. )

2mm of silicone allow quick gas exchange at the sensor with a delay in a 15-30 minute window.
My boss suggested polyvinyl alcohol as a barrier material to prevent gas exchange, and even a tiny little layer was working so well - stopping dead, no visible response after hours. I read up a bit on this stuff and found that it is also used as a humidity blocker in foods packaging. Its available cheap as a form releasing agent.

Maybe this stuff can be used to supress humidity exchange of epoxy cased parts.

[richiem]

Hi babysitter -- what's the life of PVA used this way -- will it last a long time?

[babysitter]

I just had the first experiences, on flexible substrates this form releasing agent dries but gets brittle when substrate gets moved, maybe a cover layer is required. I think about looking for known humidity sensitive parts (right in this thread?) , maybe I buy/borrow/beg/steal some and just paint them.

Also, my own experience is using it as a barrier against exchange of oxygen and co2.

(Maybe the first part i should paint is a humidity sensor?)

[Kleinstein]

For the reference part, there is the option to use a closed metal casing with the few signals going in and out through ceramic feed through capacitors. Sometimes these caps are misused as a feed through to vacuum vessels - so they should be rather thight. Keeping HF signals out is not that bad either.  I would not worry so much about humidity at the OPs - the LF noise of the LTZ1000 is still way higher than typical drift / LF Noise of good OPs like the LT1013, LT1007, ... even in plastics case.  Its the whole circuit that would profit from stable humidity levels.

I would be sceptical that there is much way for improvements through this slight compensation of residual TK effects with this extra 5.6 M resistor. Feedback from the heater is at least sensitive to the termal enviroment. So having extera / other  thermal insulation would need a different resistor.

There might be a chance to tweek / improve temperature regulation a little:
With the resistor as heating element, the regulation loop is nonlinear (in this respect the LM399 is better designed). So tuning of the temperature regulator must be done for the worst case (highest gain), which would be very low temperatures. There are two ways to improve here: either linearize the heater through some extra nonlinear circuit  or avoid the very low temperatures by an extra layer of temperature control. This extra layer could be a moderate temperature, only aktive at low to normal enviromental temperature.

[alanambrose]

Hi,

Wow - what a thread - 265 pages of dense pdf not to mention the linked docs. Took me a week of spare time reading and I feel I need to go back and read it all again to see if I missed anything
Very educational, and many thanks.

To add a bit more data and ask a couple of questions...

I just acquired 3 x 03458-66509 ref boards with the aim of building a sort of Fluke 7001 clone. Attached is a little basic measured data FWIW - no noise or stability #s yet. I see that the accepted wisdom is that HP set these to run at ~90C rather than a preferred ~60C.

I also see from the 3458a schematic (e.g. http://www.maxmcarter.com/vref/hp3458-1.gif ) that the bases of the embedded temperature sensing transistor and the temperature control transistor are connected off the board as J400(1) and J401(1). I can't quite see from the CLIP where these go to and what they're used for. Anyone else figure this out yet?

Also I need to make some covers for the refs to keep out the drafts I can't see any detailed images of the HP design, so I've spun my own (also attached).

Any feedback? - I really don't know much about thermal. I've taken some inspiration from the L399 cover which has a double wall.

Regards, Alan

[plesa]

I hope that your boards were cheaper than 03458-66509  on ebay ( 99 USD). 
I suppose that new LTZ1000 and PWW resistors from Edwin Pettis can be cheaper alternative.
Of course with proper burn-in and with lower temperature setpoint than HP used.

But everything depends on your reference purpose ( consider using multiple LM399).

[richiem]

@alanambrose -- just ignore those base connections that go off-board. I'm not sure where they go either, but I think they are used for temp monitoring in the 3458. For your purposes, they are unconnected.

I like your idea of having the three boards as a 7001/Datron 4910 type standard. I can fully recommend changing R411/15k to 13k using Ed Pettis' PWW resistors, having done this for my 3458 board-based standard.

I've used LTC2057s for buffer amps -- one unity gain to buffer the 3458 board's zener output. and one with gain to provide a 10V output. Were I to do what you are doing, I would buffer the three board individually, and have a fourth set of buffers that mix the outputs of the three boards to give an average of their individual outputs. Good luck to you on this.

[alanambrose]

>>> just ignore those base connections that go off-board. I'm not sure where they go either, but I think they are used for temp monitoring in the 3458. For your purposes, they are unconnected.

Thanks, but the reason I asked was that I had a suspicion that we might learn something useful. If HP decided to route the connections off the board they must have had a reason. Could be something simple like identifying when the LTZ was up to temperature, but could be something smart. There's a sense in this thread (and others) that HP could have been a bit cleverer with their LTZ ref board - not making a lower temp / lower drift version for metrology labs etc. However, as this is still the go to 8.5 digit meter and it has been since the early 90's, I'm happy to give them the benefit of the doubt. Will take a closer look at the CLIP.

Alan

[Dr. Frank]

Hi, these additional connections are not used in the 3458A, as far documented in its CLIP.
There's only this connection to -18V over a resistor, which is intended as a current cancellation path for the zener and collector currents, and the OpAmp supply. So the GND supply towards this PCB is virtually current free.

But this PCB may also  work well as a reference w/o this feature.

[Andreas]

  The LTZ's 1/f noise is guaranteed to be <=2uVpp, but it is on a Poisson distribution, and about 90% of them will have exceptionally low 1/f noise.  This is not something that most hobbyists will even be able to see, nor should they worry about it.

Why not?
It is not too difficult to measure the noise.
Its simply a low noise amplifier + a oscilloscope.
And its the simplest way to find out wether a reference behaves "strange" or if it works most probably on spec.
Its one of the parameters that I want to know besides T.C. and long term drift.

With best regards

Andreas

[6thimage]

Was HP's initial decision of using a 90 deg C oven temperature over 60 deg C anything to do with ambient temperature? As if they used 60 deg C the multimeter wouldn't be able to be used (comfortably) in a 50 deg C environment (i.e. racks).

I know there is a lot of magic that goes into making a good ltz1000 circuit (and I'm not trying to fanboy HP-Agisight), but it seems strange that they would have gone with a 90 deg C circuit, without knowing it has worse stability than a 60 deg C. I would have thought that the engineers at HP would have been on first name terms with the right people at Linear and, if that was the case, would this choice be purely a trade off they needed to make rather than incompetence? But this does beg the question of why they never offered a high stability option with a large note of 'keep the unit's temperature below x'.

I'm not sure if anyone has asked this (or answered this) - I've tried to keep up with this thread, but there is just so much information.

[richiem]

Max spec'd operating temp for the 3458A is 55°C -- assuming an internal temp of 15°C above ambient, that means a temp at the LTZ of 70°C -- so running the LTZ at 90°C +/- 5 makes a lot of sense. I agree that HP should have made a metrology version (do we know they didn't?) with the ref running at lower temp, which is one (but only one of several) of the things that Fluke has done with their high accuracy version of the 3458.

I take Ken's observations seriously, and he and Frank are in agreement that the "current-cancelling" of -15V thru a 2.67K resistor to ground is most likely unneeded; nevertheless, I am using that on my little LTZ box.

[alanambrose]

Well my approach is to try and extract everything from the other designs even if the conclusion is that some things may not be optimal or that the designers might have had slightly different objectives from mine. I see that while the temp sensing transistor Vbe connection isn't used in the CLIP, the heater transistor connection apparently is. It runs through a DG211CJ analogue switch and then to HEATER/R123 (page 51 on the pdf I have). Curiously, this then plunges into the middle of the 'sentry DC input amplifier' - I can't quite see why ATM. Maybe the two connections might also have been used in test / burn-in? I did notice some production-style test points on the bottom of the board also.

Re LTH A version selection and temperature objectives - I think it's fairly well accepted that in that period HP was selling a vast quantity of equipment to the military, hence the spec. Some folks on volt-nuts think that we're unlikely to see those kind of volume sales to a price-intensive customer again soon - and therefore the likelihood of a vendor improving on the tech in the 3458a is low. I guess metrology sales were fairly low in comparison and anyway HP's partner, Fluke, was king of that game - so maybe HP marketing didn't want to go there. It would also have required HP marketing dept to understand both the ref temperature options along with the market opportunity (as apparently Fluke did) and like what they thought.

Re the thought that these boards might be dodgy - I'm guessing they're either test cast-offs that have been sitting around for 20 years somewhere, or that they're spares that have been sitting in some quartermaster's stores for the same period. (They have the HP logo on.) Guess I won't know until I measure the TC / drift / noise, but also I'm planning to spin my own boards at the right time. A couple of questions if I may:

+ how do you guys measure TC? Have you got test chambers?
+ was there a conclusion from the 'best' way to get from 7.x -> 10V discussions?

Alan