Ultra Precision Reference LTZ1000

[Edwin G. Pettis]

To MAX666,

Technically speaking, they are but are not wound by hand, the hand is only used in guiding the wire as it is wound onto the bobbin and yes, there is a technique to doing it correctly, otherwise the resistor may not turn out quite right.  In PWW resistors, all the details are important to the final outcome (as it should be in any precision resistor), very small variances can lead to significant effects on the final characteristics, for instance, it can change the TCR from being very low to being significantly higher, perhaps 3 to 5 times higher is quite possible, it has happened.

Theoretically, it certainly is possible to wind more than one individual resistor on a bobbin, however there are many constraints as to whether or not it can be practically accomplished, in most cases it turns out to be impractical.  For instance, two resistors on a single bobbin almost always requires four terminals, the smaller bobbins cannot accommodate two lead assemblies at each end, the bobbin is too small.  For the larger bobbins, they must be retooled to accept two lead assemblies per end, new collets must be purchased to accept the double leads (and they are quite expensive) and welding the wires to the terminals becomes a bit restricted by having two leads near each other.  Having multiple lead assemblies also complicates the calibration process as it is very difficult to rotate a resistor with four leads sticking out of it and keep a true four terminal, low resistance contacts.

In the case of the resistors being mentioned here, the 120R0 and 12K (for example), a larger bobbin must be used as each value (using the same wire size as proposed) basically fills a bobbin (8E16 size as an example, my 805), a possible candidate may be my 807 (8E24) but that still leaves the termination problem.  I have considered such bobbin modifications in the past and other ideas to produce a possible multi-resistor bobbin design, they do have some advantages but the major drawback is the cost of implementing them.  They must be designed in such as manner as to not compromise the precision characteristics of the resistors I currently make while giving some advantage to warrant the cost of it.  You are not the first one to ask about or propose such resistor designs, in the past, the easiest solution was to put the resistors into a close fitting container and fill it with good thermal conductive filler.  This did save on expensive tooling but was still more expensive than two (or more) individual resistors.  Another method which works better than you might think is to wrap copper adhesive tape around the resistors, while this does increase the thermal time constant a bit, it works quite well if the ambient temperature does not vary wildly quickly.  As is the usual suggestion when working with temperature sensitive circuitry, encase the whole works and eliminate drafts, that is very important.

[janaf]

Welcome Lars!

A2 opamp keeps the voltage over R1 and Q1, the temp-compensation transistor (pin 4 & 5) equal at all times. That's what keeps the output voltage constant and variations in R1 are to a high degree compensated for. One of the very clever details in this circuit! Variations in "R6" would not be compensated for. I guess that's the difference. 

[richiem]

LTC2057 is not a $20 opamp -- I bought some for under $5 ea.... but they are SO8 pkg which is the smallest I can work with by hand. Having the chopper amp is essential for temp reasons unless you are going to ovenize the circuit, and the 2057 is the best chopper amp.

[Andreas]

Lars Walenius

Welcome Lars !

With best regards

Andreas

[Galaxyrise]

If the 20ohm resistor "R6" according to Andreas has 1/70 sensitivity and 120ohm R1 according to Janaf, Lymex and TiN has about 1/700 sensitivity it seems strange that they should track?

You're comparing the sensitivities from different circuits.  Changing R1 changes the zener current, which has a different result in the output voltage based on whether or not R6 is present.

[lars]

If the 20ohm resistor "R6" according to Andreas has 1/70 sensitivity and 120ohm R1 according to Janaf, Lymex and TiN has about 1/700 sensitivity it seems strange that they should track?

You're comparing the sensitivities from different circuits.  Changing R1 changes the zener current, which has a different result in the output voltage based on whether or not R6 is present.

Ok.

Anyone with an idea how the output sensitivities for R1 and "R6" changes with "R6" inserted? As I understand they need to cancel each other if the "tracking pair" should work?

[Edwin G. Pettis]

Actually, the mandrel is that part of the bobbin on which the wire is wound, a collet is what holds the lead assembly of the bobbin and the collet is screwed into a shaft driven by the motor.  The diameter of the mandrel is what limits the possibility of putting two lead assemblies into each end of the bobbin, this is a physical limitation.

Evanohm is welded using a CD welder, that is the simple part of it, the rest of the detail is in the design of the resistor, lead assemblies and electrodes.  As I told DiligentMinds, it is not the type of welding that can easily be accomplished by a novice in order to achieve a quality weld.  If you are just interested in trying your hand at making a basic resistor with a fairly low TCR, then I suggest Alloy 294 (Constantan) which can be soldered too and has a moderate resistance.

DO NOT under any circumstances use acid flux on electronics, there are no exemptions to this rule unless you are a plumber working with copper pipes.  Zeranin-30 (a variant of Manganin) can be soldered to with regular solder but these alloys must be heat treated after 'winding' to remove stress and being mainly a copper alloy will remain sensitive to additional heat, soldering can and will cause a permanent change in the heated wire.  These alloys are of very low resistance so as a matter of practicality, you cannot wind high resistance values unless you have lots of space.  If you imagine that you might be able to make a high precision wire wound resistor with very low TCR on your own, I'm sorry, it isn't going to happen.  If that was the case, everyone would be making great PWW resistors and that is just not the case.  The same thing can be said for metal foil resistors, that isn't something you're going to do on a Saturday afternoon in the garage either. 

[janaf]

Seriously guys, despite the amount of effort people are putting into the voltage references, we are not on solid ground with the LTZ1000. See current discussion on resistor sensitivity analysis and dr Frank's projection of need for improvements in the coming years.

Frankly, the datasheet for the LTZ1000 is rudimentary only. We need a totally re-written revision up to today's standard, on par with LT's current datasheet. That would be an update in several orders of magnitude in information for users. We should also have a SPICE model for the LTZ1000 itself so one could simply evaluate things like. LT surely have tools and knowledge to make measurements an to convert measurements to model.

What are the odds of this happening? Is it likely? Is it simply a cost for them that they will not recover?

If it's not likely LT will do it, I think community should and I'm willing to do my share.

- Collect and organize data that is available. A wiki?
- Do "complete" static measurements on the LTZ1000 including temperature dependency
- Create a spice model
- Measure and model aging ...

Measuring the LTZ1000 should be doable. Two transistors and a zener which all can be measured independently or together.

Any takers? Thoughts? Realistic? Has any of this already been done? Am I out in deep waters? Stepping on toes?

Does someone have the right LT contacts to forward this message to?

Janaf

[TiN]

Another reason is likely that main customers for LTZ who buy it in thousands are likely already did many years of characterization and would not need model/update on datasheet anyway.

As of us, I do want to commit in own way to get details, by doing next:

* Build prototype boards to measure real data. Done, 5 boards built (3 x ACH, 2 x CH, LTC2057 on four of them, LT1097 on one).
* Burnin 1000 hours all boards. Done.
* Build preamp for noise measurement. In progress, schematics phase.
* Charaterize refs with temperature. Need define method. Have TEC and KI 2510, waiting CNC vendor to make enclosure metal box.
* Characterize voltage dependency. 24VDC to 8VDC power in 0.1V steps (KI 2400 SMU) - pending
* Build reference backplane with 10VDC, 2VDC and 20VDC output, both positive and negative. Idea stage
* Build 24bit DAC and 24bit ADC for datalog. In idea stage
* Capture realtime data and show online during 1 year period - testing frontend with javascript now

It should be both static data hosted on dedicated public site, where anyone can gather verified data and learn thru analysis, and where newcomers can get idea of whats current status of things, without reading 73 pages of forums. While as participants progress, we keep discussion using this forum as communication platfrom.

I get feeling that some of people would like to try LTZ for their fun, but after reading page or two here where talks about custom unobtanium resistors, they got carried away rather quickly.

P.s. as end result my vision is to have practical application how LTZ ref used to generate stable current like 10.00000mA standard or <1ppm DAC/ADC system. So I  going forward to use my site as hosting for all findings/data already. Its own server, so its have no space/toolkit limitations. Trying to get ahold some javascript and python to make webpage where I can display measurement data realtime from LabView. (i am hardware guy, not software wizard (

[macfly]

......
The 50ppm/K natural [unheated] TC of the LTZ is reduced to something less [maybe 2-5 ppm/K], and so when you turn on the heater circuit [which holds +/- 1.0mK], the TC drift is now a great deal less than 0.05ppm/K [or, if you want, the heater R4/R5 ratio no longer has to be so accurate to achieve 0.05ppm/K TC].

Hello,

I have made a Q&D test with one of my LTZ1000ACH's with external heating. At a chip temperature between 40 to 45 deg. Celsius the temperature coefficient drops vom 28ppm (without cathode resistor) to < 1ppm with a 20 Ohms S102K as cathode resistor. Anode resistor was a 120 Ohms S102K.

Regards,

macfly

[TiN]

That sounds little similar to CSNG ones I got before (2.5 ohm). Ended up with around $75 per each.

[MK]

......
The 50ppm/K natural [unheated] TC of the LTZ is reduced to something less [maybe 2-5 ppm/K], and so when you turn on the heater circuit [which holds +/- 1.0mK], the TC drift is now a great deal less than 0.05ppm/K [or, if you want, the heater R4/R5 ratio no longer has to be so accurate to achieve 0.05ppm/K TC].

Hello,

I have made a Q&D test with one of my LTZ1000ACH's with external heating. At a chip temperature between 40 to 45 deg. Celsius the temperature coefficient drops vom 28ppm (without cathode resistor) to < 1ppm with a 20 Ohms S102K as cathode resistor. Anode resistor was a 120 Ohms S102K.

Regards,

macfly

Looks like you got lucky getting the value close to right.  I guess we are calling this new resistor "R6", though I think that conflicts with other resistors in the original design.

This technique will be at it's best if R1 [the anode resistor] and "R6" [the cathode resistor] track well with temperature.  At a minimum, they should be placed close together on the PC board [flat side to flat side for the VPG foil resistors].

I asked VPG how much it would be for a VHP100 in values at below what the data sheet has as a lower limit [100 ohms], and they said that they could do it, but it would be 3-4 times the price of a similar 100-ohm resistor with the same specs.

So, I started looking at alternatives.  It turns out that VPG makes a resistor that is specifically designed for low-ohms values in the range we need, and that is the VPR221Z.  If you made an R1 and an "R6" with these, their heat-sink tabs [made of aluminum] could be glued together [using thermally conductive high-temperature adhesive] with the leads on opposite sides.  After a bake at 125C for 72 hours to drive off all moisture, they would then be sealed in dual-wall PTFE/FEP heat-shrink to make them "almost hermetic".  These are 4-wire units, but it doesn't stop you from using them as 2-wire resistors [just use the 2 leads on each side connected in parallel].  That will shift the TCR a bit, but that's OK in this application.  So now, we have a hermetically sealed [well, "almost hermetic"] resistor pair that is thermally coupled.  I don't know if it would do any good [or if you would just be wasting your money], but you can ask VPG to make these TCR track to +/- 0.1ppm/K for more money [and be sure to tell them that you will be using these as 2-wire resistors so they are trimmed correctly].  They may TCR-track naturally anyway, or maybe they could be selected as an alternative?

With "R6" in place, the heater circuit's divider now has relaxed specifications, and that opens up the possibility for a number of lower-cost divider techniques that were not previously available.

If you were to use a 4 lead resistor for R1 then you could use one of the leads for the ground side as the sense for the 10 volt low.

[janaf]

Yesterday I received WW resistors from Pettis, in all took a couple of weeks including shipping to Europe.

Today, hermetics arrived from Vishay, earlier than expected, about 12 weeks in total. 

[janaf]

Sure, I can post data on these. I'm waiting for some components to build a mini-thermal chamber but depending on time, could do some quick-and-dirty test before the thermal thingie has been built.

[richiem]

I have gotten my spare HP 3458A reference board put into an aluminum case with power supply and buffer amp. Thought I'd share some initial data. The R411/R412 ratio is 12.5:1, and the buffer amp for the zener voltage output is an LTC2057 in unity gain. I'm using a metal foil 12.1k with a 402R metal film in series to get 12.5k -- I've ordered resistors from Edwin Pettis and will replace that combo with a 12.5k ww. The top of the LTZ and the bottom side of the board under it are sandwiched in open-cell foam for minimum air current disturbance yet with some heat exchange.

The main supply is from an LM317 at 17.3VDC (couldn't get to 18V with my little wall-wart transformer which puts out a *nominal* 17VAC -- I wanted this to run with mains voltage as low as 110) and, like Max Carter and HP, I put in a -15V supply from a 79L15 for the 2.67k R419 resistor to nowhere. I'm still not sure which LTZ1000 this is -- non-A or A; Dr. Frank thinks HP only used A versions.

I've been moving the box from 22°C to 24.5°C and back, monitoring temps with a calibrated thermistor, with several hours between moves. The output variation is 0.3ppm/°C as read by my HP 3458A running at a nearly constant internal temp of 36.4°C, (+/- 0.2°C). Hope this info is useful to those building an LTZ1000 based supply modeled on the Linear Tech-HP design.

It will be interesting to see what difference if any the ww 12.5k resistor makes. I put in a second LTC2057 to boost the 7.13453xV zener output to 10V. I used metal film resistors and cermet pots for feedback and cal, and unfortunately, but not unexpectedly, the 10V variation over temp is 25ppm/°C -- useless. Edwin's ww resistors and Vishay foil trimmers should help here.

[Dr. Frank]

I got the question on the other resistors. In my measurements

- R2 is slightly more important than the datasheet indicates, -0.4ppmV/100ppR compared to 0.3ppmV/100ppR in the datasheet. Not a very significant difference (apart from the sign!)
- R3 was less sensitive. I got -0.07ppmV/100ppR compared to 0.2ppmV/100ppR. A quite significant difference but low values anyway.
- R4/R5 I got slightly higher value than the datasheet and same sign, positive. Not very significant difference.

- For R2 the sensitivity decreased a bit with increased resistor value (range 40K-100K), but not very significant
- For R4/R5 the sensitivity decreased a bit with higher ratio/higher temperature setting, but not very significant
- For R3, the sensitivity was unchanged over the range (40K-100K)

Hi Jan,

your measurement results were somehow familiar to me, and now I know why..

The Chinese blog bbs38hot.net contains several similar threads about the LTZ reference, and in one the author bg2v0 also measured the influence of the different resistors.. exactly the same what you've found. See screenshot

https://translate.google.com/translate?sl=zh-CN&tl=en&js=y&prev=_t&hl=de&ie=UTF-8&u=bbs.38hot.net%2Fthread-46-1-1.html&edit-text=

Frank

[janaf]

As I got some requests, I will post more, but give me a couple of days.

Thanks for posting the Lymex data. If we multiply my data by 100, for 100ppm resistor changes we get:

-	Datasheet	Lymex	Janaf
R1	1	0.14	-0.14
R2	0.3	0.4	-0.4
R3	0.2	0.03	-0.07
R4/R5	1.0	0.95	+1.2

A couple of conclusions:
- It seems the R1 and R3 values are not as critical as given in the datasheet. Or both Lymex and I are wrong.
- Put your money on R4/R5 and R2 while R1 and R3 are less critical.

In the end, the errors of resistors are uncertain , it is not meaningful to dive too deep into the decimals.
EDIT: changed signs in the table.

Frank,

Great!
It may be the same guy "Lymex" that I compared to before.

In any case, I'm pretty confident about the data now as at least two other people get very similar results to mine.
Edit: TiN being the third person...

[janaf]

I'm appending a file with diagrams for the measurements I have done of the LTZ1000ACH. So far. To be continued and cleaned up. Most of these measurements, I have not seen published elsewhere.

I estimate that what I publish now represents less than half of the measurements that I intend to do.

If someone from LT sees this, some samples of LTZ1000ACH would be appreciated as support for the work

(EDIT: changed attached file to pdf)

[TiN]

My word does not show any data

[janaf]

Sorry, should have been the pdf. Uploaded now.

It's just the diagrams, no data.

[janaf]

It's same family as Star Office but OpenOffice 4.1. Essentially MS Office equivalent, it has less detailed functionality than the full MS equivalent but works very well. And is free. I have it on my Samsung phone too. MS and OO should be able to exchange files but in in my experience it only works to 90% .

OO suffers from the usual word processor problem that "everything moves", i.e. not good at neatly aligned, repetitive, identical plots.

[richiem]

Presumed reasons -- thermal and mechanical isolation.

[MK]

As previous poster, much less heat leak, and much smaller mechanical strain on the kovar leads, but at the expense of much increased fragility. Impressive though, but still not sure how much it gains one.

[janaf]

As said, probably the ideas is to isolate thermally. Not convinced that it really helps. Japanese? Seen some similar designs on Chinese sites.

Looks cool anyway!

[chickenHeadKnob]

It's not for mechanical reasons since the LTZ1000/LM399 aren't sensitive to stresses on their leads (Cf. Datasheet). It makes sense for thermal reasons, but although eye catching, this design is far from optimal : The aim here is to make the temperature uniform among the leads. This should be done by a lot of copper near the solders, and as small spacing between traces as possible. The "flying" traces should be made as thin as possible to avoid thermal gradients due to another thermal source/leak on the PCB. The quite large traces unfortunately ruins the benefits of the slotted areas.

35µm copper is about 10 times more thermally conductive than its 1.6mm FR4 substrate.

In attached file : A draft of what I believe more optimal.

The width of the traces was the first thing to catch my eye, vis a vis the LTZ1000 board in the new 34470