Ultra Precision Reference LTZ1000

[Andreas]

It's nigh impossible to find new parts in hermetic packages any more.  I can find some CERDIP 1013s on ebay, though.  Are these opamps humidity sensitive at all?  "For humid environments, surface coating may be necessary to provide a moisture barrier." is the only datasheet reference I can find to it (in the 2057 datasheet.)

You can also find CERDIP LT1013A on RS-Components. (from TI)
If the chip is sensitive to forces any humidity will change the characteristics. Either the PDIP package swelling or from PCB changes which are creating forces to the chip through the PINs of the package.

Even a hermetically packaged voltage reference like VRE3050A or AD586LQ or LT1236AILS8-5 is sensitive to forces through the PCB (like humidity changes) if not carefully decoupled from the board.

A chopper like the LTC2057 does not need to be in a hermetically package. Due to the self adjustment the effects of humidity will be canceled out by the chopper.

With best regards

Andreas

[mrflibble]

It's nigh impossible to find new parts in hermetic packages any more.

I came across these recently:
http://www.digikey.com/product-highlights/en/ls8-reference-family/51750

[dannyf]

if I were designing a replacement board for the 3458A,

Wouldn't that depend on why you are redesigning the board?

Many times, we don't design for the highest performing products.

[Galaxyrise]

You can also find CERDIP LT1013A on RS-Components. (from TI)

Obsolete packaging according to the datasheet.  rs-components in the US, aka alliedelec, does not have it.  I went to the german site, and there indeed was the AMJ... but boy do you pay a lot for the M version! 

If the chip is sensitive to forces any humidity will change the characteristics.
...
A chopper like the LTC2057 does not need to be in a hermetically package. Due to the self adjustment the effects of humidity will be canceled out by the chopper.

I contacted Linear Tech, and asked about this.  The answer is that the LTC2057 will self-compensate for any temperature and/or humidity caused drift-- this is caused by the die being "stressed" when the epoxy package changes dimensions, and since it is "inside the zero-drift control loop", it is compensated for.

Makes sense, and confirmed by Linear, even!  Thank you both   The 2057HV is surprisingly affordable, too.

(such as ... Peltier coolers, etc.).

This is what I've been playing with, and part of why I'm very concerned about humidity.  So far, I can get my, uh, "oven" to Ambient-10C with 1W and Ambient-18C with 4W; still a lot of things for me to learn and play with.  Temperature controlled boxes have been fascinating so far.  I'm aiming to eventually implement your suggestion from much earlier in this thread.

[Andreas]

Obsolete packaging according to the datasheet.  rs-components in the US, aka alliedelec, does not have it.  I went to the german site, and there indeed was the AMJ... but boy do you pay a lot for the M version! 

Hello,

you looked at the wrong datasheet from LT. On Texas Instruments the package is still active.
The chip is identical to the LT chip according to a interview of a LT representative in a paper.

And yes: precision is not cheap if not using low noise choppers.

With best regards

Andreas
 

[Blackart]

Gentlemen first post here.

I recently acquired a Datron 4700 and I plan to "have a go at calibrating it" so I need a 10V ref. I would then use the 4700 etc to keep my 34401A and bank of old 3400A/B/400E/F/Gs etc in line. I have been eavesdropping here and on volt-nuts for a week or so trying to get the threads/approaches clear in my mind.

I think Ive decided to go down the LTZ1000 path with ;
approx 45C ref temp
vishay S102 resistors plus a couple of others I have in stock ( better if the $ are kind when I ask texasinst for a quote )
LTC2057 opamp
unbuffered ref out
10V buffered ref out
self design PCB
powered by a double regulated low noise supply Ive already done.
lots of thermal & magnetic/RF shielding

If your listening, I was wondering if a couple of you could clear up a couple if things I can't work out from my couple of readings of the blog.

Dr Frank
You made a ref with 2 off LTZs, it look like one has the ref and heater supplies buffered by a transistor and the other is driven as per the LT data sheet. Why ? was this to compare the performance of the two approaches ? If so what was the outcome ?
OR
Was the above idea dropped and you were you seeking to combine the references as per the LT data sheet to improve stability and noise etc if so how did this work out ?

Andreas
You also added the ( FET in this case ) buffer i.e. a BF245. How do you think this performed ? as I thought that any self heating in the driver opamp would be nulled out in the FB in steady state. Or is the ref noise/drift such that the opamp has to work hard to maintain a constant current ?

regards Lucas

[Dr. Frank]

Dr Frank
You made a ref with 2 off LTZs, it look like one has the ref and heater supplies buffered by a transistor and the other is driven as per the LT data sheet. Why ? was this to compare the performance of the two approaches ? If so what was the outcome ?
OR
Was the above idea dropped and you were you seeking to combine the references as per the LT data sheet to improve stability and noise etc if so how did this work out ?

regards Lucas

Well, I really tried both designs.
The transistor gives more current capability at the output jack, as the OP07 is limited to a few mA only.
On the other hand, it gives more gain, so that at last, I skipped it again, because I feared instability of the circuit due to oscillation.

The whole circuit is quite sensitive to external AC noise, which may lead to disturbance and even unlatching of the temperature regulation.

The transistor is really not needed at that point, and the additional resistor from V+ to the reference high side unloads the Op07 also.

My advise: keep the circuit as simple as possible.

I would also not use a chopper OpAmp, as they create more noise than regular ones.

The LT1013 do not influence the stability so much, so they are fully sufficient.
ank

[jd]

I would also not use a chopper OpAmp, as they create more noise than regular ones.

With the exception of the LTC2057(HV).  This part exceeds the LT1013 in almost every spec, including noise [and it is better in *all* of the important specs for this circuit].  Note that the LTC2057 is a completely new architecture than most chopper amps [and Linear Tech is not revealing *exactly* how they did it].  For other chopper amps, I would agree with Dr. Frank.

MAX44246 looks quite useful for DC precision. It looks like it might be lower noise than even LTC2057. Also, is chopper amplifier noise not of a different character, being free of 1/f behaviour? So longer integration times continue to reduce fluctuations. Perhaps for use with a null detector/comparison circuit or 7.2:10V booster.

(But I think Franks point was that in the actual LTZ1000 reference circuit the LT1013 opamp errors are not significant in any case).

[Andreas]

Andreas
You also added the ( FET in this case ) buffer i.e. a BF245. How do you think this performed ? as I thought that any self heating in the driver opamp would be nulled out in the FB in steady state. Or is the ref noise/drift such that the opamp has to work hard to maintain a constant current ?

Hello,

there were 2 reasons for the FET:

1. to keep away heating from the OP-Amp.
    5mA * 7V = 35mW @ 130K/W gives around 5K self heating.
    Offset drift is up to 2.5uV/K = 12.5uV
    that sounds not much but:
    12.5uV/2mV/K Heater setpoint = 6mK Temperature setpoint drift.
    together with the 50ppm/K drift of the zener there is up to 0.3ppm drift of the LTZ.

2. Battery powered design with only 14V regulated voltage.
    I simply feared that with 1V less voltage against the 15V datasheet design
    there would be not enough headroom to fully stabilize the zener current loop.
    The negative pinch off voltage of the FET compensates for the lower power supply voltage.

Other effects are that with lower load current and with lower excitation of the output the open loop gain (= current regulation precision) increases.

So in a design with LTC2057 only the second point would still have some meaning.

With best regards

Andreas

[Andreas]

(But I think Franks point was that in the actual LTZ1000 reference circuit the LT1013 opamp errors are not significant in any case).

Note: In the more serious (long term stable) desings (AN86 + Datron reference board) they used a LT1013 in the now obsolete hermetically TO-99 case.
Ok this would also be a possible enhancement for the HP-reference board.

With best regards

Andreas

[Andreas]

As far as the LT1013 is concerned, the note on page 5 of the data sheet for the LTZ1000 says:
"BOTH A1 AND A2 CONTRIBUTE LESS THAN 2uV OF OUTPUT DRIFT OVER A 50°C RANGE"

Hello,

With my calculation above this is only possible with the "typical" drift spec of a LT1013AC = 0.3uV/K.
With the maximum spec of a non "A" device of 2.5uV/K the 2uV drift is already eaten up by the self heating of 5K.

With best regards

Andreas

[eurofox]

Hello Dr. Frank, Andreas, all VoltNuts,
                                                      Would it be possible to send me your most stable implementation containing all modifications (minor - major) as schematic for review - so that we can reach common sensus - I am willing to put effort in it - in order to create a nice schematic in Altium out of it and upload schematics as (*.schdoc ) and as(*.PDF) on EEVBLOG so that people on the forum have a reference and do not get lost in the big forest ... - as a next step we can make up a list of best engineering practices to implement the LTZ1000(A) as it should be done with and without extra temperature controlled mini-oven ...
I do not really need a nice CAD drawing, something hand-written (scanned) to start with is fine, as long as the paper contains all major & minor details ... it is OK for me.

Thank You !
ps.: You can mail it to forum or send me a PM.
                                     

Good idea because now information is scattered.

I already have a word file with a lot of information that I collected and I have myself a LTZ1000 with a slotted PCB and isolation.
I use the extra 70k resistor to lower the working temperature of the LTZ1000.


eurofox

[Andreas]

Good idea because now information is scattered.

I think all information is within this thread.

And: there is no single truth. Every cirquit built here has its special targets.
It makes a difference if you want to have a "primary standard" a "working standard" or a "transfer standard".

With best regards

Andreas

[TiN]

And collecting related data will be thru personal preference prism, some might think about items which are not important to others, and vise versa.
But in case such work to be done - do it public, so others can benefit.

I don't have any experience yet, and collecting ideas and data, so following this thread as well. Just calibrated my EDC MV106 according to my calibrated 2001, which is only 4 days after cal in Tektronix/Keithley lab.
My first reference will be LTZ1000A, with direct output. MV106 was around 120ppm off. I have it constant powered on, so just another reference for future tests.

Also curious, why nobody tried implementing reference with resistor networks? Datron, Fluke in their 8508A are using networks, which can be explained by matched tempco. After all, on output it's important
to keep constant resistor divider ratios, not actual resistance values of individual resistors. And networks fit this job best, as their elements are fitted on same substrate, and drift same or very close,
reducing mismatch significantly.
Am I missing something?

Maybe better option would be order needed network from VPG? I'm thinking of doing so, but again, i have zero hours of practice with LTZ, only diode resting in a bag waiting for PCB.

P.S. Fluke 8508A used AD823A and LTC1150 in their reference + custom R network.

[Andreas]

Am I missing something?

Hello,

for first: usually you will have to order large quantities of those "specialized" networks. (so no hobbyist will afford it).
second: the tempco will only cancel out if the resistors are of similar kind. (resistor film/foil thickness).

If you pair a large (70K) + a small (120R) on the same network. The network will either use a very large area (costs) or the tempco will not cancel out.

With best regards

Andreas

[Dr. Frank]

Also curious, why nobody tried implementing reference with resistor networks? Datron, Fluke in their 8508A are using networks, which can be explained by matched tempco. After all, on output it's important
to keep constant resistor divider ratios, not actual resistance values of individual resistors. And networks fit this job best, as their elements are fitted on same substrate, and drift same or very close,
reducing mismatch significantly.
Am I missing something?

Maybe better option would be order needed network from VPG? I'm thinking of doing so, but again, i have zero hours of practice with LTZ, only diode resting in a bag waiting for PCB.

P.S. Fluke 8508A used AD823A and LTC1150 in their reference + custom R network.

The answer is obvious:
- very high cost for low volume production
- useful for the temperature regulator only, i.e. the 12k over 1k divider

If you study the calculations in the datasheet, or what I have calculated in this thread, you would recognize, that especially this divider ratio of 1:13 should be very stable, by matching the T.C. of both resistors.
Vishay offers such matched dividers with 0.2ppm/K, and these should also be oil filled for stability of ~4ppm/6yrs.  Resulting impact on the LTZ circuit would be 0.002ppm/K and 0.007ppm/yr.

The 70k resistors don't need to be that stable, as all their drifts are attenuated by a factor of 500.

The 120 Ohm resistor should have low drift values, i.e. low T.C. and low long term drift, i.e an oil filled VHP202Z should be used. The 100:1 attenuation of its drifts would give an impact of 0.01ppm/K and 0.004ppm/yr. Useless to match with the others; and also there is no matching counterpart left.

If an amplifier to 10V is added, this should also be based on a divider in one case. But this would of course add 100 times higher instabilities to the 10V output compared to the naked 7.2 V output of the LTZ1000.
 
So you easily see, that the array of the Datron designs is not necessary, and used perhaps only for convenience of the PCB design.

Frank

[Blackart]

Thanks for the responses.

I was worried about the chopper noise and since Ive never used a chopper amps, I have no real experience on the actual spectral makeup ( if any ) of the left over chopper noise. Also I looked at the Maxim MAX44246 data and compared it to the LTC. My worry is that the maxim device doesn't seem to like driving capacitive loads, unless buffered by a resistor c100Ohm. Which just adds more tempo issues to deal with. Since Dr Frank raised concerns about the ref circuit susceptibility to RF instabilities I think the LTC might be the one to try.

Given that others on the forum have successfully tried the LTC1013 I think Ill try the LTC2057 If only to add some more experience the thread.

I was wondering about the 400k tempo resistor required for the LTC1000A implementation. I have not found much discussion on the threads about it apart from it being miss applied in a couple of instances. Since the data sheet notes it is for "TC compensation" - OF WHAT ? the opamps or the LTC1000 ref. i.e. should it be located anyware special i.e. near the REF can or near the opamps.. any thoughts ?

regards Lucas

Oh and with reference to Gazelles question and I agree with the others; both forums ( inc VOLT-NUTS ) are worth the read. Not only for the REF discussions but the insights into the quest.

[Andreas]

I was wondering about the 400k tempo resistor required for the LTC1000A implementation.

Hello,

According to the data sheet the 400k resistor is only necessary for the LTZ1000 and not for the LTZ1000A.
The feed back is with heater power (i.e. the LTZ serves as temperature sensor).

On my LTZ1000A I have tested the TC without the resistor.
over a 10-42 °C range there where around 1.7ppm shift (including noise and hysteresis) which corresponds well with the 0.05 ppm/K from the data sheet.

With best regards

Andreas

[branadic]

If you can live with 30K instead of 70K for the two collector resistors, you can get 5 resistors in a 16-pin ceramic LCC.  60K resistors are twice as wide, so you would have to move up to a 24-pin ceramic LCC or a 14-pin side-brazed ceramic DIP.

Have you ever tried this possibility of individual VSM85, 86, 87, 88, 89 network from Vishay? Can you tell something about the prices for a single piece? I'm looking for 5x 5kohms or 5k/20k (20k made of 4x 5k internally) network, for a gain setting application.

Edit: No need to be a network, can be an array too.

[jd]

I was wondering about the 400k tempo resistor required for the LTC1000A implementation.

Hello,

According to the data sheet the 400k resistor is only necessary for the LTZ1000 and not for the LTZ1000A.
The feed back is with heater power (i.e. the LTZ serves as temperature sensor).

On my LTZ1000A I have tested the TC without the resistor.
over a 10-42 °C range there where around 1.7ppm shift (including noise and hysteresis) which corresponds well with the 0.05 ppm/K from the data sheet.

With best regards

Andreas

Hi Andreas,

I found that you can use the resistor to fine-tune the temperature coefficient in the LTZ1000 circuit, so start with 400 K and put the circuit in a temperature controlled environment (I used a beer fridge that could heat/cool  ) Then you might change it to e.g. 470k to reduce the TC below even 0.05ppm/K.

This will also take care of any concerns about opamp TC.

The resistor itself does not have to be ultra-stable as far as I can see.

[jd]

The answer is obvious:
- very high cost for low volume production
- useful for the temperature regulator only, i.e. the 12k over 1k divider

Hi Frank,

Have you looked at LT5400? http://www.linear.com/product/LT5400 Networks of 4 resistors,  2ppm / 2k hours long term ratio stability. And not that expensive really.

A unique part as far as I have seen. On paper they look very interesting for making ratios, provided you can get lucky with the values available. But look, you can make the temperature controller divider using R4=(10+1) in series and R5=(10+1) in parallel. = 11k and 0.9090909k, equivalent to 12.1k + 1k. Perfect!

John

[jd]

Hello John - the two 70k resistors can be build with LT5400-2 in parallel and LT5400-1 in series, in theory at least but need to be tested in the field. [100k||100k]+[10k+10k] but their 8ppm/K is NOK.

Hello gazelle,

No no no, I was thinking for ratio use only! As you mention the absolute TC is mediocre and the absolute value long term stability is completely unknown (I asked them). But for ratio use it ought to be very good.

The R4:R5 ratio is little bit more tricky - you want to have a matched tracking ratio, all resistors need to be part of same device to maintain matched tracking ratio - meaning to take LT5400-1 and LT5400-4 to build your 12.1:1 ratio - that would offer 0.2ppm/K matched tracking.

Yes sorry, from memory I thought there was a single device with 2x10k and 2x1k but it is 2x100k+2x10k, absolute values too high I think.

I don't think your two-device solution gives matched tracking unfortunately.

In fact I have been working with the -6 which has 2x5k and 2x1k. This gives a ratio of 11k:1k which still works but you have to reduce the 70k collector resistor to 10k say. However my results are not so great, I got about 3ppm drift in the first 6 months or so. But there could be other causes which I am still looking into...

One possibility is the absolute values change too much, since the divider is loaded slightly by the base current this would shift the operating point. Or it could be some other blunder.

So for the four resistors R2 ... R5 in LTZ1000 ref. design can be replaced with LT5400-x network, and R1=120 ohm should remain a high quality type Vishay PG preferable with Kelvin Varley connections.

:-)

[Dr. Frank]

The answer is obvious:
- very high cost for low volume production
- useful for the temperature regulator only, i.e. the 12k over 1k divider

Hi Frank,

Have you looked at LT5400? http://www.linear.com/product/LT5400 Networks of 4 resistors,  2ppm / 2k hours long term ratio stability. And not that expensive really.

A unique part as far as I have seen. On paper they look very interesting for making ratios, provided you can get lucky with the values available. But look, you can make the temperature controller divider using R4=(10+1) in series and R5=(10+1) in parallel. = 11k and 0.9090909k, equivalent to 12.1k + 1k. Perfect!

John

What for?
My design goal was < 1ppm/yr., and I achieved that without any schmuck devices or exotic PCB slots.
Use ordinary precision wirewound resistors and solid thermal management. That's it.

Stability is now < 1ppm / 4yrs.

Frank

[branadic]

My design goal was < 1ppm/yr., and I achieved that without any schmuck devices or exotic PCB slots.
Use ordinary precision wirewound resistors and solid thermal management. That's it.

Maybe you have one of those tiny little beasts that tend to have less drift compared to the average? You can't say for sure that it is just the way your circuit is build up, because you can't make a statistical announcement to that, right? So maybe you had also a portion of luck with your reference?

[Dr. Frank]

My design goal was < 1ppm/yr., and I achieved that without any schmuck devices or exotic PCB slots.
Use ordinary precision wirewound resistors and solid thermal management. That's it.

Maybe you have one of those tiny little beasts that tend to have less drift compared to the average? You can't say for sure that it is just the way your circuit is build up, because you can't make a statistical announcement to that, right? So maybe you had also a portion of luck with your reference?

Maybe no scientifically correct statistics.... but I have 2 working references, really existing, and very stable, at least.
The third one is the modified reference in my 3458A, which also is very stable, without any Schickimicky gadgets...
And the fourth one is at "babysitter", a 1: copy of this design, which shows drift < 0.3ppm after 1/2 year.
All those really existing devices all fit within the theoretical calculations and estimations, i have done.

Everything else, I see here on the net, is theoretical stuff only, most of it even without any solid background. Only nice gadgets, without any hint or even proof, that those are useful.

Up to now, I haven't seen any real stability measurements from somebody else.. again.. I have to stress..only speculations all around.

So, you better show something practical,i.e. real circuits with real improvements over my stuff, instead of bringing up such weak arguments.

Frank