Ultra Precision Reference LTZ1000

[borghese]

You have to load the appropriate DRU file for 1 layer; for example "Multi-CB_1L_125um-Tracks_OL35um.dru". File attached, just rename from TXT to DRU
Regards

[3roomlab]

I have assembled and tested my 5 new LTZ1000 reference boards.

They all are based on Andreas schematic, providing additional EMI filtering for great immunity against RF and mains disturbances.

ive been playing with eagle, i wonder, has anyone been succesful in totally laying this entirely on 1 side? i must say i have no success.

Dr. Frank has!  https://www.eevblog.com/forum/metrology/mx-reference/msg1297126/#msg1297126

edit: actually he used two jumper wires, so *almost* entirely on one side.  I put a copy of his design on OSHPark: https://oshpark.com/shared_projects/gl2aXmlh

i know he did, i quoted his post in my question. i should re-word the Qn as : did anyone else have success in a strictly 1 layer layout w/o using jumpers?

this is my whimsical layout which i did for fun and somewhat curiosity
https://3.bp.blogspot.com/-fUTVzhfdWqY/Wjuw9s3vnXI/AAAAAAAAC30/BwFihqLtVVcEBqPCOMeiwlxS7IgqzYitQCLcBGAs/s1600/LTZ1k_11%2Bcopy.gif
i included the "RR0" which on LTZ pdf page6, is for finding an unstabilized TC compensation (but now i think it is a useless inclusion :/). the idea is to do the construction manhattan style, so no leads will be visible on the rear. the size is based on a IP67 plastic box 100x68mm. i dont suppose i have seen anybody create their favourite project box as a piece of component?

in my version, i had to use ground plane to connect capacitors (i also have to "jump" VREF to the 2nd op amp). which lead me to think, since capacitors do not really need to track temperatures and influence overall tempco, it should be possible to flip all the capacitors to the flipside of the PCB and have them out of the way of all the resistors on the front side.



*update
well i wasnt quite satisfied, and i think i managed to really fit all into 1 layer.

https://2.bp.blogspot.com/-ggDXXGuKW-A/Wjvgm3jeewI/AAAAAAAAC4I/rGwKasaRdf8qRVw13cVOsFrf3OUn19GxgCLcBGAs/s1600/LTZ1k_11B%2Bcopy.gif

[Galaxyrise]

The gain of the temperature compensation transistor reduces to about 50 when the base and collector are at the same voltage (which is where this circuit biases the transistor)-- so the high hfe listed in the data sheet is misleading. 

Isn't this the condition in which I measured 200?

[Kleinstein]

For the sensitivity to the OP it is not Hfe that matters, but the voltage gain of the transistor stage. This is about 6.5 V / (kT/e) or about 200, depending on the temperature. The early effect might reduce this gain a little, but not that much. The input side is rather low impedance and the collector side is usually much higher impedance - so the actual Hfe does not matter as long as it's in a reasonable range and not much below 1.

[TiN]

3roomlab, or cellularmitosis should make single-layer LTZ ref with large pads, so we can order metallized ceramic board to make hermetic LTZ ref .
Put some VAR resistors, get kovar tubes for feedthru copper pins and seal it all in vacuum or argon filled tube. 

Likely there would be no difference in performance, but it would look neat. Maybe toss an orange LED or two on the base to give it warm fuzzy glow too  .

[beanflying]

3roomlab, or cellularmitosis should make single-layer LTZ ref with large pads, so we can order metallized ceramic board to make hermetic LTZ ref .
Put some VAR resistors, get kovar tubes for feedthru copper pins and seal it all in vacuum or argon filled tube. 

Likely there would be no difference in performance, but it would look neat. Maybe toss an orange LED or two on the base to give it warm fuzzy glow too  .

I would prefer a 5 digit Nixie voltmeter on the front please 

[branadic]

Prema Semiconductor used a slightly different circuit to power the LTZ1000 from 12V in Prema 6048. I found two circuit diagrams with slighlty different resistor values in two different manuals. Would be interesting to see what they finally assembled.

-branadic-

[3roomlab]

i dont know if my eyes are playing tricks ... but what does the red resistor do?

[floobydust]

Not an expert here, but in ref PCB layouts, there's two grounds- the high-current heater return and the reference (-).
They are separated to minimize any heater-induced I2R voltage drop across ground from shifting the 0V reference point and causing uV offset error.

Your PREMA schematic just seems to use one symbol though, as some CADD software cannot understand two ground (nets) that are connected together at the PSU central grounding point. I don't have a 6048 or 6047 PCB to look at both sides.

The 10R resistor between them is usually to avoid big trouble if one ground is open circuited, and to lower Q for any RF picked up between the two ground traces.

[3roomlab]

Not an expert here, but in ref PCB layouts, there's two grounds- the high-current heater return and the reference (-).
They are separated to minimize any heater-induced I2R voltage drop across ground from shifting the 0V reference point and causing uV offset error.

Your PREMA schematic just seems to use one symbol though, as some CADD software cannot understand two ground (nets) that are connected together at the PSU central grounding point. I don't have a 6048 or 6047 PCB to look at both sides.

The 10R resistor between them is usually to avoid big trouble if one ground is open circuited, and to lower Q for any RF picked up between the two ground traces.

i could not get my head around it. so i dump something into FEMM 4.2 to visualize it
pic1 describes the approx currents and the 10R between the 2 sides, each with its own 0v (cyan)
pic2 is the same currents, but all traces are approx 6mil (no 10R)
pic3 same again, left side is approx 200mil trace, right is 6mil trace (no 10R)
pic4 is 6mil only 1 GND (no heater GND)

the simulation idea behind 6mil and 200mil trace comes from saturn PCB
6mil 1 oz PCB 1 inch long loaded @ 0.1mA returns a potential diff of about 0.011v
230mil 1oz PCB 1 inch long loaded @ 5mA returns a potential diff of about 0.011v (2oz PCB will only need about 150mil wide)
esp for VREF end, i wonder how much of these traces should be balanced between sourcing and draining. it would be a crazy matching exercise, this reminds me of that zero ohm thread, trying to find the zero current point inside a thick copper piece.

i also assume that the GND return current posing the most problem is the heater as the current varies with changing temperature.

but since swapping these around in simulation for GND, it appears as long as all returns appear at exactly the same star point, it doesnt seem to need a 2nd heater return? what did i miss? (assume the left side GND is the master)
maybe the better question is, since pin 3 potential is referenced to the GND thru R1, how could a seperated heater return create a more stable GND point since it is the only GND? if GND shifts, wouldnt VREF also shift and result in no change? (i assume a low noise LDO is on board and the V supply is always "constant")

i went a little further in FEMM 4.2 simulation. the animated gifs alternate between 2 diff current returns. however the more i look at these, i seem to confuse myself more

assuming zero volts is the black hole, it prevents a current return disturbance on the right from affecting the 1 on the left.


i need to emphasize, i am assuming by simulations.

this is with/without the 10R linker. i am unsure what to make of it.


i can assume since all GND labels are not really zero, what does come close to a full negation? huge capacitor as sinker?


(the txt file needs to be renamed *.FEC for use in FEMM 4.2)

[floobydust]

Colourful sims, what I expect for showing the electric field in copper PCB traces. Learned about the "river mouth" being something to consider.

My theory for the Prema 10R resistor - it does nothing at DC as the PCB trace resistance is comparatively low, unless for open-circuit protection on the Kelvin zener (-).
Or, it's to fool the CAD software. But at high frequencies, RF, the twin trace inductance is now at play and I suspect that another reason for it.

Copy pasta from http://www.electronicdesign.com/analog/what-s-all-ltz1000-stuff-anyway:

"Some folks put a dummy resistor in the schematic to separate those grounds into different nets. However, Mark Thoren, staff scientist at Linear Technology says, "Dummy resistors separating 'grounds' are a recipe for disaster, and I see disasters related to this practice all the time. One of the first things to look for when reviewing a customer's schematics is multiple ground symbols without a good reason for them."

'Do be aware of the return currents causing drift problems. [Carl] Nelson comments, “Current out of pin 7 is about 6.5 V/R2 + 6.5 V/R3 = 186 uA. Current out of R5 is about 550 uA.  Current from R1 (same as the Zener current) is about 4 mA.  These currents should be routed separately to the low side of the reference output. If a single trace is used, and the trace resistance is 0.4R , voltage drop in the trace will be 1.9 mV. Copper has a thermal coefficient of 0.0038/°C, which generates 7 uV/°C in the trace, equal to about 1 ppm/°C in the reference.  This is about 20 times higher than the expected drift of the part’s design intent.”

[guenthert]

Copper has a thermal coefficient of 0.0038/°C, which generates 7 uV/°C in the trace, equal to about 1 ppm/°C in the reference.  This is about 20 times higher than the expected drift of the part’s design intent.”

There he lost me.  What "thermal coefficient" is meant?

[Andreas]

If a single trace is used, and the trace resistance is 0.4R

How long are your traces?
I usually use minimum 24 mils which have 0.9R per meter with 1 oz copper.
Usually you should use Kelvin sensing for the zener voltage.
Within the resistor loops you have at least a reduction factor of 100 for the resistance changes.

There he lost me.  What "thermal coefficient" is meant?

0.385%/K is the thermal resistance change of pure metals like copper, platinum etc.

with best regards

Andreas

[TiN]

Happy New 2018 to all voltnuts. Wish you all guys short LTZ lead-times and smooth stable voltages.
Little present on my LTZ data. I had my references turned off for few weeks, and just in mid-December I put them all back in new box and powered backup.
Since then all 8 references are scanned non-stop using HP 3458A and Keithley 7168 nV scanner card.

So far:



Modules:

Channel 1: KX 7V LTZ1000 reference.
Channel 2: KX 7V LTZ1000 reference.
Channel 3: KX 7V LTZ1000 reference.
Channel 4: HP 3458A STD A9 reference
Channel 5: xDevs.com FX 10V reference prototype.
Channel 6: KX 7V LTZ1000 reference prototype.
Channel 7: KX 7V LTZ1000 reference prototype.
Channel 8: xDevs.com FX 10V reference prototype.

P.S. Over last year I got 8M of pageviews, which is amazing to me, given the narrow niche of my projects. 

[Kleinstein]

Looks like the 3458 got an extra 0.2 to 0.3 ppm shift by Christmas. At least all the curves seem to shift in parallel - so the more likely cause would be the meter.

[mimmus78]

Looks like the 3458 got an extra 0.2 to 0.3 ppm shift by Christmas. At least all the curves seem to shift in parallel - so the more likely cause would be the meter.

I think the meter shift seems to track temperature ... It's hard to say for me with all this faded colored dots. I'm daltonic, so it's even harder for me ...

This almost demonstrate (what Frank keeps saying) that at lowerish temperature there is not zener histeresys.

The 3458a was always on during this time?

[TiN]

I think that was shift caused by ACAL DCV, triggered by temperature change.
The meter and setup was moved 2 days to another room prior the log start.

Room is freely moving ambient, no airconditioning, zero airflow around the references box.

[Kleinstein]

ACAL causing the shift sounds very plausible.

The different refs seem to show quite different initial drift from recovering from the time off. It might be interesting if this somehow correlates with temperature setting or A and non A versions.

[Dr. Frank]

I think that was shift caused by ACAL DCV, triggered by temperature change.
The meter and setup was moved 2 days to another room prior the log start.

Room is freely moving ambient, no airconditioning, zero airflow around the references box.

Happy New Year, Illya!

Very interesting experiment..

Could you please assign the oven temperatures of the 8 different references, and maybe their initial reference values?
 
Especially CH2 and CH3 need over a week or so to stabilize from a ~2ppm change..
I assume, these references which need long time to stabilize, run at higher oven temperatures?

After a power interruption, even after several days, all of my 8 LTZ1000 references, which run at 45..55 and at 60°C (inside my 3458A), stabilize to their initial value within < 0.2ppm after a few hours, latest, and stay there.

I'm just sending a friendly volt-nuts colleague one of my LTZ references in cold condition, and I'm really curious about the round-trip comparisons of the reference voltage, at first after a few days at his location, and then on return in my lab.
This round-trip will collect possible shifts due to vibrations also, like my 3458A encountered on the last volt-nuts gathering two years ago.

Frank

[mimmus78]

I think that was shift caused by ACAL DCV, triggered by temperature change.
The meter and setup was moved 2 days to another room prior the log start.

Room is freely moving ambient, no airconditioning, zero airflow around the references box.

Happy New Year, Illya!

Very interesting experiment..

Could you please assign the oven temperatures of the 8 different references, and maybe their initial reference values?
 
Especially CH2 and CH3 need over a week or so to stabilize from a ~2ppm change..
I assume, these references which need long time to stabilize, run at higher oven temperatures?

After a power interruption, even after several days, all of my 8 LTZ1000 references, which run at 45..55 and at 60°C (inside my 3458A), stabilize to their initial value within < 0.2ppm after a few hours, latest, and stay there.

I'm just sending a friendly volt-nuts colleague one of my LTZ references in cold condition, and I'm really curious about the round-trip comparisons of the reference voltage, at first after a few days at his location, and then on return in my lab.
This round-trip will collect possible shifts due to vibrations also, like my 3458A encountered on the last volt-nuts gathering two years ago.

Frank

Also resistors type could be interesting ...

[Andreas]

This round-trip will collect possible shifts due to vibrations also, like my 3458A encountered on the last volt-nuts gathering two years ago.

Ups,
now I´m feeling guilty:
https://www.eevblog.com/forum/metrology/lm399-based-10-v-reference/msg840100/#msg840100



So we should do the next meeting at your location?

with best regards

Andreas

[martinr33]

On TiN's fantastic - and live - data logging set!

We're seeing a lot about the meter here, at these levels of resolution. I think we need a calculation to take out the temperature variance of the 3458a. It looks like it is around 0.5ppm per degree K (that's just from eyeballing the drop on December 24). I think that this is more significant than the drift on at least some of the references. Note how much larger it is for the 10V references over the 7.xV references.

As the chart stands, references that have a temperature coefficient that matches the 3458a look rather better than they should.

Once the temperature coefficient is established, it is probably better not to do autocals.

There's also the question of drift in the 3458a ADC.  I know that TiN has looked at this, but for long-term logging, this is also a factor.

Here's a quick correlation analysis, checking each reference against the recorded temperature:

ch1   ch2           ch3           ch4           ch5           ch6           ch7   ch8          k16v   k10v
59%   -84%   -81%   -83%   -70%   -23%   91%   -65%   2%   99%

Channel 7 most closely tracks the temperature of the 3458a, assuming a positive temperature coefficient for the meter. If the meter has a negative tempco, then references 2,3,4 are closer. Therefore, if the 3458a was at a constant temperature, one of these three may well be the best reference from a stability point of view. That is, of course, assuming no material drift in the meter.

Put another way - we are looking for the "golden reference" which is stable over both  temperature and  time. Even more golden would be stability across the dimension of space - which would make it easy to ship.

My experience is that the 732a is a really good example of such a reference. Probably why it is still around. Although it is about 100x heavier than these references, and really does not like going dark.

[branadic]

So we should do the next meeting at your location?

with best regards

Andreas

I'm in with a calibrated Keithley 2002, our 3458A, our calibrated Prema 5017 SC, hopefully with my till then calibrated private Prema 5017 SC and some LTZ1000 and LM399 based references. 

The nice thing is with Andreas PC software for 3458A, the GPIB adapter I can bring along and the ambient sensor module we are able to monitore live data together with ambient properties (temperature, humidity, dew point and pressure).

-branadic-

[Dr. Frank]

Ups,
now I´m feeling guilty:

So we should do the next meeting at your location?

with best regards

Andreas

Hello Andreas,

no need at all to feel guilty.. The 3458A is not a standard!

This shift was an interesting experiment.. and covered by the other 3 references @ home..
For the next time, I need an algorithm for detecting and valuating such drifts and drops inside such a reference group for proper calculation of the expected mean value..

Didn't somebody mention recently, that calibrated 3458As encounter a shift of up to 5ppm, just by transportation?
So I wonder, how transporting instruments like these might work at all.

In the end, it was a really nice meeting, and I propose to repeat it this year, maybe here, near Frankfurt where 2..3 other well-known volt-nuts from this area may also be willing to participate.

Frank

PS: Two references are just on their way to a round trip via dhl, so I'm very curious, how their measured values can be reproduced afterwards.

[TiN]

Dr.Frank,
I will take bank apart once I get back Taiwan next week, and will list all configurations (opamps, resistors, LTZ type) per each channel. Two of KX modules are with AUGAT socket for zener. 

martinr33, my 3458As have no measureable tempco, but before you go into gossip theory, do not forget that all references are connected to meter thru scanner card, NOT directly. And I did not measure impact of the scanner JFETs on tempco/drift yet Also all modules, except A9 had thermal stress at start from wiring connections (not using any connectors at KX outputs, only direct solder copper twisted pairs between scanner and board). FX 10V modules had zero aging at all, with fresh LTZ chips.

I periodically was checking drift and stability of the meter, it's still accurate and stable. In fact, just before leaving to vacation, I had calibrated two KX LTZ1000CH's refs from CalMachine and shipped to him.

Assigned value: KX1 module (large input power binding posts): 7.141467 VDC , https://xdevs.com/ckx1_test5_tc/
Assigned value: KX2 module (smaller input power binding posts): 7.157411 VDC , https://xdevs.com/ckx2_test10_tc/ 

These modules arrived to CalMachine, but since he's too lazy shy busy, he just got few hours on each module, using his calibrated (i think <30 days ago) 3458A. Here is his data:

CSV-data.

Plots:





So output deviation fits my expectation of 2ppm DCV at my lab accuracy, with good margin:

KX1 module 7.157411 VDC (xDevs lab) vs 7.157415 VDC (CalMachine lab), +0.56 ppm off.
KX2 module 7.141467 VDC (xDevs lab) vs 7.141472 VDC (CalMachine lab), +0.70 ppm off.

Not that bad, given international air shipping in winter and different calibration source (my 3458As calibrated vs 732B (cal August 2016 by Tektronix USA) in January/February 2017). CalMachine's meter calibrated by Keysight. Also CalMachine's temp is lower, I calibrated values at +24C, not 20. So tempco may contribute another 0.2ppm error.