[FT] xDevs.com KX LTZ1000A fairy-tale, or the story of little jumper.

[mimmus78]

O men two weeks is the definition of "this will take forever" ... I think with slow ramp up you can go with just 24h.
Lucky you that have the scanner card ... I already finished my budget for the next couples of years so I can measure only one at time.

[TiN]

No scanner card used in this case. Three separate meters are the jedi way. Need to get moar 8.5-digit meters, so I can test 4 resistors together!   
I tried faster ramps, but it's just too small margins to tell what change from what.

[Kleinstein]

If it just for the temperature effect one could go faster and maybe also do a continuous ramp instead of the step like profile. However it is not only temperature that effects the resistors, but also indirect effects, like expansion / shrinking of the PCB which can be in part due to humidity effects. So it is not easy to get a pure temperature effect. Also Temperature gradients can have an influence - this is likely much of the overshoot seen on temperature steps.

A scanner will add some errors - so for highest precision it is better to use the DMM directly. It somewhat depends on the required resolution and noise: one might get away with a good scanner. At least with the slow temperature profiles noise is not that critical - so it is only a possible longer term error (e.g. thermal EMF of the scanner) that is a really problem. On he other side a scanner could reduce some of the thermal EMF errors by reversing the resistor.

[TiN]

Since most of my little modules PCBs happen to have same width, I decided to make an hermetical service enclosure for additional thermal testing.
We talked about this wild idea few pages ago, so this will also be used for that.



It's simple copper tube 42mm in diameter, 130mm length with caps. Material of choice is copper, so I can solder it to make gas and liquid tight. This is required to allow submersion into LN2 dewar, while keeping cryogenic liquid out of direct contact with PCB module inside the probe.



Inner tube wall has two slots to support and hold PCB, which just slides in. Caps are easy to solder to the tube, longer copper thin tube is 10mm OD, to have power, signal wiring and cabling escape the dewar and probe PCB.



I'll also test low-noise preamps in this chamber once they ready, to see how temperature affects the noise figures.

[ManateeMafia]

Very nice TiN. Looking forward to some real "low thermal" copper put to the test.

Any plans on controlling condensation?

[TiN]

It's easy to pressurize inner volume with nitrogen. Just after nice few hours prior bake with everything inside the probe.

First test would be just blank PCB with thermal sensors to measure curves and temperature change speed. Perhaps by using smaller PCB with bit of insulation and internal heater I could also adjust temperature in wider range.

[David Hess]

I have used liquid nitrogen cooled environmental chambers before with the board exposed without any protection.  Once my boss came in while I was doing a cooling test to destruction and asked how cold it was.  "I do not know because the thermometer only reads down to -199F but since the liquid nitrogen is pooling inside the chamber and the board is sitting in it, I assume -320F."  The precision amplifier and comparator circuit never failed and continued to meet its specifications.

Aluminum electrolytic capacitors (and I assume wet tantalum capacitors but I never tested them) obviously do not function at such a low temperature but the design recognized this and used ceramic and solid tantalum capacitors which had no problem.  None of the common plastic or glass packaged surface mount parts I was using had any issues but I expect that thermal cycling would have eventually broken something.  The total error of the LT1013 class operational amplifiers I was using increased but not in any unexpected way.

For a hermetically sealed design using copper and tin based solder, how quickly does tin pest become a problem?  Is this an application where an antimony or bismuth bearing solder should be used?

[mimmus78]

So I managed to get my own way of measuring TC of resistors working in a reliable and automatic fashion.

I use what I consider an easy to build but "sophisticated" method to heat the resistor that leave almost
just few seconds of lag from the temperature measurement point and the R variation. As you can see on
the second chart TC curve is almost a straight line  ... this should means temp. and res. variation track very
closely. I still have to check some parameters to prove this method is deftly OK and than I can describe
how I did it.

Anyway apart from heater, the measurement are done with a K2001 in 4W configuration with offset
compensation to avoid thermal EMF to upset readings.

My thermal cycle consists in heating from room temperature to room temperature + 5°K and stay there for
some minutes, than increase other 30°K (and stay at this point for more minutes) and than back to initial
room temperature + 5°K (and rest there for more other minutes). Rump up and down are made almost as
fast as I can. The rump up is 1°K per minute circa and rump down is a little slower than this.

All measurements where done with stable/stabilised room temperature with a max variation of 0.5° C.

I started checking some UPW50 few days ago and I get

• 10.005,05 @ 26°C
• 10.006,02 @ 56°C

that is something like ~ 3,2 ppm/K that is also right in the typical specs of the resistors.

The strange thing is that I found something physical is amplifying the TC effect, I observed that after heating
the resistance value always "overshoot" and after cooling down it undershoot. I got more 0.1ppm/°K of
difference if this parameter is measured before or after the rest period (maybe also 0.3 ppm with some
resistors). 0.1 ppm/°K is not much if you are matching 3ppm/°K resistor for the LTZ1000 divider, but I'm still
curious what is it. Anyway if you find a 1 ppm/°K resistor this 0.1 pmm/°K became very important and can
cause significant error in TC measurement (I think this effect will be reduce with low TC resistors but this is
still to be proved).

So the TC of this resistors seems to be a combination fo TC of the wire and TC of some physical stress
applied to the bobbin during heat and cool down cycle with the latter requiring more time to settle down
than just plain TC of EVANOHM.
 
I read Andreas thread about TC of resistor, but I cannot find any measurement where the DUT was leaved
at the new temperature to rest for some hour. Because of this I cannot tell if this behaviour compare the the
chart Andreas proposed in his thread or not.

I found also many of this resistor seems to go down in value after many thermal cycles. This last went down of
20 ppm, but this drift can also be caused my meter that is not stable. I have to remember to check multimeter
offset reading against my calibrator (that has some VHP resistor inside as reference) at beginning and end
of measurement to be sure shift is of the DUT and not by ME.

Any comment is appreciated.

Domenico

PS: just noticed on second chart that measurements gets noisier at the end of heating a cooling process
... so it can also be my measurement technique that is wrong.

PPS: just realised that at the end of the heating/cooling cycle there are a big number of points because
temperature variation is more slower than in the middle of the chart ... so  if we use temperature as X axis
it's normal that it appears to be more noisier at the end of each cycle.

[Andreas]

Hello,

pictures please.

And I think that the "warm drift" is due to some relaxation effect (or humidity effect) of the epoxy.

With best regards

Andreas

[mimmus78]

-- Andreas --
initially I was thinking too to humidity. But I run more than 20 cycles on this resistor and this relaxation effect at begin and end of the cycle is almost exactly the same day after day, from the first cycle to the last one. Unless it's a sponge I guess it cannot be so sensitive to humidity.

-- as for my setup --
it will run all the night to trace the next resistor so I'll try to take some photos tomorrow evening when I will exchange the resistor under test. I hope to have some time to describe why this elementary setup works so well and make a post ... this will be good if you would like to reproduce the same process. In any case, if I will not find the time to do the post, I will just post the photos on your post.

I have some initial conclusions from my initial TC exploration of this UPW50. Almost any value I tested is +2.2ppm to +3.2ppm except for the unfortunate case of the 1K resistor that are all -1ppm.
Still to be determined if they drift apart ... I will check my multimeter in a few days and with all data collected I should be able to track any drift  ...


At the end if I use this three resistors:

• 2K (+3ppm)
• 10K (+3.1ppm)
• 1K (-1ppm)

for the ts divider of the LTZ1000 I get a total error of just a little bit more than 3.2ppm/°K.

This value is than attenuated by a 75 times factor to the Vzener and it become less than 0.05 ppm/°K. This value should be not too much but it's still something to consider in the hole error budget.

I ordered another 1K UPW50 resistor from another supplier in the hope to get one with positive coefficient (it should be here tomorrow).
I think in any case I will assemble the KX reference this week end with what I have ... than maybe I will exchange this 1K resistor in future when I will have one with some positive coefficient.

[mimmus78]

Andreas: I think I lost from the LTZ1000 thread why the LT2057 was discarded at the end by you, may you just repeat the reason?

[Andreas]

Hello,

my UPW50 were all in the range of +0.6 .. +4.8 ppm
(so with a relative large stray from device to device).

https://www.eevblog.com/forum/metrology/t-c-measurements-on-precision-resistors/msg503546/#msg503546

with best regards

Andreas

[Andreas]

Andreas: I think I lost from the LTZ1000 thread why the LT2057 was discarded at the end by you, may you just repeat the reason?

Hello,

It wasn´t me.
I have 2 references running with LTC2057 (#5 + #6) against 2 references with LT1013A in CERDIP8 (#3 + #4).
Up to now no winner. The lowest ageing drift seems to be on #4 + #6. (with same datecode of LTZ1000A)

If I look closer to the data:
the devices with LTC2057 can be operated at lower voltages.
and have (by accident or systematically?) lower T.C. (before adjustment with R9).

with best regards

Andreas

[mimmus78]

Sometimes (often?) the hysteresis is larger when having a low TC within a resistor type.

I added another sample but I think you get the point ... actually I measured hysteresis in the first cycle is 33% to 50% of TC.

Maybe we need to change test procedure:

 - What happens if we use only 10°K as thermal cycle with max 1°K per hour of dt?
 - Does hysteresis effect is reduced with lower and slower thermal cycle?
 - If we use fast and slow thermal profile, does resistor settle down to the same value after 24h resting?

To test latter point I was thinking to do:

   - move from t.amb to t.amb + 8 with 1°K x hour
   - rest there for 2 hours
   - move from t.amb + 8 to t.amb + 4 with 1°K x hour
   - rest there for 3 hours
   - move to t.amb + 40 with more than 0.5°K x minute
   - rest there for 1 hour
   - move back to t.amb + 4 with more than 0.5°K x minute and rest there for many hours and check what happens

[TiN]

Alright, long story comes to and end, module was shipped and on it's way to the end user

Area around reference chip filled with isolator and enclosed into metal cage for shielding/air draft protection.



Similar on the bottom side.



Final assigned value: +7.1637446 VDC, +/-0.6ppm at +24.0 °C ambient.
Pre-shipment datalog capture proven stability within 0.5ppm over 5°C ambient temperature variation.

Hope we hear from new happy owner in few months, once module make it's way into Belarus 

[mimmus78]

Not nice ... but sure enough stable.
Just realise my thermal chamber used for resistors TC with its 3 cubic centimetres volume is too small for testing this modules :-)

PS: a 3458A is near to begin another journey to his new destination

[Kleinstein]

A not so nice professional look could be an advantage, when it comes to customs.

For the hysteresis, there are several possible effects to cause is. Some of them, like uneven thermal expansion might be influenced by temperature gradients and thus by the speed of temperature changes. But generally I would not expect a big influence of the heating rate. The rate of cooling from high temperature might have an influence, as relaxation processes might get increasingly slow on cooling. So once you are something like 40 K below the last temperature where the system was in internal equilibrium, it may take very long to catch up and may not reach that state even after days of waiting. Fast cooling from high temperature can make a difference - I know it does to the properties of epoxies, when cooling from above 100 C.

[mimmus78]

For the hysteresis, there are several possible effects to cause is. Some of them, like uneven thermal expansion might be influenced by temperature gradients and thus by the speed of temperature changes. But generally I would not expect a big influence of the heating rate.

I started another very slow test this night 1°K per hour slope and it seems I'm getting completely different values. I too didn't expected so much difference ... let see tomorrow what it turn out after temperature is 10°K up. Actually with 5° K delta is only 0.2ppm/°K VS 1ppm/°K I got yesterday I cannot explain all this difference. Also Edwin explained many times that fast or slow temperature delta should not make much difference with a good resistor.

[dr.diesel]

Look what showed up today!   

Digikey stuff has been here for a while now, just waiting on resistors.

Thanks again to mimmus78 and TiN

[TiN]

And finally I got around and updated Gerber files to revision B03. Nothing serious, just widen some gaps, thicken few traces and uodated silks/masks and bit of plane pours for more lax DFM. Gerber files hosted the same place as before.

No plans to order/build boards yet though, I need catch up on few other things and build reliable noise measurement setup.

[mimmus78]

The slow rump up give me exactly the same TC results in the 26-56°C temperature points as the fast rump up so my result are as we all expected.
What was different was that the slow rump up give you the possibility to better appreciate and measure TC with better accuracy at every °K.

The fist 10°K the TC was of -0,38ppm/°K and than it accelerated step by step to -1.6ppm when the temperature was approaching the 56°C.
This has fooled me yesterday in thinking I had completely different data because I made preliminary estimation of TC only on the first 10°K.

I will continue to do experiments ... I'm still not convinced what are those resistance recovery that happens when after a thermal cycle we
leave the DUT rest for many hours at the new temperature. I suspect those recovery are physical compression/relaxation of the resistor
itself and that they also cause the resistor to drift.

[TiN]

I've heard member SvanGool got his boards manufactured, using B03 latest Gerber.

He shared this photo with me:



Guard ring bit overly thin, but that shouldn't affect anything much.

Btw, anyone get references based off mimmus78 PCBs going?

Happy holidays everyone! 

[SvanGool]

You can find TiNs KX-ref PCB Version B03 here: https://oshpark.com/projects/HfKcqjV3 at OSH Park (with TiNs permission).
They are still UNTESTED.

These are the standard 4-layer boards with ENIG(gold)-finish.
The minimum order at OSH Park is three boards, these cost in total $49.80 ($16.60 per board), including standard (international) shipping, excluding any potential taxes or other costs.

[mimmus78]

TiN many buddies received boards, but nobody still built it in public. I'm going to build first sample soon. I received Edwin resistors yesterday (3 sets) ... as soon as I find 2 dammed resistors for the temperature set divider with same tempco I will start.

Inviato dal mio Nexus 6P utilizzando Tapatalk

[mimmus78]

Ahh forget to order couple of caps and the TC resistor ... and mounted second op amp reversed!



Inviato dal mio Nexus 6P utilizzando Tapatalk