PCR versus TCR

[Zeranin]

I need to read everything again to be sure I precisely understand what's going on. That said, if my understanding of the 'experimental' setup is correct and the drifts are happening when I think they are I have a suspicion that the problem is caused by creep in the bonding layer. That is, you hit it with a 16A square wave, the shunt heats up, the aluminium heats up and sets up a stress on the bonding layer because of differential expansion of the shunt and aluminium. Then the bonding layer slowly relaxes giving the observed shift as the stress in the shunt changes with movement of the binding layer. Very much a first guess but I think it's consistent with the observed behaviour, if I've understood the observed behaviour properly.

I understand what you are saying, and it’s a good suggestion. However, the electrically insulating layer that bonds the zeranin to the aluminium substrate is organic and would soften more at higher temperature, so we would expect any such creep would be faster and further at 47 DegC compared to 20 DegC, and yet no difference is observed. I would also need to be convinced that such a theory matches the observed direction of the drift, and approximately predicts the correct magnitude. Keep in mind also that the zeranin temperature rise is only 0.25K, so the induced strain would be very small, and just as well. I’ll bet creep effects would start to happen for temperature swings of ten’s of degrees. This is certainly not the way to build a metrology style resistor, but most metrology style resistors won’t cut the mustard here because the significant self-heating at 16A would produce an unacceptable temperature rise. Whatever else you can say about this zeranin resistor, you won't find another resistor whose temperature rises by only 0.25K on application of 25.6 Watts, ie, a thermal resistance from zeranin to backing plate of 0.01 K/W.

[Zeranin]

... a possible solution would be to use a heatsink material with the same thermal expansion coefficient as Zeranin (18ppm/C). Silver looks like a perfect match with 18ppm/C and even plain copper (19ppm/C) looks much better than aluminium with 23ppm/C.
Cheers
Alex

I think you have missed the point here, and my own suspicion is that using a heatsink material with the same thermal expansion coefficient as Zeranin will not help. If the heatsink and zeranin change temperature in unison, as when the R-T curve is measured, then difference in expansion coefficient will indeed modify the manufacturer R-T curve that was taken with naked zeranin. However, the R-T curve that I measured is with the zeranin already bonded to the aluminium substrate, and I use this actual R-T curve to select an operating temperature (~34 DegC) where the actual dR/dT is almost zero. The effect of the different thermal expansions is simply to modify the R-T curve, and shift the position of the sweet spot where dR/dT=0.

It is my suspicion, and some others have said this as well, that what causes the unexpected resistance drift is the temperature rise of the zeranin with respect to the constant-temperature substrate, and the stresses and strains that this creates are not related to the difference in expansion coefficients. This temperature rise is tiny, just 0.25K, so one could be forgiven for thinking it should have a negligible effect. The thread may still have some way to run to fully explain what is going on.

[Alex Nikitin]

Another suggestion: as the dissipation in the shunt goes from 0 to maximum and the temperature rises by 0.25K, it creates a shift in the temperature distribution in  the heatsink and a shift in the temperature regulation loop. Also 0.25K makes for ~5ppm linear extension in Zeranin...

Cheers

Alex

[Zeranin]

Another suggestion: as the dissipation in the shunt goes from 0 to maximum and the temperature rises by 0.25K, it creates a shift in the temperature distribution in  the heatsink and a shift in the temperature regulation loop. Also 0.25K makes for ~5ppm linear extension in Zeranin...

Cheers

Alex

A fair suggestion. However, the x12 Peltier modules sitting on the opposite side of that 12mm thick constant temperature aluminium heatsinking plate are arranged in a grid that nicely and uniformly covers the area of the zeranin shunt, deliberately designed that way to ensure an even temperature across the entire area of the shunt. To further iron out temperature differentials, there is the 12mm thick aluminium plate itself, plus that 12mm thick copper plate on the other side of the zeranin resistor, used to clamp it down to the constant temperature aluminium plate, so all in all we can be confident that the entire zeranin resistor is damned close to being at a single, uniform temperature.

You are correct that 0.25K makes for 5ppm linear expansion in zeranin ..... an astute observation, but as yet, only hand waving. You will need to stitch that into a full, detailed explanation of what is going on.

[amspire]

To further iron out temperature differentials, there is the 12mm thick aluminium plate itself, plus that 12mm thick copper plate on the other side of the zeranin resistor, used to clamp it down to the constant temperature aluminium plate, so all in all we can be confident that the entire zeranin resistor is damned close to being at a single, uniform temperature.

Is there a layer between the copper layer and the Zeranin to allow for the different expansion rate of copper to aluminium? Copper has a lower expansion rate then aluminium, so it may be compressing the top of the Zeranin which would lower the resistance. Perhaps if you can try different top plates, you can get one to give the best coefficient. Zinc, for example, is a higher expansion rate then aluminium so it should work towards a positive temperature coefficient.

[Alex Nikitin]

To further iron out temperature differentials, there is the 12mm thick aluminium plate itself, plus that 12mm thick copper plate on the other side of the zeranin resistor, used to clamp it down to the constant temperature aluminium plate, so all in all we can be confident that the entire zeranin resistor is damned close to being at a single, uniform temperature.

Is there a layer between the copper layer and the Zeranin to allow for the different expansion rate of copper to aluminium? Copper has a lower expansion rate then aluminium, so it may be compressing the top of the Zeranin which would lower the resistance. Perhaps if you can try different top plates, you can get one to give the best coefficient. Zinc, for example, is a higher expansion rate then aluminium so it should work towards a positive temperature coefficient.

It is certainly a good question why different materials are used on two sides of the shunt. Why not copper on both sides?!

Cheers

Alex

[d-smes]

Is there a layer between the copper layer and the Zeranin to allow for the different expansion rate of copper to aluminium? Copper has a lower expansion rate then aluminium, so it may be compressing the top of the Zeranin which would lower the resistance. ...

There must be something there or the Copper would electrically shunt the Zeranin!   Regardless, I like this theory.  If the Copper is bolted to the Aluminum at the corners, then the assembly would bow as it heats- The Al expands more and faster than the Cu both because of the higher expansion rate and because the Al is a fraction of a degree warmer than the Cu.  With the Al now being slightly longer than the Cu between clamping points, the assembly curls/bows (long dimension is no longer straight).

Having just written this, resistance shift due to the bending should have showed up in the R-T test.  Maybe it did which is why the clamped R-T didn't match the Zeranin in free space R-T.  So we're back to the internal heating of the shunt with current as causing some additional stress / strain forces that causes the shift.

I like Alex's suggestion of using Copper on both sides; especially since the Cu has a closer TCE to Zeranin.  In addition, I'd suggest cooling both sides symmetrically and use the same 0.07mm thick electrically insulating film on the clamping side.  If that doesn't help, maybe you need to invest in another L+N shunt and use that in your machine (or attempt to copy how it is made).

[Zeranin]

Here is a more precise description of how the beast is built. There are 3 major components.

(a) The 12mm thick temperature-controlled aluminium heatsinking plate, with an array of Peltier modules on the underside. This plate is 350 x 230mm, which is wider than the 100mm wide resistor. This plate doubles as the base of the temperature controlled box that houses the electronics for the current driver, which is why it is 230mm wide. This arrangement produces a temperature controlled box for the electronics 'for free'. I don't have easy access to 12mm thick copper plate of this dimension, so was made of aluminium. The current driver will soon need to be delivered to the customer, so there is not time to rebuild this plate in copper, and in any event, I would hesitate to embark on such a major and expensive design change unless firmly convinced it would make a difference.

(b) The zeranin resistor assembly. This consists of the etched ziz-zag patern of zeranin sheet, bonded to a 1.6mm thick sheet of aluminium (AKA the substrate), of dimension 250 x 100mm. The 0.07mm thick heat-bonded film provides electrical isolation b/w the zeranin and aluminium.

(c) The 350 x 100 x 12mm thick copper clamping plate.

The object is to minimize self-heating induced temperature changes of the zeranin. Therefore it was decided that the zeranin (rather than the aluminium substrate) would be in contact with the constant-temperature heat-sinking plate, separated of course by a 0.06mm thick electrically insulating PVC plastic film, with small quantity of thermal grease. Thus, the zeranin sees aluminium on both sides. The copper clamping plate is in contact with the aluminium substrate, and x8 of M4 screws pull the clamping plate down onto the constant-temperature heatsinking plate, with the zeranin resistor assembly sandwiched between. The screws pass through the zeranin resistor assembly, with mating 5mm clearance holes etched in the zeranin sheet, and subsequently drilled through the substrate.   

At this stage, it is not practical to change the metals that have been used. I originally requested that the substrate be made of copper, but that was not possible, as the etching process would have removed the copper, as well as the zeranin!

This project has a long history. It was originally designed to use a Powertron FHR4-80370 shunt resistor, but this was found to be unsatisfactory, with a drift on the leading edge of the 16A square wave of almost 15ppm. We had a custom version of this resistor made at great expense from Zeranin, yet found it performed no better than the cheap bog-standard foil, again reinforcing the point that the R-T curve has nothing to do with the drift that we observe. By this time I was sure in my mind that the problem was due to the temperature rise of the zeranin with respect to the substrate, so designed a 'super-sized' version of the 80370, with almost 3 times the surface area of zeranin. The thermal resistance from zeranin to substrate scales as the surface area, so this new resistor with x3 surface area results in a x3 reduction in the temperature rise of the zeranin. Sure enough, the new resistor does perform x3 better, with a drift of only 5ppm vs the original 15ppm. Further improvement could be made by scaling up even larger, but the existing super-sized resistor is already at the limits of the equipment used to manufacture it, and I would also have to completely redesign and rebuild the heatsinking plate and peltiers.

Further improvement could be made by decreasing the thermal resistance of the 0.06mm thick PVC plastic film between the zeranin and the constant temperature heat-sinking plate. The best material to use here is the thermally conductive version of the common Kapton (Polyimide) 'HN' film. The thermally conductive version is known as 'Kapton MT' with a thermal resistance x3 lower than the common HN version. Unfortunately, the MT version is hard to get, and the largest sheets I have been able to obtain (without spending k$ for large quantities) are A4 size, which is just slightly too small, damn it.

The customer can live with the 5ppm drift, and typically the largest step change in current is <8A, with drift of 2.5ppm. That said, I live by the philosophy that if I can measure an imperfection, then it should not be there. 

I'll mention that most precision designers don't use shunts at >10A, precisely because it is so damned difficult to avoid self-heating induced changes in temperature and resistance. The usual approach is to use a magnetically based sensor that does not suffer from any self heating at all, such as the Danfysik (now LEM) 'Ultrastab' range of current sensors. Look inside a million$ MRI machine, or commercial current driver for magnetic coils used in science research, and you will find a Danfysik current sensor. I have used these sensors in the past, but they have issues of their own. I have placed a Danfysik current sensor in series with the 16A current driver output, and was horrified to find that the measured current drifted by 200ppm (not a misprint) over the first 300 ms after a step current change, then overshoots, then converges to an approximately stable value. I contacted the manufacturer, who at first blamed my measurements, but after a month of email communication, were eventually able to independently verify that I was correct. To this day there is no mention of this imperfection in their data sheets though. The technical term that describes this effect is 'settling time'. They need to measure and specify the settling time to 100 ppm, 10ppm, 5ppm and 1ppm, but the trouble is, if they were to do so, then demanding customers may not buy the product. Or maybe they would anyway, because you can't buy a shunt resistor that does not drift immediately after application of 16A or more. The customer is presently using a 16A current driver that I built many years ago, employing a Danfysik Ultrastab current sensor, so will be well satisfied with better than order-of-magnitude improvement. The Ultrastabs are noisy, too, with the new current driver having better than order-of-magnitude lower noise as well. 

At an academic level, I am never happy with any observed effect that I do not understand, so regardless of customer needs, I would go mad if I did not at least fully understand what was going on. I'll get to explaining a possible detailed explanation, really just a refinement of some of the ideas already presented, and see if it withstands scrutiny.

Cheers, Colin 

 



 

[Zeranin]

.... maybe you need to invest in another L+N shunt and use that in your machine (or attempt to copy how it is made).

A good point, but it is not that easy. The reference shunt is 10 mohms, so the heating is a mere 2.5W at 16A. It is also huge, the naked manganin strip being 100mm x 0.7mm x 1400mm, folded 6 times. I also temperature control the manganin strip to within 0.02K, just to be sure. Self-heating induced changes in temperature and resistance (at 16A) are truly negligible.

Unfortunately though, I can't use a 10 mohm shunt in the current driver, because the voltage developed (0.16V at 16A) is too low, and would result in excessive noise. Therefore the zeranin shunt in the current driver is 0R1, giving 1.6V of signal, and improving the SNR by an order of magnitude. Also, thermoelectric effects become very significant with a 10 mohm shunt, and I need to work hard to keep that under control when using the 10 mohm L+N at 16A, an issue that I can do without, and that all but disappears with a 0R1 shunt.

Life was not meant to be easy.

I do have an identical 0R05 zeranin shunt that would have the initial drift to 2.5ppm, at the expense of slightly higher noise. Re 0R1 or 0R05, its 6 to one, and half a dozen to the other.

With unlimited time and budget, I would use a large, naked oil-immersed Zeranin shunt, with pumped circulation of the oil past the zeranin to reduce thermal resistance to an acceptable level. The whole show would be temperature controlled with the sensor attached to the zeranin.

[amspire]

It is pretty difficult to track down temp coefficient issues when the cost means you cannot keep making experimental shunts until you crack the problem.

If it is possible to run test with a temp sensor moved to various locations - a few different places on the aluminium, the copper, the Peltier cooling. Then apply a load and plot the different temperatures versus drift, you may find a correlation between one of the temperature drift curves and the shunt drift. It can help narrowing down the problem.

The other thing that could be useful is if it is possible to etch out a very thin extra Zeranin resistor - say along one edge of the current shunt resistor. You could then be monitoring the resistance changes in this thin resistor at the same time as load is applied to the main resistor. This would probably make it easier to see the Zeranin resistance changes with temperature, and it would also show of the effect is inherent in the bulk TCR of the Zeranin, or if the effect only shows in the track with the power applied - like a PCR effect.

If the monitor resistance track does match the main shunt resistor for TCR, then this monitor resistance track could be used to generate compensation for the TCR in the main shunt.

Richard

[zlymex]

Further improvement could be made by decreasing the thermal resistance of the 0.06mm thick PVC plastic film between the zeranin and the constant temperature heat-sinking plate. The best material to use here is the thermally conductive version of the common Kapton (Polyimide) 'HN' film.

Have you tried the blue silicone pad? although may be thick(I'm sure there will be thinner version), it conduct heat very efficiently and also with very high insulting resistance(I measured at >1E12 in anyway)
http://www.aliexpress.com/item/FREE-SHIPPING-400-200-2-5mm-Blue-Thermal-Conductive-Silicone-Pad-mat-for-computer-and-laptop/651156177.html

And because it is soft, any gaps will be filled by apply not very large force, also the joint will be seamless if smaller pieces to be used.

[Zeranin]

Zeranin-30 is in the Manganin family of resistance alloys-- and as such has a very large pressure coefficient.  So, never mind the barometric pressure changes, which could be bad enough, the pressure changes on the element from the difference in TCE between copper and aluminium could account for the "PCR" seen.

Yes, I too have wondered about the effect of the pressure provided by the clamping plate, but we need to think more carefully. The pressure exerted by the plate is at all times provided by the x8 of brass M4 screws, that pull the plate onto the constant-temperature aluminium heat-sinking plate, with the zeranin resistor assembly sandwiched between.
The total cross sectional are of these screws is negligible compared to the area of zeranin, so these screw can never produce a large force, relative to the area of zeranin. You appear to be analysing the problem as if the zeranin, the aluminium substrate, and the copper plate, were sandwiched between two totally immovable plates, in which case temperature changes in the zeranin, alum-substrate or copper would indeed result in massive forces, but that is not the case at all. For all practical purposes, the zeranin/substrate/copper are unconstrained and free to expand and contract as they choose in the direction of their thickness – the clamping screws could as well be made of plasticine in this respect, because cross sectional area of the screws is so negligible compared to the area of the zeranin.
As further, very convincing evidence of this, I can even loosen off these screws, and observe no measurable effect in the resistance of the zeranin shunt. Another enticing theory bites the dust.

If we agree that the zeranin is effectively unconstrained in the Z-direction, and I insist this is the case, then at first glance the explanation for the 5ppm drift is easy. The zeranin is mechanically constrained in the  X &Y directions, in the plane of the material, because it is mechanically bonded to an aluminium substrate that is much thicker than it is. Easy peasy. Zeranin expansion coefficient is +18ppm/K. The zeranin heats by 0.25K with respect to the substrate, and as a result, expands by 4.5ppm (18x0.25) in the Z direction, but is prevented from expanding in X and Y, and this will have the result of decreasing the resistance by 4.5ppm, just as observed.

Unfortunately there is a serious problem with this explanation. What’s good for the goose is also good for the gander. We would expect the same behaviour when measuring the R-T curve, but not so in practice. When the zeranin temperature is changed in unison with the heatsinking aluminium plate, as per the R-T curve, there is essentially no change in resistance, yet there will indisputably still be the same expansion of the zeranin in the Z-direction, with any increase in temperature. The zeranin will still be constrained in the X & Y directions, though there will now be a small change in length in X and Y due to the different COEs of zeranin and aluminium. However, this won’t have any effect on the zeranin resistance, because an equal change in X and Y (a length and crosss section term) cancel, the increase in length cancelling the increase in width as far as resistance is concerned. Drats! You can try until the cows come home, but this explanation just won’t work. Sure, you can explain as above why the zeranin changes resistance when it self-heats above it’s substrate, but then you won’t be able to explain why the same thing doesn’t happen when both are heated in unison.

Curiouser and curiouser said Alice. (a favourite quote of mine from Alice in Wonderland) There must be something else going on.

[Zeranin]

Further improvement could be made by decreasing the thermal resistance of the 0.06mm thick PVC plastic film between the zeranin and the constant temperature heat-sinking plate. The best material to use here is the thermally conductive version of the common Kapton (Polyimide) 'HN' film.

Have you tried the blue silicone pad? although may be thick(I'm sure there will be thinner version), it conduct heat very efficiently and also with very high insulting resistance(I measured at >1E12 in anyway)
http://www.aliexpress.com/item/FREE-SHIPPING-400-200-2-5mm-Blue-Thermal-Conductive-Silicone-Pad-mat-for-computer-and-laptop/651156177.html

And because it is soft, any gaps will be filled by apply not very large force, also the joint will be seamless if smaller pieces to be used.

I always regarded the rubbery thermal interface materials as inferior, and some of the older products of this type were almost useless, appealing to those that wanted a quick, easy, non-messy solution, whereas my own philosophy is to use the best, meaning lowest thermal resistance available, period. That said, I'm amazed by the high thermal conductivity of the materials used in some of these higher performance rubbery sheets, even if the large thickness offends me. The product you reference has a thermal conductivity of 1.5 W/mK, whereas the plastic film that I used to mount the zeranin shunt is a mere 0.19 W/mK, but MUCH thinner. Even so, when I do the calculations, I find that a 0.5mm thick film of your material would perform as well as the 0.06mm PVC film that I used. I really should make an effort to use something better. I was not able to get hold of a sufficiently large sheet of Kapton MT, that would gain me a x3 advantage, but there are rubbery sheet materials out there with thermal conductivities as high as 3.0 W/mK, and some available in thinner than 0.5mm, so I could do maybe x3 better than now by shopping around for one of the thinner, very-high-conductivity rubbery sheet materials. Thank you for alerting me. Such materials might have creep problems in this application, but never know without trying.

[Zeranin]

I think I explained the effect you are seeing in that post.  When the shunt self-heats, there is a temperature difference between the substrate and the element.  When it is heated in an oven, there is no temperature difference.  Mystery solved.  This temperature difference is something that you can reduce by using better materials and/or better construction techniques-- but at what cost?  AND, you don't have time to do this anyway as per your previous post.  The way to fix it is to compensate for the effect (in hardware and/or software).

There is a shunt that was designed by VPG for the CERN project, with help from John Pickering.  You can search for the paper about it on the Internet.  The new shunt is the CSNG series, which is [arguably] the best shunt on Planet Earth.  That said, even the CSNG series has a PCR.  You are simply not going to make this go away, no matter what you do.  So, if I were you, I would quit fighting it, and just compensate for it [somehow].

When the shunt self-heats, there is a temperature difference between the substrate and the element.  When it is heated in an oven, there is no temperature difference.

I agree with this 100%, and from a practical point of view, this tells us what options are available for reducing the problem. However, It is quite a stretch to say Mystery solved.

Beyond 'handwaving', no one has offered a detailed explanation of what is going on. No one has explained the nature of the stresses and strains that result in the 5ppm drift that I observe, or why the R-T curve does not apply, and so on. I deliberately posted an analysis of the stresses and strains that we should expect, pointing out that the stresses and strains that we predict and expect specifically do not explain what is happening, which specifically makes this a case of Mystery not solved, doesn't it?

The best way to fix it is to reduce the root cause of the problem, only then should one consider adding 'corrections', at least that's my philosophy.

I am most interested in those CSNG shunts that you mention. They are relatively recent, and I was not aware of them. The PCR claims to be 4 ppm/W, which is only 20 times worse than the 0.2 ppm/W (5ppm/25W) that I achieve, and certainly the best off-the-shelf shunt that I have seen, and perhaps the best off-the-shelf shunts on Planet Earth. Apparently you can get (on special order, at unknown price) x6 of them on a single substrate, so it would be possible to match what I have now with x3 or x4 of those. There are complications paralleling multiple 4-terminal resistors, basically you need a separate differential amplifier for each, but that is quite practical if you only need 3 or 4. However, to do significantly better than I presently achieve by using those VPG shunts instead would be very messy and expensive, and probably not practical.   

[Alex Nikitin]

I still think that your amplifier noise can be improved and the shunt voltage (and dissipation) can be lowered, reducing all power-related effects. The self-noise of the shunt is very low compared to 40 Ohm or so of the amplifier equivalent noise. Thermoelectric effects may be easier to deal with.

Cheers

Alex

[zlymex]

I have cut open one of the CSNG shunts six chip version, 0R0500, looks good inside.

[Zeranin]

I have cut open one of the CSNG shunts six chip version, 0R0500, looks good inside.

I am absolutely amazed! Did you just happen to have this CSNG resistor sitting on your bench? How many more have you got, and how and why do you come to have them? Assuming you bought it, do you mind me asking how much it cost?

[Zeranin]

I have cut open one of the CSNG shunts six chip version, 0R0500, looks good inside.

The Kelvin sensing appears to be incredibly primitive. It looks as if the sense connection is made to the high-current conductor, just beyond where the wires enter the package. The sense connections should be at the resistive foil, not several inches back along the tinned-copper high-current connecting wire. What am I missing?

[Zeranin]

I still think that your amplifier noise can be improved and the shunt voltage (and dissipation) can be lowered, reducing all power-related effects. The self-noise of the shunt is very low compared to 40 Ohm or so of the amplifier equivalent noise. Thermoelectric effects may be easier to deal with.

We appear to be at cross purposes. The Johnson self noise of the 0R1 shunt itself is so low as to be neglected. If we lower the shunt resistance, then the shunt noise is of course even less, but that's irrelevant. It doesn't matter how you cut it though, if you lower the shunt resistance, then you get less signal, and this degrades SNR at the shunt preamp output. Reducing the shunt to 0R05 (which I have) wouldn't degrade the noise performance dramatically, and may provide a better all round balance.

[zlymex]

I have cut open one of the CSNG shunts six chip version, 0R0500, looks good inside.

I am absolutely amazed! Did you just happen to have this CSNG resistor sitting on your bench? How many more have you got, and how and why do you come to have them? Assuming you bought it, do you mind me asking how much it cost?

I have another CSNG shunt(also six-chips) together with 20+ pieces of similar shunts(in the same datasheet as CSNG): VCS331Z,  VCS332Z and VFP4.
I cut it open about 2 years ago, someone sent it to me for free just let me cut open because it was new, strange and we have not seen inside photo of it before. I cannot remember the exact price of my other CSNG because I have bought many other shunts in these several years time, price tag roughly around $5 to $80 each

[zlymex]

The Kelvin sensing appears to be incredibly primitive. It looks as if the sense connection is made to the high-current conductor, just beyond where the wires enter the package. The sense connections should be at the resistive foil, not several inches back along the tinned-copper high-current connecting wire. What am I missing?

One current conductor is on top side of all the chips, covering with thick tin. partially can be seen.
The other current conductor is similar, go thru all the chips at bottom side, torn from the chips.
Kevin sensing are two thin wires(one can be seen) soldered to the middle of the two current conductors.

[Alex Nikitin]

I still think that your amplifier noise can be improved and the shunt voltage (and dissipation) can be lowered, reducing all power-related effects. The self-noise of the shunt is very low compared to 40 Ohm or so of the amplifier equivalent noise. Thermoelectric effects may be easier to deal with.

We appear to be at cross purposes. The Johnson self noise of the 0R1 shunt itself is so low as to be neglected. If we lower the shunt resistance, then the shunt noise is of course even less, but that's irrelevant. It doesn't matter how you cut it though, if you lower the shunt resistance, then you get less signal, and this degrades SNR at the shunt preamp output. Reducing the shunt to 0R05 (which I have) wouldn't degrade the noise performance dramatically, and may provide a better all round balance.

What I've meant that your amplifier appears relatively noisy (~80nV RMS noise in 10kHz BW, or 0.8nV/rtHz, equivalent of ~40 Ohm resistor noise at 300K) . If you use multiple amplifiers in parallel to achieve a lower noise, this figure looks excessive.

Cheers

Alex

[Kleinstein]

The Kelvin sensing looks a little primitive, but it is not that bad as one might think at first. The copper / solder part is mainly responsible for spreading the current over the parallel connected resistor elements. As the length of the copper path is the same for all elements the current is evenly divided in the resistors, even if the copper part changes resistance. Then there is not much tin /copper in the path sensed by the voltage sense connectors. I agree it could have been done better, but as long as the parts stays stable this should not be such a bad thing. The copper/tin part might add a little to the TC, but not much. 

The main part I would be afraid of would be aging in the tin part, as this is a relatively low meting point alloy with normally a fine structure, it can change even at room temperature or slightly above. As it's potted there should be no tin whiskers growing.

As for the amplifier a value of 0.8 nV / Sqrt(Hz) is not that bad, but a parallel connection of 4 amps could give you about halve the noise and thus allow for the reduced shunt size. It's a trade of in spending money for the shunt or the amplifiers.

[d-smes]

If we agree that the zeranin is effectively unconstrained in the Z-direction, and I insist this is the case, then at first glance the explanation for the 5ppm drift is easy. The zeranin is mechanically constrained in the  X &Y directions, in the plane of the material, because it is mechanically bonded to an aluminium substrate that is much thicker than it is. Easy peasy. Zeranin expansion coefficient is +18ppm/K. The zeranin heats by 0.25K with respect to the substrate, and as a result, expands by 4.5ppm (18x0.25) in the Z direction, but is prevented from expanding in X and Y, and this will have the result of decreasing the resistance by 4.5ppm, just as observed.

Unfortunately there is a serious problem with this explanation. What’s good for the goose is also good for the gander. We would expect the same behaviour when measuring the R-T curve, but not so in practice. When the zeranin temperature is changed in unison with the heatsinking aluminium plate, as per the R-T curve, there is essentially no change in resistance, yet there will indisputably still be the same expansion of the zeranin in the Z-direction, with any increase in temperature. The zeranin will still be constrained in the X & Y directions, though there will now be a small change in length in X and Y due to the different COEs of zeranin and aluminium. However, this won’t have any effect on the zeranin resistance, because an equal change in X and Y (a length and crosss section term) cancel, the increase in length cancelling the increase in width as far as resistance is concerned. Drats! You can try until the cows come home, but this explanation just won’t work. Sure, you can explain as above why the zeranin changes resistance when it self-heats above it’s substrate, but then you won’t be able to explain why the same thing doesn’t happen when both are heated in unison.

Zeranin- Thanks for bringing your problem to the forum and for your detailed descriptions and responses.  Everyone loves a mystery!

I would like to question your assumption that the Zeranin is constrained in the X and Y axis.  As-built, the Aluminum X & Y are locked to the Zeranin X & Y through the heat-bonding, electrically insulating film.  But that film has mechanical compliance and its own TCE (which is safe to ignore).  The mechanical compliance can be thought of as an array of stiff springs connecting each Al X-Y coordinate to the same Zeranin X-Y coordinate at room temperature (as-built).  Now when heated uniformly in an oven (no current), the Al and Zeranin want to grow to two different sizes because of the differing TCEs but they can't because they are constrained by all these stiff springs.  So, does the Al get compressed to match the Zeranin hot dimensions or does the Zeranin get stretched to match the Al hot dimensions?  My guess is that the Al, even though it's softer, is thicker and ultimately has the higher stiffness.  So the Zeranin gets stretched when the assembly is heated.  But the springs of the insulating film also get stretched such that the hot X-Y coordinates no longer match; the Zeranin is stretched to slightly smaller X-Y dimensions than the Al.  In this regard, the rubbery thermal interface may have and advantage because it has more "give" (weaker springs).

Also recognize that in the dimension of the zig-zags of the Zeranin, there are gaps which will take up a lot of the differential TCE in that dimension.  However, these put additional strain on the corners where the zig bends to become a zag.  But I agree, that within each 25mm zig and zag the Zeranin is still getting stretched in both dimensions.

Now apply current.  The Zeranin gets 0.25K warmer and would have slightly larger dimensions, if it weren't constrained, by an amount I'll call dx and dy.  In this case, the Al X-Y coordinates stay the same (assumed constant temperature) but what will the Zeranin's X&Y's become?  I assume the spring constant of the insulating film remains constant and stretches the Zeranin by the same amount.  It follows then that the differential heated dimensional difference of dx and dy shows up as Zeranin's X&Y coordinates growing by dx and dy (in reality, slightly less than dx & dy due to slightly less spring force).

I believe there is also a slight temperature difference between the cooled side/face of the Zeranin and the top clamp side of the Zeranin.  This means the clamped face area expands slightly more than the cooled face causing the Zeranin to want to curl or cup (convex on the warm side, concave on the cool side).  Given the thinness and good thermal conductivity of the Zeranin, this is probably a second-order effect at best.  But it also results in different stress/strain and dimensional changes that may explain the powered vs un-powered difference in R-T.

[Zeranin]

One current conductor is on top side of all the chips, covering with thick tin. partially can be seen.
The other current conductor is similar, go thru all the chips at bottom side, torn from the chips.
Kevin sensing are two thin wires(one can be seen) soldered to the middle of the two current conductors.

I feel much happier now we can clearly see that the sensing has been done properly. I could not believe that it was done as badly as first appeared. Thanks so much for providing these pictures.