EEVblog #544 - Fluke 5450A Resistance Calibrator Teardown

[Chipguy]

Hey folks,

Dave's video gave me an idea to research something:
The question was : Can you order custom made resistors with affordable prices / small numbers.
The reason why I asked myself the question is that I have an upcoming project where I need resistor values like:
1 / 2 / 4 / 8 / 16 / 32 / 64 / 128 Ohm .... you can see, it's binary

I found this:
http://www.vishaypg.com/foil-resistors/videos/?video=33

They make you 0805/1206/2010 and 2512 resistors with 0.01% tolerance and 0.2 ppm TC.
They can be ordered via digikey for example.

I have not found other distributors than Digikey that process the order for end customers.... yet.

[ddavidebor]

how much this little jewels cost?

[Chipguy]

how much this little jewels cost?

About 11 to 19 EUR each.

[ddavidebor]

Well, that's pretty good

[codeboy2k]

I'd like to have on hand some of those bulk, uncut resistors

I can measure myself and have my own microscope. 

[hpux735]

Hi everyone! 

When I saw this video and Dave's call for information about the resistors, I immediately thought of my friend Tom's brother.  He has worked at Fluke for 25 years, and improved a process for building high-precision resistors for the standards industry.  He hasn't had a hand in this product, or it's resistors, but he personally knows the people that almost assuredly assembled them.  This thanksgiving, Tom got a raft of information about the process and the story of one of the nude virgins in utopia-land (which is Everett, Washington in case you were wondering.). 

As several of the responders have mentioned, the form used on the low value resistors is mica, not plastic.  Also, those resistors are almost certainly out of spec now that the case is open.  Resistances that low are very sensitive, and can be changed by dust on the surface.  Not that Dave probably cares that much, I assume he bought that unit for our viewing pleasure (thanks, Dave!).

The can resistors are wire-wound using enameled nichrome wire.  They are made individually, by hand, using lathes.  The resistance per foot of the wire is well known, and used as a first pass for the desired resistance.  The resistors are hermetically sealed in glass that has the same coefficient of thermal expansion as the nichrome wire.  I'm not clear on the details, but they are terminated using a technique called "water welding".  I can't really figure out what that is, but apparently it's '60s era tech.  The final resistance is fine-tuned by adjusting the length of the lead that is terminated.  Apparently, this is where the comment about the nude virgins in utopia-land was hilarious because, predominately, the women that make these resistors are about 60.  One woman in particular has been making resistors for Fluke since she was 18.  After about 40 years, she's getting close to retiring.  Also, the nichrome spool that they've been using to make the resistors is about exhausted, so they will become awfully hard to source in the near future. 

It's possibly that I'll get some of the raw materials in the mail soon.  If I do, I'll take high quality pictures and post them.

With regard to the pairing of the resistors:  They are indeed paired, but not for the reason you may think.  They select resistors based on the TCR (Temperature coefficient of resistance, I assume).  They determine that across a +/- 20 degree C range (let's assume) that 3 resistors have a +.3 ohm change and one has a -1 ohm change.  When you place these 4 resistors together, they will average near a null temp-co over the range.  The nichrome wire used in these resistors is terrible for temp-co.  If they are heated +20C, the wire expands and decreases the resistance.  It may remain this way for a week or two before it returns to spec.  This is why resistor pairing for temp-co is so important in this application.

In case you were wondering about the resistors used in late-model standards, they are still hermetically sealed in glass, but have a MUCH better temp-co.  The resistive material is sputtered (I believe) onto one side of the glass package.  On another sheet of glass, an inert coating is applied.  These pieces of glass are mounted to each other face-to-face (coatings on the inside).  The perimeter is also sealed with glass.  At this point, the resistance is roughly correct.  To get it perfect they laser ablate the resistive material to trim the resistor.  His contribution to the process is tuning the chemistry of the materials such that a minimum of glass is ablated with the resistive material, which increased yield from about 5 to 80 percent.

[EEVblog]

Awesome info, thanks!
I was coming to the conclusion that the Vishay precision resistor group did these resistors. Great to hear Fluke did them!, and that the group is still going?

[hpux735]

Yes, the group is still going. It seems they're still considering whether they should buy another spool of the nichrome wire...  That's assuming they can get a new crop of nude virgins, that is.  The hermetically-sealed laser-trimmed resistors are still made in house at Fluke in Everett (and also Scotland, I think).

[BravoV]

@hpux735, thank you, what a great infos you have there, and please keep them coming ! 

[EEVblog]

Yes, the group is still going. It seems they're still considering whether they should buy another spool of the nichrome wire...  That's assuming they can get a new crop of nude virgins, that is.  The hermetically-sealed laser-trimmed resistors are still made in house at Fluke in Everett (and also Scotland, I think).

I'm very surprised the group hasn't been given the arse by the "The Danaher System", as equivalent resistors could be outsourced to Vishay.

[SeanB]

Don't think the Vishay ones are accurate enough or stable enough to use as a primary standard. Might be some legislation calling for a certain thing, like with RAMBUS.

[hpux735]

Yes, the group is still going. It seems they're still considering whether they should buy another spool of the nichrome wire...  That's assuming they can get a new crop of nude virgins, that is.  The hermetically-sealed laser-trimmed resistors are still made in house at Fluke in Everett (and also Scotland, I think).

I'm very surprised the group hasn't been given the arse by the "The Danaher System", as equivalent resistors could be outsourced to Vishay.

From what I understand, the Danaher purchase didn't change the group all that much.  I know that the employee perks (such as the employee store) got much worse.  It used to be that employees could buy equipment a vast discounts.  Now, it's more like a yearly auction format.

[Dr. Frank]

As I recently acquired a Fluke 5450A, I'd like to revive this old thread, adding some interesting information.

My instrument came from the German Air Force, last calibrated in 2003 (coincidentally by Air Force cal lab, where myself worked 35 years ago).
It was obviously stored in a mining tunnel depot since then, and now sold as surplus.

The manufacturing date is about 1987/88, 2 years older than Daves instrument.

When I checked the resistors by a 34465A and a 3458A, I found the same drifts, as Dave did, mostly all lower resistance values drifted upwards, 1k and 1k9 more than the others, and the high ohmic ranges, 10M, 19M and 100M read extremely low, between -1000 and -5000 ppm.

So I quickly found the fault by 'nosing': The transport box and the instrument itself smelled like an old, musty cellar.
It was obviously stored under high humidity conditions for an extended period of time, which might well cause leakage paths.

When I opened  the instrument, see picture 1, it contained moulded ELECTROL reed relays for these high ranges, instead of the COTO types, like in Daves instrument.
Moulded components can very well suck water vapour inside, which would explain, where the leakage is located. 

So I heated the relays indirectly by an incandescent lamp to about 55°C, for about 5h.

These gross errors now have vanished, still reading a bit low, to the same level as shown in Daves video.
I will come back to that point later.


During the heating period, I monitored the 10k value, which was extremely stable to < 2ppm, meanwhile the temperature of these resistors also had increased by about 15°C.
That means, they really have a T.C. of around 0.15ppm/K.
Great performance, but how is this accomplished?

I now had a closer look & calculation on the T.C. matching, which is really very tricky.
You can see in the pictures, that the individual T.C.s are written on the resistors, with sign and in multiples of 0.25ppm/K.

So, the two 450k resistors are marked P0.75 and N0.50, which denotes +0.75ppm/K, and -0.50ppm/K.
For the calculation of the combined T.C. of the resulting 900k resistor, the difference of their T.C.s has to be averaged also,  so the resulting T.C. is 0.125ppm/K only, opposed to the expected 0.25ppm/K.

For the lowest resistor arrays of this chain, 10 Ohm, 9 Ohm, 81 Ohm, one can easily see the individual T.C. markings.
The paralleling of four resistors gives 4 fold improvement of the T.C., and also mitigates self heating effects when higher measurement currents are applied.

These big vias are used as the resistor reference points with Kelvin connection on the PCB bottom side.

This pairwise matching / averaging 'trick' can be found in all Fluke instruments, like the 332B/D, 752A, and the 742A, which explains their extremely low T.C.s, even after all these years.


I took the readings of all resistors from 10 Ohm to 9 MOhm and calculated the overall T.C.s.

So the 5450A resistors have T.C.s of -0.54ppm/K worst case, and 0.1ppm/K for the 10 kOhm resistor, which is in full agreement with my rough measurement.

[Dr. Frank]

Part 2: Calibration

The 5450A consists of two nested, multifold  10:1 resistor dividers.
Therefore in CAL mode, it can very conveniently be calibrated, by using a stable DCV source and a precise 10:1 transfer instrument.

Instead of the bulky setup in the manual, I only used the 5442A calibrator and the 3458A, using its extremely linear A/D in DCV mode.

You first compare an external 10k reference resistor to the 5450A 10k and 19k resistors, and calculate their absolute values.

After that, the lower or upper decade values (i.e. 1k or 100k) are measured by making precise 10V:1V voltage comparisons.
The 3458A is capable of making 0.3ppm precise transfers.

Within 4 steps, the extreme values are calibrated, at about 1.2ppm accuracy, due to the cumulative ratio error.

The comparison against the 12 year old calibration revealed a typical drift of less than 1ppm/year.

On the highest resistors, 100 MOhm, leakage currents of the 3458A (up to 20pA) may deteriorate the accuracy, up to +/- 180ppm.

By using 50V for the excitation voltage (which can be measured precisely to < 1ppm), this error is reduced to about 36ppm.

Paralleling the 3458A alternatively to the upper or lower resistor at 50V and 5,55555V, giving the same ~ 5V reading (generating identical leakage current), allows to make precise corrections for the real resistor values, and also to solve for the leakage current.
Latter was estimated to be between 2 and 11 pA.
(See attached xls calculation.)

So even the 100M resistor could be determined to an uncertainty on the order of about 10ppm.. The fully corrected and the 50V excitation method differed by 5ppm only!

As a surprise, the 3 upper resistors were now in accordance to their nominal values, a few ten ppms off only.

Despite that result, the HP instruments both still measure very low values.
These are about within their specified accuracy of 50 and 500ppm

I assume a systematic problem with the HP resistance circuits, more probable than residual internal leakage currents in the 5450A.

I further assume, that the internal isolation of the HP instruments is too low, as the whole Ohm measurement circuitry involves the front / rear switch, the PCB, overload protection, and so forth.


As a conclusion this 5450A resistance calibrator performs much better than its specification:

The T.C. is about 5..10 times better.

The uncertainty specification may also be better, to the 1ppm level.

This is simply due to the nowadays easily available Ohm Offset Compensation method.
Impacts from thermal voltages were back then included into the spec, instead of removing them during qualification.
 
Frank   
   

[3roomlab]

very nice info dr Frank, did you need to physically apply IPA to clean anything inside?

[Dr. Frank]

very nice info dr Frank, did you need to physically apply IPA to clean anything inside?

Definitely not!
I did not touch anything inside.
The PCB was relatively clean, but something like dust or metallic particles were visible all over the top side and around the reed relays.
I removed that carefully with a fine, clean brush.

Frank

[zlymex]

This is the first time I read this thread, a lot of useful info, thanks every body!
I have three of these 5450A, all in good condition. One of the amazing thing about it is the 81 meg WW, I cannot image how the hell they made it. Even the diameter is 0.00004", 5km of wires is needed!
For the past 10 years, I've been gathering information on these hermetically sealed WW resistors. I believe its still unrepeatable in many aspects. It's true(and pity) that fluke, the main company used so many of these, no longer use these resistors any more except in some hi-end products such as 732B(6.2k, 11.42k, 180, 1k, 8.823k), 742A(20.01 Ohm, 39.992k) and 752A(119.8k, 120k).

Modern calibrators such as Fluke 5720A(the grand son of 5450A, if I may say so) inherent many things from 5450A. Here is the photo of the Ohms board of 5720A and fragment schematics from it's serving manual


WW have been replaced with those film resistors, and big relay replaced by small latching relays, basic principle remains the same. As for the performance, some range better but others inferior. They still use WW resistors for 1 and 1.9 Ohm range.

[RobK_NL]

Interesting!

That 90M part looks like it's a standard Caddock device; probably TF series.

[zlymex]

Interesting!

That 90M part looks like it's a standard Caddock device; probably TF series.

I'm sure you are right. I don't have front photo of the 90M, but I have a back photo of Caddock TF656N.

[quarks]

Does any know where to get the 4.5 MOhm resistors or a replacement?

[zlymex]

Does any know where to get the 4.5 MOhm resistors or a replacement?

That would be very difficult because hi-value WW resistors are difficult to make. Afaik, they only obtainable by taken out from other 5450A. Fluke replace those 4.5M with film type in their new design.
Here are some of my old stuff.

[Dr. Frank]

Hello lymex,

it's very nice to read about your great experience here @ eevblog!


Did you measure a noteworthy change on these high ohm resistors from the Fluke 5450A, i.e. the 4.5MOhm and 81 MOhm resistors, towards smaller values, being about 200ppm or even 500ppm lower than specified?

And how did you measure these resistors, as for example all the HP DMMs are not capable of making precision measurements on high Ohm resistors?

As I wrote, I calibrated these resistors inside the 5450A by using 50V as a stimulus and measuring the 10:1 ratio .
This gave calibration values, which were in the correct ballpark, and the annual drift even on the 10M and 100M resistors was not so high.

When measured with my 3458A in 2W Ohm mode, I found the same big decrease in the 10M and 100M values, like quarks and also Dave did, which I still cannot explain, especially not the discrepancy between the 50V calibration method and the 5 V Ohm method inside the 3458A... Possible leakage currents inside the 5450A should be the very same in both configurations.


Has anybody of you, lymex and quarks, measured the 5450A high Ohm resistors with a Fluke 8508A or similar, using its high precision high Ohm mode, which sources up to 200V as compliance voltage?
This 8508A is really superior over the 3458A, concerning such resistor measurements...

Frank

[zlymex]

Hi Frank,

That is a good question, and had been my headache for sometimes. You are right, the resistance of 3458A has been the weak point not only for high ohmetic but even 10k is inferior than other 8.5 digits DMM. Therefore, I use my 3458A mainly at its best - 10V range.

I had a plan several years ago to make a guarded high precision hi-resistance bridge, to compare resistor in 10M to 1T range(but only decades), testing voltage would be 100V, which is the standard test voltage in here for hi-resistance. The idea is based on the Guarded Active-Arm Bridge(see page 31 of NIST TN1458). I'm also material ready for build guarded hamon transfer standard of 10M, 1G and 100G, but I just bite off more than I can chew. I vanished from this forum for about a year and also very quite in that period. Since I come back, I'll finish what have been left over.

I build a very stable 100V standard before by cascading 10 LT1021B to facilitate the 100V required to measure mega ohms, but again, no serious measurements done on those 4.5M resistors.

I did measured those hermetic resistors in the box in lower voltage that most of them show a slight positive increase of about 50 to 360ppm. No negative change of more than 100ppm ring a bell. But I cannot remembered that I use 3458A to measure 5450A for 1M or above.

[quarks]

Has anybody of you, lymex and quarks, measured the 5450A high Ohm resistors with a Fluke 8508A or similar, using its high precision high Ohm mode, which sources up to 200V as compliance voltage?
This 8508A is really superior over the 3458A, concerning such resistor measurements...

Hello Frank,

no I did not use "HiVOhm" of the 8508A because max. allowed voltage is 50V (at least that is stated in the 5450 manual). 
But I am glad you asked, because of that I meassured the Fluke 5450A with my Fluke 8508A and the Fluke 5440A-7002 cables again and found all values to be in spec.

My guess is, although I used the very same cable, most likely my setup (1 or 2 years ago) was just not clean enough. Because today I cleaned my 5450 inside and also the cables. 

bye
quarks

 

[Dr. Frank]

Hello Frank,

no I did not use "HiVOhm" of the 8508A because max. allowed voltage is 50V (at least that is stated in the 5450 manual). 
But I am glad you asked, because of that I meassured the Fluke 5450A with my Fluke 8508A and the Fluke 5440A-7002 cables again and found all values to be in spec.

My guess is, although I used the very same cable, most likely my setup (1 or 2 years ago) was just not clean enough. Because today I cleaned my 5450 inside and also the cables. 

bye
quarks

Hello quarks,

that's an interesting information, that your 5450A NOW seems to be in specification.

Would you mind to test, if there is a difference in reading, between the 8508A and the 3458A for 10M and 100M?

I still suspect the 3458A on these high ohm ranges, especially as I found this contradiction, i.e. the volt ratio method (10:1 divider at 50V) gave reasonable values for these resistors, instead.

HP had defined barn-door wide high Ohm specs, for some good reasons, I assume..

And a last comment: The 10k to 1M Ohm ranges in contrary are quite precise on the 3458A, and not as bad, as lymex stated.
You have to use the instrument properly, e.g. by using appropriate delays on OCOMP.

Cheers, Frank