EEVblog #730 - Thin Film Resistor Networks

[EEVblog]

How ceramic backed precision thin film resistor networks are constructed and laser trimmed.
Examples come from the resistor networks in the 7.5 digit Keithley DMM7510 Multimeter, manufactured by Fluke.

[Willy.nebula]

Ok is it me or does that first resistor network in the video look like it has a face in the top right corner 

[VK3DRB]

Excellent video Dave! You didn't waffle on - it was both informative AND interesting.

Years those, the bonding technology between the pins and the ceramic circuit was not near as good as today. The old AWA Thorn 4KA colour TV chassis was plagued by these connections going open circuit intermittently, causing "random" intermittent CRT convergence problems. The English 4KA's were a dreadful design, generally ending up on the rubbish tip only a few years after they were bought. A quick and dirty attempt to fix the ceramic contacts was hitting the lead with a soldering iron for about 1 minute. Sometimes it worked, sometimes not.

The algorithms for how to trim resistor networks quickly would be interesting in itself.

[SNGLinks]

I wonder if they use glass so they can laser trim them after encapsulation?

[nitro2k01]

Just to nitpick a little. Around 3:15 you said that the portion of metal on the left adds resistance. It really adds conductivity/decreases resistance compared to if that part was isolated, though, since it adds however many micro or nanoohms in parallel. (Not trying to claim you don't know of course. Only a minor objection to the wording.)

Otherwise a great video.

[Dr. Frank]

Dave, very interesting and informative video!

Some remarks, though:

FLUKE made these glass covered resistor networks for their high grade calibrators and references in first place.. You'll find these in the 732B, the 5720A and also in the 8508A 8 1/2 digit "Reference Multimeter".
These networks give these instruments their superior stability.
The reason, it's an open, glass covered design, is simple.. The LASER trimming is done after the sealing of the element, through the glass plate, and in some cases obviously in situ, in the finalized instrument at the final testing place.
Imagine, you use that network for final trimming of the 732B output, i.e. absolute 10.00000V and 1.00000V.

The High Voltage divider should have exactly 9.9M / 100k .. I assume that there was some protection circuitry inside the instrument, giving these few kOhm instead.

The reason, Fluke  manufactured this special network, has two reasons or advantages over usual design ocompany Caddock, where they bond 2 distinct resistors back-to-back.
First, both resistors have identical T.C., as they have been sputtered from the same alloy target (NiCr or TaN), so the T.C. tracking is already near perfect.

But they also overcome the power difference T.C. effect in a usual HV divider..
The line width of each resistor seems to be the same everywhere.
The length of the 9.9M resistor path should be exactly 99 times the 100k resistor path.
So the power dissipation on each resistor element is equal.
That means, that the 9.9 M and the 100k resistor will heat up to the exactly same temperature when 1000V were applied. Additionally, the high thermal conductive ceramic substrate equalizes any temperature gradients, but on a slower time constant.
That means, that the 100:1 ratio will be constant under all circumstances, to < 1ppm..
Also dynamically, if you want to digitize an alternating 1000V level  on a short time scale.

Most other 6 1/2 ... 8 1/2 DMM suffer from that power dissipation effect.
They need a quadratic compensation, by calibrating at 1000V, 500V, and maybe 100V, to compensate later numerically for this change in ratio over power.
The 3458A does not have this compensation, because maybe it is not feasible to calculate the quadratic compensation for 100kHz digitizing at these high voltages, but it therefore has a mediocre 12ppm additional error at 1000V.

So, this Fluke Thin Film network really is absolutely perfect.

Frank

[EEVblog]

The algorithms for how to trim resistor networks quickly would be interesting in itself.

Yes, I'd love to see the manufacturing and cal facility and procedure for these.
The laser machines that do this are not trivial either.

[EEVblog]

Just to nitpick a little. Around 3:15 you said that the portion of metal on the left adds resistance. It really adds conductivity/decreases resistance compared to if that part was isolated, though, since it adds however many micro or nanoohms in parallel.

I assume you means 13:15. Yes, my goof, I was trying to say adds resistance in parallel as you say. Not the best wording at all.

[EEVblog]

The reason, it's an open, glass covered design, is simple.. The LASER trimming is done after the sealing of the element, through the glass plate, and in some cases obviously in situ, in the finalized instrument at the final testing place.

I had suspected that but didn't think it would be the best approach, perhaps due to heat buildup or other issues with the sealed glass. It didn't occur to me it could have been trimmed after installing in the product.

The High Voltage divider should have exactly 9.9M / 100k .. I assume that there was some protection circuitry inside the instrument, giving these few kOhm instead.

Yep, likely.

But they also overcome the power difference T.C. effect in a usual HV divider..

I thought about going into that, but the video was already much longer than what I had planned.

The line width of each resistor seems to be the same everywhere.
The length of the 9.9M resistor path should be exactly 99 times the 100k resistor path.
So the power dissipation on each resistor element is equal.
That means, that the 9.9 M and the 100k resistor will heat up to the exactly same temperature when 1000V were applied. Additionally, the high thermal conductive ceramic substrate equalizes any temperature gradients, but on a slower time constant.
That means, that the 100:1 ratio will be constant under all circumstances, to < 1ppm..
Also dynamically, if you want to digitize an alternating 1000V level  on a short time scale.

Most other 6 1/2 ... 8 1/2 DMM suffer from that power dissipation effect.
They need a quadratic compensation, by calibrating at 1000V, 500V, and maybe 100V, to compensate later numerically for this change in ratio over power.
The 3458A does not have this compensation, because maybe it is not feasible for 100kHz digitizing at these high voltages, but it therefore has a mediocre 12ppm additional error at 1000V.
So, this Fluke Thin Film network really is absolutely perfect.

[/quote]

Great added detail, thanks.

[G7PSK]

My guess is that the areas where you can see what looks like burning is metal vapour on the glass if the trimming is done after the glass is sealed onto the resistor network so the lines that look clear and sharp are cut before hand or nichrome vapour is transparent.

[EEVblog]

FLUKE made these glass covered resistor networks for their high grade calibrators and references in first place.. You'll find these in the 732B, the 5720A and also in the 8508A 8 1/2 digit "Reference Multimeter".

Do you know if it's the same part reused, or custom values/config for the DMM7510?

[EEVblog]

My guess is that the areas where you can see what looks like burning is metal vapour on the glass if the trimming is done after the glass is sealed onto the resistor network so the lines that look clear and sharp are cut before hand or nichrome vapour is transparent.

Yep, could well be the case.

[HighVoltage]

Very nice video and really good comments here
Thanks!

[Dr. Frank]

FLUKE made these glass covered resistor networks for their high grade calibrators and references in first place.. You'll find these in the 732B, the 5720A and also in the 8508A 8 1/2 digit "Reference Multimeter".

Do you know if it's the same part reused, or custom values/config for the DMM7510?

I assume, they used a similar  100:1 divider in the 5720A..
Actually, it's a 70k / 7M there, for a 100 fold amplification of 11V reference to 1100V output.

I always wondered, how they could use a single component for that crucial circuit element in the 5720A.. And how Fluke could make an Artefact High Voltage Autocalibration  to that degree of uncertainty,where HP in the 3458A failed at the exactly same feature.. Now I know!

See also the Fluke 5720A manual / schematics and the Theory of Operation...

For the other network.. There are several pictures of the similar components from the 732A, 8508A, but I doubt that   they reused these in the Keithley instrument.. As there the networks were used for the LTFLU or the LTZ1000 references..
Maybe you can determine, for which mode Keithley uses this network, as current shunts, or for the resistance mode.

Frank

[German_EE]

Thank you for an interesting video. I'm tempted to try this for myself using the laser cutter at my local Maker Space but I don't think it will cut nichrome or copper.

[fcb]

Great video. Where's the teardown/review of the 7.5 digit mm?

Gunhaver posted a comment on the YT comments "Danaher no longer owns Tektronix! The sale may not be completely final yet, but they sold off the Tektronix brand and part of Fluke Networks along with a bunch of other assets."

Any news?  Couldn't find anything recently on Google.

[EEVblog]

Great video. Where's the teardown/review of the 7.5 digit mm?

Not finished yet. Already got almost an hours worth of raw footage, it's getting out of hand.

Gunhaver posted a comment on the YT comments "Danaher no longer owns Tektronix! The sale may not be completely final yet, but they sold off the Tektronix brand and part of Fluke Networks along with a bunch of other assets."
Any news?  Couldn't find anything recently on Google.

All I could find was that they sold off Tektronix networks in Oct:
http://www.oregonlive.com/silicon-forest/index.ssf/2014/10/danaher_sells_communications_b.html

[smoo]

The algorithms for how to trim resistor networks quickly would be interesting in itself.

Yes, I'd love to see the manufacturing and cal facility and procedure for these.
The laser machines that do this are not trivial either.

Laser trim machines are not that complicated.  First you have the laser itself, which does require a big old power supply and water cooling, also some circuitry to pulse the laser.  Then the laser pulse has to be directed where you want it, either using prisms mounted on a table that can be moved in the x, y and z axis's, or mirrors mounted on galvos.
For the final in circuit trimming it may even be done manually with an operator moving the laser beam one pulse/step at a time while watching the readout on whatever instrument in use.

[dentaku]

Thumbs up for an interesting video and knowledgeable forum comments.
 

[Prime73]

thanks for an interesting video Dave.
On a side note a quick question - I've noticed in your videos you use different DMMs all the time. What's your reasoning when you choose a DMM to measure something? For instance in this particular video you've used Brymen BM257. Why this particular one and not say yours Fluke 85V ?

[EEVblog]

thanks for an interesting video Dave.
On a side note a quick question - I've noticed in your videos you use different DMMs all the time. What's your reasoning when you choose a DMM to measure something? For instance in this particular video you've used Brymen BM257. Why this particular one and not say yours Fluke 85V ?

Whichever one is closest to hand that has the probes I need usually, unless I have a specific need for a certain meter. Meters get thrown around the lab all the time, probes go missing and magically reappear etc. Even more so now with Dave2 in the lab a lot.

[Tek_TDS220]

He has a name: "Dave2" !!!!

[FrankBuss]

Interesting video. First I didn't understand why the resistance is ohms/square, but I think I found the solution: Let's assume one 1mm^2 square resistor, which has a resistance of n ohm. Now if you have a 2mm^2 square, you could divide it into two 1x2mm^2 parallel resistors. Obviously each one has a resistance of 2*n ohm, and in parallel it is again n ohm. That's neat and not obvious at first glance.

But why do they need to calibrate the resistors at all? As a software guy, I would just use some low temperature drift resistors, measure them and calculate the rest in software.

[TiN]

Hm, LTFLU voltage reference in DM7510? That's surprising Expected to see LTZ there.

Thanks for video, we desperately want moooore!

[zapta]

Interesting video. First I didn't understand why the resistance is ohms/square, but I think I found the solution: Let's assume one 1mm^2 square resistor, which has a resistance of n ohm. Now if you have a 2mm^2 square, you could divide it into two 1x2mm^2 parallel resistors. Obviously each one has a resistance of 2*n ohm, and in parallel it is again n ohm. That's neat and not obvious at first glance.

This was the most fascinating part of this episode. I presume it's measured from an entire edge of the square to the entire opposite edge.

You can also mention in your proof that the voltages between the two parallel resistors are identical at each point and thus there is no current between them. Otherwise, it would affect the analysis.

Also, I presume that the resistance/square depends also on the thickness. Is the resistance per cube (face to opposite face) also fix? If so it may be an even more canonical characteristic.