how to hand match resistor to tighter tolerances, <1%?

[AndrewT]

Originally Posted by star882

You definitely cannot guarantee 31 bits of accuracy but you can get the precision.

31bit is better than 1 in 2billion or <0.5ppb

[ratza]

...and that means noise problems. LOTS of.

[star882]

That's why you need to take a lot of samples and average them.

In fact, by adding a controlled amount of "noise" to the input of an ADC, it's possible to increase the effective resolution at the cost of much increased acquisition time.

[Hendrixon]

Originally Posted by AndrewT

Hi,
it depends on the values of the resistors you are matching.
A 3.5digit DMM that reads down to 200.0ohms can match resistors greater than 500r to better than +-0.2%
from 51 ohms to 470 you can get better than 0.5% matching.
At lower resistances, or for closer matching, you need to modify the measuring procedure.

Connect a string of resistors together in series. Ensure the leads on each are of similar length. Yes, it is that sensitive.

Pass a current through the resistors so that each resistor sees a voltage drop of 195mVdc.
Now set your DMM to 200.0mVdc.
Measure the voltage drop across each resistor. Note the voltages.
Select matching resistors and group them.
It's that easy to get better than 0.05% matching.

A suitable CCS can be made from a 317 with a (switchable) fixed resistor in the output lead. If you can feed the CCS and resistor string from a variable regulated supply, then even better. It avoids changes in the CCS current with power dissipation.

I like that one a lot!!!
It's easy and fast to get 0.1% match sets.
As you said, 0.05% is also at a "cheap DMM reach" (but it does need a stable current/voltage source for that one).
Untill now I used Rob Elliot's way (Wheatstone bridge) to find matching sets:
48V Phantom Feed Supply for Microphones
With that one I could match up to 0.01% (and less!) with just an el cheapo DMM... but for these toleranses it takes time for the bridge to balance each time (about 20 seconds!) since it's wayyyy too sensetive even for tiny heat changes.

That "resistors string" trick suits my needs

Thanks.
Btw, do you have a "capacitor string" trick as well?

[tomchr]

Originally Posted by star882

It's possible to build a high precision current source with an opamp, MOSFET, precision resistor, and precision voltage reference.

Please run a sim of this where you plot the output current versus temperature, supply voltage variation, and part-to-part variation. Take the tempco of the various components into account. Then tell me that this is still a "high precision current source". Better yet. Build it and test it in the thermal chamber of your lab. Sorry! I'll have to wave the BeeEss flag on that one... It CAN be done, but it is nowhere near as simple as you indicate above.

Having spent a couple of years in the precision amplifiers group of a major semiconductor manufacturer, the majority of this time as the lead design engineer on a precision op-amp product that, among other specs, guaranteed Vos < 5.0 uV and TC(Vos) < 20 nV/deg C, I'll certainly be the first in line to testify that achieving anything "precision" requires paying a lot of attention to every detail. This includes counting every thermocouple in the test path. Think about it... Every time you have two dissimilar metals joining, you have a thermocouple. Bond pad to bond wire; bond wire to lead frame; lead frame to lead frame plating; lead frame plating to PCB trace plating (gold or tin); PCB trace plating to PCB trace (usually copper); etc., etc. I'm not arguing that it cannot be done. After all I did it... But it's not trivial at all!

If you are interested in board-level precision circuitry from a practical perspective, I highly recommend Walt Jung's book, "Op Amp Applications Handbook". Especially, I found chapter 7 "Hardware and Housekeeping Techniques" to be very insightful. I'm not affiliated with Walt Jung or the company he works for.

As for measuring voltage, a high precision ADC would get a higher resolution than most multimeters. (A good one with 4096 "counts" is only 12 bits, and precision ADCs are commonly available to as much as 31 bits!)

Please look at the ENOB, INL, DNL specs. You'll notice that just because the ADC spits out 31 bits, doesn't mean it's delivering 31-bit performance. Also, the "many-bit" converters (anything above 12-14 bits) tend to be delta-sigma converters. While they do provide good performance, they're not without their own slew of issues.

To get back to the original topic - matching of resistors... I did this recently myself using a 6.5 digit DMM. As long as you're looking for higher value resistors, this is pretty feasible without using kelvin connections - though, you probably won't get the full 6.5 digit precision. But I was aiming to get within +/-0.5 % of 100 kOhm so I could live with that. Within about 10 minutes I had the pair I needed. One thing I found, though, was that the distribution of the batch of +/-5 % resistors I had was rather skewed. Most of the resistors were about 2.5 % low. It certainly wasn't a random, normal distribution centered around 100 kOhm, though, with the small sample size I tried (maybe 30 resistors), I guess I can't make the call from a statistical stand point. I just found it "interesting"...

~Tom

[forr]

I use a 5 digits Metrix DMM and get quite easily 0.3% precision.
A bunch of resistors in series using a breadboard feeded by a voltage or current stable source and getting the voltage across each of them using DMM should be both fast and more precise.

[Conrad Hoffman]

Tom brings up an interesting point. I've matched a lot of resistors and usually buy bags of 200 from Digikey. One would think the distribution would be random, but I've gotten some rather skewed results. If I wanted not only matching, but a particular value, it could take a huge number of resistors to do it. Far better to just buy what you really need from Vishay or similar.

It's a decade old, but I did get permission to reprint my old Mini Metrology Lab magazine article on my web site. Definitely not the last word, but it might provide some useful info.

Best,
Conrad

[tomchr]

Digikey (and others) sell 0.1 % resistors. They're actually not outrageously expensive. If that provides enough precision, you're set. If not, then you need to ask yourself if you're doing something wrong or if you really need that level of precision. A lot of times I find that I can design quite precise circuits with imprecise components by selecting a circuit topology where the precision depends on the matching between components rather than their absolute value. And should absolute accuracy be required, one can always add a trimpot. Though, I bet the tempco of a trimpot is significantly different from that of a standard metal film resistor so you might be shooting yourself in the foot if your circuit needs to operate within a large temperature range.

Someone mentioned scraping resistor material off to increase the resistance of a fixed resistor. I've tried that as well. Works great. I think the resistors were carbon film (el-cheapo 5 % type). I wouldn't use this for anything in mass production, but for a quick demo - sure why not... Just seal the resistor with some enamel so it doesn't soak up water. But I have no idea what that would do to the tempco (probably not much) or long-term stability (probably throw that right out the window).

~Tom