Resistor Grading

[David Hess]

A good, used 34401A is cheap these days and will give you 6 1/2 digits.

They do look like a good deal, especially since a full service manual is available.  I was thinking that an older model would be easier to service.

[Doctorandus_P]

I am a bit curious about the distribution of these resistors.
In the old days it was common for 5% resistors, to always have a tolerance that is greater then 1%, because all <1% resistors were binned by the factory and sold as 1% resistors. But as far as I know that is not done anymore because production processes have improved to the point that they just manufacture what they need, but maybe <0.1% is still sorted this way.

[trobbins]

It looks to me like quality control of processes, on a manufacturing batch or session basis, provides quite a tight distribution around some varying mean.  Sort of akin to the control process making a slight adjustment after some quanta of parts made.  And that would infer that when a batch or session has resulted in say 0.5% then the whole batch is then 'binned', rather than individual parts.  The span of values I hand measured was 92% within +/-0.25% for 100x 402R 1% bandolier, and 96% within +/-0.18% for 200x 10k 1% bandolier.  That relates to part value variation and not an absolute value or tolerance spec, but seems reasonable for binning say parts to a +/-0.5% tolerance, when the same part has 1%, 2% and 5% spec rating options.

My initial effort on tempco similarly indicated that a statistically significant sample of tempco measurements on such parts then likely holds for all the parts on a bandolier - ie. a close distribution around a mean, rather than a scatter of pos and neg tempcos over the full range of the tempco spec.  I haven't done a sampling of tempco of the 402R parts, but they seem to be well within their +/-50ppm/C spec, albeit noticeably higher than observed for the 10k parts I just got.

[Kleinstein]

For finding matching parts it can make sense to record the actual values and than only sort out the best matching parts afterwards.
For really accurate measurements one may have to wait a few 10 seconds or so for the temperature after handling stabilizeses. So the measurement may take some time.

An impedance bridge may well have more resolution / stability than a 4.5 digit meter in ohms mode. It can also depend on how the values fits to the ranges.

Quite often resistors from the same batch already match quite a bit better than there nominal tolerance.
There are different way how resistor values are made / checked. Especially tighter tolerance tend to be individually trimmed (e.g. via laser) and the time spend there sets the accuracy / tolerace class. Lower grades may be produced in batches and only sample checked for adjusting the process for the next batch. Here batches further out may end up as high tolerance (or different value) parts.

[EC8010]

... but maybe <0.1% is still sorted this way.

I can answer that one. I measured one hundred 1M 0.1% metal film resistors and saw a Gaussian distribution - centre values were not missing. What I also found was that standard deviation was 0.032%; they were defining their tolerance at 3 sigma. But they did not define what their tolerance meant. In fact, barring a comment I saw many decades ago in a Mullard book, nobody defines what their tolerance means.

Measuring VTO for Supertex DN2540N5 depletion MOSFET, the centre values are missing.

[Alex Nikitin]

For finding matching parts it can make sense to record the actual values and than only sort out the best matching parts afterwards.
For really accurate measurements one may have to wait a few 10 seconds or so for the temperature after handling stabilizeses. So the measurement may take some time.

I use fabric gloves and try not to touch resistors, only probes (it still transfers some heat if you hold these continuously, so there is a noticeable drift for a few seconds for resistors with larger TC). For resistor measurements I find my oldest meter - HP3456A - is the best, even the Keysight 34465A is not quite as convenient. 1ppm resolution is sufficient for all but the very best resistors measurements and Offset Compensated Ohms are useful so for this application I would recommend the HP3456A over 34401A, for example. On the photo is how I grade resistors on tape, in this case the RN55C is 50ppm/C grade, so numbers are in 10s of ppm, i.e. +14 means 10.0014K measured and -14 is 9.9986K. For better grades (i.e. <5ppm/C TC) the numbers could be in ppm.

Cheers

Alex

[David Hess]

I am a bit curious about the distribution of these resistors.

Based on just looking at the number of resistors in each bin, it looks close to a normal distribution.

In the old days it was common for 5% resistors, to always have a tolerance that is greater then 1%, because all <1% resistors were binned by the factory and sold as 1% resistors. But as far as I know that is not done anymore because production processes have improved to the point that they just manufacture what they need, but maybe <0.1% is still sorted this way.

I have only read about that happening and never encountered it myself.

I am suspicious about some capacitors because the same series of capacitors can be available in 3 different tolerances.

An impedance bridge may well have more resolution / stability than a 4.5 digit meter in ohms mode. It can also depend on how the values fits to the ranges.

My manual impedance bridge would be too inconvenient if I had to measure different resistors, but in this case the values are all close together so only two dials would need to be adjusted at most, so maybe it is worth trying it.

[Kleinstein]

For just looking at matiching resistors one can often keep the bridge resistors fixed and just use the reading of the null-detector. The reading is somewhat nonlinear and the scale may not be well known. For the small relevant range to reading of the deviation is usually still OK.

[TimFox]

One can certainly calibrate the null meter for deviation from nominal value for a given expected value.
General Radio used to make such bridges with detectors to evaluate deviations.

[David Hess]

One can certainly calibrate the null meter for deviation from nominal value for a given expected value.
General Radio used to make such bridges with detectors to evaluate deviations.

If it came to that, I could make a linear null meter amplifier for my impedance bridge to drive a multimeter.