of tolerance and accuracy

[stickjam]

Working on a design requiring bipolarizing the output a multiplying DAC, the application notes indicate the use of very high precision resistors, this got me thinking about component tolerance ratings and accuracy.

Taking tempco and other environmental factors aside, conventional wisdom and statistics would say that if you needed a resistor whose value was within 0.1% of 1Kohm; by testing a large batch of 1K resistors rated at 1% you might be able to find around 1 out of 10 of them falling within the 0.1% accuracy range.

However, considering that a manufacturer offers several categories of 1/4watt metal film resistors in different tolerances of 5%, 2%, 1%, 0.5%, 0.1%, what are the chances that a 1% resistor is also guaranteed to be off by at least 0.5%?  (a 5% resistor will always be off by at least 2%, etc.)

Is this a reality in resistor/capacitor manufacturing?  If so, are any specific brands and lines better than others in this regard?  Just asking because I'm in a restocking mood.  ;D

 

[abbey road d enfer]

You take a bunch of 5% resistors and measure them. You will probably find one out a hundred that measures between 999 and 1001, which is +/- 0.1%. But the manufacturer does not guarantee it (neither do I  ), and probably the tempco will be such that holding it in your hand will put it off by 10 ohms.
Another scenario is that you find none close enough to 1k, but all of them centered tightly by a few ohms around let's say 1030 ohms. This is because manufacturers have integrated in one parameter (tolerance) two issues: precision and accuracy. In fact their precision is much better than 5%, it may be 1% actually, but accuracy is off by maybe 2-3%.
A manufacturer can guarantee +/-5% by strict control of the mfg process. But 1% and better is achieved by actual measurement and selection. This is true for carbon or metal film. Some high precision resistors use a different process that allows trimming in the course of mfg (wire-wound or laser-trimmed film).
The problem with precision resistors is their cost: you pay a lot because it's not mass-production.
Where do you need high-precision:
Input CMRR: you need at least four high-precision res; in fact if you use 0.1% res, you can only guarantee 48dB CMRR. Cost $2. And then for HF CMRR, you need two 0.1% caps. Doesn't exist.  So you shortcut with 1% caps (only good for 34dB CMRR @HF). Cost another $2.
I say, use 1-2% res and a trimmer, 10% caps and a trimmer: total cost $1.50. Labour? Irrelevant for DIY. Less than $1 for a boutique mfgr. Mass-production? Have it made by robots, Chinese, or change the specs!  ;D
Other applications may really justify the use of high-precision components, but in my case, in 37 years of design and manufacturing, I swear I've never used anything else than 5% res and 10% caps. All things that needed more precision, I have put trimmers or provision for multiple caps. I have also spent a lot of time designing procedures and test gear that allowed positive calibration and trimming. I must say mfg quantities were typically 100-500. Not really mass-production.

 

[JohnRoberts]

I don't know for a fact but i would be surprised if 1% resistors are hand selected by manufacturer's. 

I would not be surprised to learn that extremely tight tolerance parts are measured and computer trimmed with laser or whatever while still on the production line, if the process can't be managed tightly enough without human touches.

5% resistors I bought from ROHM ( a japanese company) were typically 2% or better back in the '70s. I know there were cheaper 5% resistors that lived up to the tolerance marking.

JR

 

[stickjam]

This question came to mind when thinking of how the semiconductor industry sorts CPU chips: all the performance levels of the same family come from the same design, die and wafer.  When the finished chip is tested, the clock rate is ramped up until some diagnostic begins to vary from ideal.  Chips that start to fail or drift off-spec earlier are stamped with a lower recommended clock speed or feature set and sold at a lower price than the fewer ones that passed all diagnostics when run at higher rates, thus selling for more.  Many folks overclock their PCs because a CPU rated for 1.8GHz because it dissipated a few extra milliwatts over spec may run very happily at 3GHz with a water cooler.

I just wondered if whether this same kind automated testing and sortation practice extends to the passive component realm.  If multiple tolerances, say 5%, 2%, 1% came off from the same manufacturing line, one could reasonablly expect that *none* of the 5% resistors would ever be accurate to within 2%.   

Just contemplating the omnipresent evilbay 50-value assortments to refill the experimental stockpile, versus going for a more targeted and varied assortment of the KOA Speer 1% 100PPM metal films from Mouser.  After all, I always seem to need one of those oddball standard values like the 51's, 75's or 91's

 

[Rochey]

I just wondered if whether this same kind automated testing and sortation practice extends to the passive component realm.  If multiple tolerances, say 5%, 2%, 1% came off from the same manufacturing line, one could reasonablly expect that *none* of the 5% resistors would ever be accurate to within 2%.   

This is how it was explained to me in college. (but that was 10 years ago.)

 

[stickjam]

I just wondered if whether this same kind automated testing and sortation practice extends to the passive component realm.  If multiple tolerances, say 5%, 2%, 1% came off from the same manufacturing line, one could reasonablly expect that *none* of the 5% resistors would ever be accurate to within 2%.   

This is how it was explained to me in college. (but that was 10 years ago.)

That's what I was afraid of.  I have a hunch that since resistors below 1% tend to have a different look (numbers instead of color code) that a batch of 1%ers would be most likely to contain resistors with the closest to dead-on values.

Maybe I should order a huge quantity of 5%, 2% and 1% of the same value from the same manufacturer, along with three .001% resistors to round out a Wheatstone bridge .  Then I'd keep my daughter busy on an experiment in statistics.  If they're graded through automated testing, I'd predict the 1%'s to be distributed in a single cluster of values, with the 2% and 5% resistors value distributions showing a narrow and wide hole in them, respectively.  On the other hand, I'd have to interface the bridge to a Nintendo DS to hold her attention that long.  :

 

[PRR]

> requiring bipolarizing the output a multiplying DAC, the application notes indicate the use of very high precision resistors

For why?

If you use a 0V-5V DAC and need +5V/-5V with half-count EXACTLY at zero, yes, that's precise.

But if you are making audio, even if you "need" zero-reference, that can often be 10mV or more away from zero, because we can always block it with a cap.

Any time an AUDIO puzzle seems to require 0.1% 100ppm parts, there's usually a better way.

I remember when 10% resistors were never within 4%. But I don't think 1% are made the same way as 5%.

Carbon-film is hard to compute but can be very repeatable. Carbon-composition was plagued by hasty grinding and mixing and cooking; film goes right to a turbulent gas so all parts in a pot come out the same. I kinda suspect they no longer measure and sort: if the first few samples meet spec, probably the whole pot does, and not running every part over a meter sure saves money.

Assuming you can make millions of parts to get benchmarks, accuracy comes down to the consistency of your gas and valves. So you probably have a big pot burning street-gas and hand-valves making cheap parts, and your best workers on new plumbing with a gas-checker making the more precise parts.

 

[stickjam]

> requiring bipolarizing the output a multiplying DAC, the application notes indicate the use of very high precision resistors

For why?

If you use a 0V-5V DAC and need +5V/-5V with half-count EXACTLY at zero, yes, that's precise.

But if you are making audio, even if you "need" zero-reference, that can often be 10mV or more away from zero, because we can always block it with a cap.

I wouldn't be that anal either except the thing I'm breadboarding is a piece of test gear - a microcontroller-based function generator to replace that Quick-n-Dirty one I accidentally toasted with B+.