# Resistors - binning and weird distributions

I read about a little quirk related to binning resistors in a comment on this recent question.

Some manufacturers will sell, for example, 1% and 5% resistors that are really made in the same batch. When the resistors are being sorted by value, the more accurate ones are put into a 1% category and sold for a slightly higher price, and the less accurate ones are sold as 5% resistors.

This method of sorting guarantees that 5% resistors running through this process will never be within 1% of their nominal value. In other words, a 1 k$\Omega$ +/- 5% resistor would have a resistance in the range [950, 990] or [1010, 1050] - but never [990, 1010].

Does this actually happen? I guess that the parts are still what you're paying for, but it seems really odd that a 5% resistor would have 0 probability of being within 1% tolerance.

• How is it odd that a manufacturer would want to maximize profits while still giving the consumer components within spec? This is why you buy components whose worst case will work in your application. – I. Wolfe Mar 2 '15 at 15:05
• This reminds me of a story I heard long ago. A manufacturer was selling parts with a guaranteed failure rate of 1 part in 100. 100 units fit in a box. One of the customers noticed that there was always a defective part in a particular location in the box. The manufacturer was actually testing all parts, and placed one defective part in each box of 100. – JRE Mar 2 '15 at 15:25
• Because the sales guys had made the customer a lower price for the 1% failure rate. Since the manufacturer was doing 100% testing of the parts anyway (other requirements for that product and too expensive to change methodology for one or two customers,) they just dropped in one deader in 100 to justify the lower price. – JRE Mar 2 '15 at 15:30
• @Kynit to save having to throw it away and replace it with a working one. – Will Mar 2 '15 at 15:30
• @JRE: The story I heard was that an automotive company ordered a lot of one million chips from a Japanese manufacturer and specified that at most 0.01% could be defective. A crate arrived which contained, along with thousands of tubes of chips, a baggie with 100 chips. When asked what those were, the manufacturer indicated those were the "defective" ones [thus making clear that they were confident that the remaining 999,900 chips were good, and perhaps indicating that they'd felt insulted at the thought that they would ship less than 100% perfect product]. – supercat Mar 2 '15 at 17:20

The point is that it might happen.

The right way to look at this issue is to never assume anything about the distribution of error within the specified tolerance band. Always assume the distribution will be the most inconvenient for your design, because you don't know that it isn't.

Speculating on how something behaves beyond the tolerance specified in the datasheet is pointless and will only get you in trouble.

• I guess that's fair - I'm sitting here picturing a nice, smooth bell curve, but there's nothing saying that this has to be the case. – Greg d'Eon Mar 2 '15 at 15:14
• @Kynit A smooth bell curve cannot be assumed without knowing the specifics of the process. The distribution might be asymmetric to begin with, etc. It is also not very likely that the distribution's peak will fall on the desired value. That's a good outcome, but rarely achieved. When you look at the distribution of the 1% parts, you'll see that it's skewed. – Kuba hasn't forgotten Monica Mar 2 '15 at 15:20

This method of sorting guarantees that 5% resistors running through this process will never be within 1% of their nominal value.

This is a false conclusion.

For example, if 90% of the production parts fall within 1% of nominal value, but only 10% of the sales volume is for parts with 1% spec, then about 89% of the "5%" parts will still have values within 1% of nominal.

• Yeah, this has happened with microprocessors. They're tested at the highest frequency, those that fail are retested at lower frequencies. If they pass they become lower rated processors. There aren't enough of the lower rated ones to fill out the demand, thus overclocking them often works. – Loren Pechtel Mar 3 '15 at 2:27

In my experience it does not happen (resistors are trimmed to value in modern times, not manufactured then sorted), but that doesn't mean that the center of the distribution will be the nominal value nor that the distribution will necessarily look like a normal curve.

For example, on 5% resistors the trimming equipment may be set to the nominal value but the delays in the cutting mean that the value is systemically higher than nominal, or perhaps there's a small calibration error in the measuring equipment- a span shift and/or a zero shift, plus some variation between measurements. So long as the typical value is within 1-2% of the nominal they're going to run with it because the variations from unit-to-unit will not result in any significant amount of scrap.

There is usually some advantage in doing things like using 100K || 100K , 100K and 100K + 100K as MS bits in a DAC (all from the same batch), but even that is taking a chance- you have to test and be prepared to do rework. Sometimes it's worth it if parts are expensive compared to labor. Seldom so in developed countries.

Yes, it may and it does happen, but also other things may happen.

For fun I recently wrote some code to semi automatically measure and archive resistor values of 100 1% 150kΩ resistors (with some accurate enough equipment). I don't have data anymore at hand, but the values were in a distribution that looked like half a part of gaussian, ranging from 148.5Ω to 149.3Ω.

As Olin said, expect anything to happen, even to find occasionally values a bit outside of the tolerance.

I know that used to happen in the 1940s - 60s etc. I remember reading about that in my Dad's old electronics magazines such as Practical Wireless, etc.

However, to the best of my knowledge, @Spehro Pefhany is correct, and resistors are trimmed to spec since at least the 80s. It does not make sense to trim and measure in two separate steps - they are measured, trimmed and probably packed one-time.

## terminology.

"If the manufacturer sorted out the 5% resistors and tagged the remainders as 10% you would end up with 10% resistors with a bimodal distribution... Also known as a "Rabbit ears distribution" - a Gaussian with the centre chopped out" -- "10% tolerance resistor... ...what is it good for?"

Does this actually happen?

## sometimes no.

"a 5% resistor will probably between 1% and 5% (less some guard band) - if it were less than 1% it would generally be sold as a tighter tolerance ... I understand the theory behind why a company might do this but have yet to see any evidence that it's actually done. In fact, the results from the EEVblog 'experiment' ... indicate that this sort of binning was not done on the resistors Dave tested." -- Reddit: Quick question regarding resistor tolerance.

## sometimes yes.

"I measured each of the resistors using my ... multimeter ... it does seem to have two distinct "bumps" in the histogram." -- Paulo Oliveira

"In one of my school labs (~ a dozen years ago) I tested several dozen relatively low tolerance resistors and found a bi-modal distribution just within the +/- tolerance values that looked as though the manufacturer had done equivalent testing to bin them." -- Dan Neely