Re: Practical resistor accuracy distribution

On Sat, 27 Apr 2024 17:41:10 -0400, Joe Gwinn <joegwinn@comcast.net>
wrote:

On Sat, 27 Apr 2024 14:24:43 -0700, John Larkin
<jjSNIPlarkin@highNONOlandtechnology.com> wrote:
>

On Sat, 27 Apr 2024 16:47:29 -0400, Joe Gwinn <joegwinn@comcast.net>
wrote:
>

On Fri, 26 Apr 2024 19:38:07 -0700, John Larkin
<jjSNIPlarkin@highNONOlandtechnology.com> wrote:
>

On Fri, 26 Apr 2024 20:16:48 -0400, Joe Gwinn <joegwinn@comcast.net>
wrote:
>

On Sat, 27 Apr 2024 00:54:41 +0200, Klaus Vestergaard Kragelund
<klauskvik@hotmail.com> wrote:
>

Hi
>
I needed to look into distribution of accuracy of resistors.
>
Vendors don't have any info, but luckily some nice people online have
done measurements:
>
<https://www.eevblog.com/forum/projects/smd-resistor-distributions/>
>
<https://lambdafox.com/resistor-tolerances/>
>

The thing to be aware of is that resistor value distributions are
often wildly non-Gaussian.  The classic example is where there are
grades, say 1%, 5%, and 20%.  The 5% group will often have a hole that
happens to match the 1% distribution perfectly.
>
A lot of modern resistors are made with a relatively wide
distribution, and then machine-sorted into bins.  In this case, many
of the bins will have uniform distributions.  And so on.
>
Joe Gwinn

>
I suspect that they may laser trimmed and shipped, not binned.
>
One batch on a reel could be very close to one another.

>
Also possible, and maybe a combination of both?
>
But the fundamental point is that one must measure to be sure, unless
one is a large enough customer that the manufacturer will simply tell
you.
>
Joe Gwinn

>
We just assume that all 1% resistors are within 1%.
>
We buy 0.1% and some 0.05% parts too. Since most of our products have
software-based calibration, we mostly care about resistor tempcos, not
so much tolerances.

>
Yes, and calibration will wash all those strange-distribution woes
away.
>
Joe Gwinn

One could cal out tempcos, but test time would be outrageous.

We are bidding on one unit that might need that. Yuk. The alternative
is to, basically, ovenize the PCB.