Bimodal error distribution?

From: Paul Koning <pkoning_at_equallogic.com>
Date: Mon Aug 2 14:26:17 2004

>>>>> "charlesmorris" == charlesmorris <charlesmorris_at_direcway.com> writes:

>> Since actual error distributions on crystals are bimodal, not
>> Gaussian, you would normally get just about that error.

 charlesmorris> This sounds interesting. Any ideas why?

Sure. Resistors are the same for the same reason: the manufacturing
process trims them, and when the measurement says the thing is in
spec, the trim stops. So you tend to get values close to the
tolerance limits.

Another mechanism that produces this: if you sort devices into several
bins according to how tight the tolerances are, you end up with a hole
in the middle of the distribution because all those parts go into the
tighter tolerance (higher price) bin.

          paul
Received on Mon Aug 02 2004 - 14:26:17 BST