# What is the relevance of “typical values” in datasheets?

When working on a design I always work with minimum/maximum values from the datasheet (whichever is worst case), never typical values. I was reminded when in another discussion the leakage current for a BAS416 diode came up: 3pA typical, 5nA maximum. That's a factor 1000! In this case I surely would dismiss the 3pA.
What's the relevance of "typical values" in general? Do you use them in a design?

-

One thing to bear in mind when reading technical datasheets, is they are finalised by marketeers!!! not engineers. If it was up to engineers to publish datasheets you would only have the min/max relevant values and more information about the statiscal variations that you can expect.

Many people are involved with the writing of the datasheets, and ultimately it is people in marketing that have the final say!

So when you read a typical value that is so far away from the min/max values that is simply marketers doing there thing. Generally when doing a parametric search typical values are what are listed and get you in to review the datasheet. Latter on you discover that 3pA is a maximum of 5 nA, sometimes latter in the design process!!

I will generally review the min/max values to really appreciate the range and perform worst case calculations/montecarlo analysis to really work out the expected performance!

-
Often, the only useful information on the first page is the part number - The rest is marketing mumbo-jumbo. –  Kevin Vermeer Dec 5 '10 at 19:51

The word typical is "content free". Real statistical estimates require at least a range, and a confidence. Better yet, a density function.

For instance, quantity X falls between 3.7 and 4.3, with 95% confidence means something, whereas quantity X is typically 4.0 is absurd. What does that mean? What is the probability that X lies between, say, 3.999 and 4.001?

If there is any earnestness in such a "typical" claim, the interpretation should be like this. Since the 4.0 is given as two significant figures, it means that there is some high confidence (like 95%: two standard deviations, or 99.7%: three standard deviations) that it will not fall bellow 3.95, which would cause it to be rounded down to 3.9, nor rise to or above 4.05, which will cause it to be rounded off to 4.1.

That is to say, if, say, 95% or more of the time a parameter measurement, when rounded to two significant figures, does not show 3.9 or less, nor 4.1 or greater, then we have justification in claiming that it is "typically 4.0" (but not necessarily that it is "typically 4.00").

I don't know of any source of assurance that datasheets apply this sort of standard to "typically".

-

You may optimize for the typical case. For example if I need 5 A max but 2 A typical, I will design/buy a 5 A supply but choose the one that is most efficient at 2 A.

-

Often when companies spec those values they have not done that much work determining it. I have talked to many companies that their specs are purely off of simulation of what variances they will have in production.

If you are making a million phones, check the spec on what you are buying, determine what is feasible. if you have to trash 1 phone out of 1/1000 because you get a perfect storm of devices being out of spec, that is probably much better than manufacturing problems. If the diode is almost always worst case, get a different diode.

If you are making 3, fab them for typical, allowing some tolerance, and if it ends up having a major variance in one part, replace it with another. If it is mission critical part, check it first.

If over-spec is not an issue for your 3, just spec worst case. You will spend more on components then you need, but you are building three, 10 dollars only makes 30. No real cost lost. If you are making millions, you need to confirm conformance yourself.

-
Someone downvoted me. I could see many reasons someone would disagree with me, although I would like to know which one they chose. –  Kortuk Dec 4 '10 at 23:34