The datasheet of the AD5621 states the following offset error range:

typical value +/- 0.063 mV

maximum +/- 10 mV

As this is a very broad range I'd like to know under what conditions we have to

expect the maximum value.

In our application we use the DAC output as a kind of reference voltage.

+/-0.5mV offset would be acceptable.

The ambient temperature range is between 0°C and 60°C.

Who has experience with the AD5621 and can help on that topic?

Regards

Robert

Robert,

I've been in the industry for many years and have been in a lot of data sheet meetings before product release.

There is no standardization on what "typical" means. Is it an arithmetic average? Is it an average of the absolute value? Is it the mean plus one standard deviation? I've seen all used.

Is the typical from one lot? From three lots? From "skew" lots? Skew lots are lots where some process parameters are deliberately offset. For example, with CMOS parts, a run with strong P and weak N, and then with weak P and strong N.

So now we have some numbers for mean and standard deviation. The size of the standard deviation can be greatly affect by the accuracy/precision/repeatability of the tester. E.G., it's very difficult to measure femtoamps of

bias current on an op amp. How do we set the min/max limits? This is usually done with Cpk's.:

Process capability index - Wikipedia, the free encyclopedia

Again, this can vary, because one mfgr/product line may use 1.3 for a Cpk, and another might use 1.67

or 2.0.

Some parameters can be obtained from corner simulations, high/low temperature, voltage, beta, etc.

Again, this is a judgment call so the mfgr will get 100% yield (or 99%) over every fab run for years.

Here's the bottom line: I have handled a large number of cases where the customer was in production for several years, got a new date code, and had 20-30% fallout at final board test. 95% of the time, the "bad" parts were returned for FA and found to be good parts and passed our final test tape. Digging into the problem, it turned out that the customer had designed to a "typical" number. Remember, if a "typical" is a straight average, half the parts will be better, and half will be worse. The only thing you can count on in a production environment is the min/max in the spec table. Not the numbers in the "features" section of page 1, or the typicals, or the graphs.

I have pondered for years what a customer can do with a typical number; my only conclusion is get in trouble.

If you want, we could change the typical to +/-0.6, or even +/-6mV; the situation still doesn't change.

If you need a tighter part to make your design work, you will have to test and eat the fallout that cannot be returned, calibrate at end of line, or select a different part.

Harry