Page 117 - The Art and Science of Analog Circuit Design

One Trip Down the 1C Development Road

of which specs are important, and which are not, comes into play. It
makes no sense to "over spec" a parameter that the customer is not con-
cerned about because later it could cause a yield problem. On the other
hand, it is important to spec all parameters so that any "sports" (oddball
parts) are eliminated, since they are usually caused by defects and will
often act strangely. The idea is to have all functional parts meet spec if
they are normal.

Data Sheets

The data sheet is the most important sales tool the sales people have.
Therefore it is important that the data sheet is clear and accurate. A good
data sheet is always late. I say this based on empirical data, but there
seems to be a logical explanation. The data sheet is useless unless it has
all the minimums and maximums that guarantee 1C performance; as soon
as those numbers are known, the part is ready to sell and we need the
data sheet. Of course it takes time to generate the artwork and print the
data sheet and so it is late. One solution to this problem is to put out an
early, but incomplete, data sheet and then follow it a few months later
with a final, complete one.
Analog ICs usually operate over a wide range of conditions and the
typical curves in the data sheet are often used to estimate the IG perfor-
mance under conditions different from those described in the electrical
table. The generation of these curves is time consuming and, when done
well, requires a fair amount of thought. Human nature being what it is,
most people would rather read a table than a graph, even though a table is
just an abbreviated version of the data. As a result, the same information
is often found in several places within the data sheet. I am often amazed
at how inconsistent some data sheets are; just for fun, compare the data
on the front page with the electrical tables and the graphs.
Beware of typical specs that are much better than the minimums and
maximums. I once worked with a design engineer who argued that the
typical value should be the average of the distribution; he insisted that the
typical offset voltage of his part was zero even though the limits were
±4mV. Most companies have informal definitions of "typical", and it
often varies from department to department. George Erdi added a note to
several dual op amp data sheets defining the typical value as the value
that would yield 60% based on the distributions of the individual ampli-
fiers. I like and use this definition but obviously not everyone does, since
I often see typicals that are 20 times better than the limits! Occasionally
the limits are based on automatic testing restrictions and the typicals
are real; for example, CMOS logic input leakage current is less than a
few nanoamps, but the resolution of the test system sets the limit at 1
microamp.

100