herunterladen
Keywords: Fab Process Standard Deviation, Six Sigma
APPLICATION NOTE 4489
TYPICALLY TESTING “TYPICAL” IN A TYPICAL
MANNER
By: Bill Laumeister, Strategic Applications Engineer
Abstract: Typical (typ) is typically the most misunderstood word in Integrated Circuit (IC) testing. Typical
values can’t be tested directly because they are a statistical value. We discuss how to read a data sheet to
understand what parameters are tested and under what conditions. Fabrication (fab) parameters are
highlighted including process standard deviation, fab process corners, six sigma quality guard bands and
defective parts per million opportunities (DPMO) concepts.
A similar version of this article appeared September 15, 2014 in CTIMES.
Circular Logic
“Is this a riddle?” you say. No. This title sounds like it is circling, real circular logic, but it makes a point.
Typical (typ) is typically (ah, yes, now I am having fun with you!) the most misunderstood word in integrated
circuit (IC) testing. There are other words to describe its concept: representative, emblematic, usual,
normal, standard, mainstream, average, mean, and conventional. Confused? In the world of ICs, typical is
defined as having the characteristics of the group of parts. Fine but, as the old English adage goes, that is
as “clear as mud.” Let’s tell a sly secret of IC testing: typical on an IC data sheet means NOT TESTED.
There, the secret is out. So why do IC manufacturers bother to state typical values? Let me explain.
Typical Values and a Range of Variety
Typical IC values cannot be tested directly because they are a statistical value. For example, it is like saying
the average height of a human adult is 5 feet 5 inches. Measuring any single person will not allow you to
establish a mean, an average, or a typical height. An anthropologist could measure the height of every race
of mankind or statistically measure a sample of the population. Then a statistician, knowing the size of the
sample, could calculate the confidence level of the average. This process is statistically the same for ICs.
An IC designer can statistically predict a typical value based on simulation test results. Once again, typical
is meant to give general guidance to the circuit designer.
1
Page 1 of 10
Verzeichnis
