Page 212 - Troubleshooting Analog Circuits

APPENDIX F
How to Get the Right Information From a Data Sheet

Not All Dat8 Sheets Are Created Alike, end False Assumptbns could Cost an Engineer Time and Money
By Robert A. Peame
When a new product arrives in the marketplace, it hopefully Every year, lor the last 20 years, manufacturers have been
will have a good, clear data sheet with it. trying to explain. with varying success, why they do not mea-
The data sheet can show the prospective user how to apply sure the Zi,, per se, even though they do guarantee it.
the device, what performance specifications are guaranteed In other cases, the manufacturer may specify a test that can
and various typical applications and charactmistics. If the be made only on the die as it is probed on the wafer, but
data-sheet writer has done a good job, the user can decide cannot be tested after the die is packaged because that
if the product will be valuable to him, exactly how well it will signal is not accessible any longer. To avoid frustrating and
be of use to him and what precautions to take to avoid confusing the customer, some manufacturers are establish-
problems. ing two classes of guaranteed specifications:
SPECIFICATIONS The tested limit represents a test that cannot be doubt-
The most important area of a data sheet specifies the char- ed, one that is actually performed directly on 100 percent
of the devices, 100 percent of the time.
acteristics that are guaranteed-and the test conditions that
apply when the tests are done. Ideally. all specifications that The design limit covers other tests that may be indirect,
the users will need will be spelled out clearly. If the product implicit or simply guaranteed by the inherent design of
is similar to existing products, one can expect the data the device, and is unlikely to cause a failure rate (on that
sheet to have a format similar to other devices. test), even as high as one part per thousand.
But. if there are signifcant changes and improvements that Why was this distinction made? Not just because customers
nobody has seen before, then the writer must clarify what is wanted to know which specifications were guaranteed by
meant by each specification. Definitions of new phrases or testing, but because the quality-assurance group insisted
characteristics may even have to be added as an appendix. that it was essential to separate the tested guarantees from
For example, when fast-settling operational amplifiers were the design limits so that the AQL (assurancequalily level)
could be improved from 0.1 percent to down below
first introduced, some manufacturers defined settling time 100 ppm.
as the time after slewing before the output finally enters and
stays within the enor-band; but other manufacturers includ- Some data sheets guarantee characteristics that are quite
ed the slewing time in their definition. Because both groups expensive and difficult to test (even harder than noise) such
made their definitions clear, the user was unlikely to be con- as long-term drift (20 ppm or 50 ppm over 1.000 hours).
fused or misled. The data sheet may not tell the reader if it is measured,
However, the reader ought to be on the alert. In a few cas- tested or estimated. One manufacturer may perform a 100-
es, the data-sheet miter is playing a specsmanship game, percent test, while anather states, “Guaranteed by sample
and is trying to show an infetior (to some users) aspect of a testing.” This is not a very comforting assurance that a part
product in a light that makes it look superior (which it may is good. especially in a critical case where only a long-tern
be, to a couple of users). test can prove if the device did meet the manufacturer’s
specification. If in doubt, question the manufacturer.
GUARANTEES
When a data sheet specifies a guaranteed minimum value, NPICALS
what does it mean? An assumption might be made that the Next to a guaranteed specification, there is likely to be an-
manufacturer has actually tested that specification and has other in a column labeled “typical”.
great confidence that no part could fail that test and still be It might mean that the manufacturer once actually saw one
shipped. Yet that is not always the case. part as gocd as that. It could indicate that half the parts are
For instance, in the early days of op amps (20 years ago), better than that specification. and half will be worse. But it is
the differentialinput impedance might have been guaran- equally likely to mean that, five years ago, half the parts
teed at 1 MR-but the manufacturer obviously did not mea- were better and half worse. It could easily signify that a few
sure the impedance. When a customer insisted, “I have to parts might be slghtly better, and a few parts a lot worse;
know how you measure this impedance.” it had to be ex- after all, if the noise of an amplifier is extremely close to the
plained that the impedance was not measured, but that the theoretical limit, one cannot expect to find anything much
base current was. The correlation between Ib and Zi,, per- better than that, but there will always be a few noisy ones.
mitted the substitution of this simple dc test for a rather If the specification of interest happens to be the bias current
messy, noisy, hard-tfjinterpret test. (lb) of an op amp, a user can expect broad variations. For
example, if the specification is 200 nA maximum, there
might be many parts where lb is 40 nA on one batch (where
the beta is high), and a month later, many parts where the lb
is 140 nA when the beta is low.

I99