Well as you know, only the electrical test limits (in theory) can be
trusted. The curves are guidelines. The story I was always told about
electrical limits is the customer gets the right to return a part for a
replacement if the part fails electricals. Since nobody does incoming
inspection these days, that means a lot of crap gets shipped if the
vendor has poor quality. [Note the manufacturers flow usually has a QA
test for each lot on a sample basis to insure the test hardware wasn't
fubar. QA test is probably over temperature. ]
Actual testing for commercial parts is limited to basic function, often on
wafer. Industrial parts get tested at 25 C and maybe at high temp,
usually after packaging, no real process difference except test failures
are usually dropped back into commercial bins. Military grades require
much more bookkeeping. Generally only "selected wafers" get a chance to
be Military grade.
Anyway, I see your point and the datasheet doesn't make sense. I can't
think of anything in a ceramic package that would cause it to leak more
than plastic. I assume they don't put carbon black in this plastic
package, but that could make it worse than ceramic.
Some manufacturers put goop over the chip prior to the plastic going
around the leadframe. I assume that goop has high resistivity.
The goop would have to be much less expensive that package plastic or
preserve exotic properties like ultra low leakage.
The goop is something to do with packaging in general, not low leakage.
Not everyone uses it.
Wafer test is simply to save packaging parts that will fail at final
test anyway. Nobody, or at least no place I ever worked, packages parts
and didn't test them. Plenty can go wrong in backlap, bonding,
Everything with electrical limits is tested at ATE at all grades at room
temp for packaged parts. If you look carefully, you may see GBD
(guaranteed by design). That can mean a lot of things. If it is a
capacitance at a pin, the assumption is if the wafer passed parametric
testing, then the pin capacitance will be totally predictable, hence
GBD. If the part has a reference in it. the drift at elevated
temperature in theory can correlate to the drift at cold temperature.
All the tests have guardbands. Depending on the company, some test wide
at wafer then accept the rejects at final. You do this is the chips are
expensive. That is, you are willing to spend some money packaging
borderline parts in order to get product to sell. The other scheme is to
test tight at wafer and then have looser limts as the part goes down the
test flow. That is, the test limits are tighter than the datasheet spec
at wafer. At QA, the test limits are exactly what is on the datasheet
less the bench to ATE correlation error. That is, somebody should be
able to bench test the part and have it pass publsihed electricals.
If you are not familiar with parameter wafer testing, the devices on the
test pattern have to meet test criteria before product wafer testing is
done. If something fails the parametric test, then a decision is made to
see if the parts are OK to sell in terms of reliability. That is, say
the oxide breakdown was out of spec. You would probably reject the wafer
just because you don't trust it. Put if a parameter is off on a device
you don't use (say epifet), then the wafer can go off to production