[sdiy] IC parts binning

[Paul Schreiber]

> But do You really think those opamps made by TI were made with a
> different DIE per marking?
>

Same die, they are screened after DIP molding but before marking (also 
called "binning").

When I was an IC designer at Maxim Semiconductor, we had many products that 
were based off a single die.
For example, we had dual and quad 12-bit DACs, each with 3 different INL 
specs (6 parts total) but they
were all the same die. The LTX testers would do a wafer probe to see if all 
4 DACs were OK or 2 out of 4, then we would package them, and then the 
"binning machine" ran the INL/DNL tests and the laser would automatically 
write the part number on the package based on the outcome of the tests.

For the precision voltage references, we had this 1 "master die" that via 
laser 'zapping' on the top metal layer, we could make *72* different part 
numbers off it :) The extreme precision references (like 3ppm drift, 0.025% 
absolute accuracy) required a second level of 'Zener zap' and testing (up to 
25 seconds per part which is a lifetime in the IC manufacturing world) and 
those are the $9 parts versus the $3 parts. You are not paying for 'yield', 
you pay for test time. The breakdown for most ICs are:

die cost = 30%
package cost = 20%
test time = 50%

If you are Maxim/TI/ADI/etc and you are running millions of parts a month, 
you measure test time in *microseconds* and the test engineers are graded 
based on average testing times (which means you trade off the number of 
tests versus your bonus versus getting fired because you didn't test 
*enough*).

Paul S.