Q-YIELD
FAQ
Problems Involving Bins
 Suppose
we apply 3 tests to a die. If it fails Test A, we put it
in BinA, otherwise we try Test B. If it fails Test B, we
put it in Bin B. Otherwise we try test C. If it fails test
C, we put it in Bin C. Otherwise it is "good"
and we put it in Bin D.
Suppose we now wish to examine the reason
why a die is in Bin B. Obviously, one major reason is that
it passed Test A. If we were looking for clues as to why
the die is in Bin B, we would expect to find in the list
clues as to why it passed Test A mixed up with clues as
to why it failed Test B.
Ideally we would like to look at Test B in
isolation. i.e. Consider only those dies which passed Test
A, and therefore took Test B and either passed or failed
it.
However, die are often tested in batches, and the results
are only known at a wafer or batch level.
Our data may initially look something like
this:
| Wafer |
Bin A |
Bin B |
Bin C |
Bin D |
Etc. |
|
0001
|
23%
|
10%
|
3%
|
64%
|
|
|
0002
|
18%
|
9%
|
3%
|
60%
|
|
|
0003
|
35%
|
11%
|
2%
|
62%
|
|
|
etc.
|
|
|
|
|
|
But if we consider this data, we realize that on Wafer
0001. Only 77% actually took Test B.
The actual failure rate for Test B is 10 / (100-23), and
the actual failure rate for test C is 3/(100-23-10).
This leads to adjusted results which look something like
this:
| Wafer |
Bin A |
Bin B |
Bin C |
Bin D |
Etc. |
|
0001
|
23%
|
13%
|
4%
|
64%
|
|
|
0002
|
18%
|
11%
|
4%
|
60%
|
|
|
0003
|
35%
|
17%
|
4%
|
62%
|
|
|
etc.
|
|
|
|
|
|
 Back
to FAQ
Next page 
|
