Q-YIELD™ FAQ
Should I use Lot level
or Wafer Level Data?
When examining process
data, it is possible to work at either lot (batch) or wafer
level. Which should we choose?
There is no simple answer to this question.
Suppose we have six lots, three of which are
bad, and three of which are good.
Analyst A works at the batch level and concludes
that there is not enough data for a statistically significant
result. Indeed, Q-YIELD warned of this possibility when
importing such a small amount of data.
Analyst B works at the wafer level
and projects some batch measurements onto each wafer record.
We now have 24*6 = 144 wafer records. Analyst B finds a
relationship which can explain why 72 of these 144 wafers
have a high defect density. This relationship is statistically
highly significant.
In one sense nothing is different: in each
case we processed 6 batches of 24 wafers.
The question we have to ask ourselves is this: is the evidence
from each wafer statistically independent? i.e.
was the information used to draw our conclusions based upon
the same observation or a different observation.
If the information was obtained by aggregating
data batch and then projecting it onto each wafer, then
the data was not statistically independent, and Analyst
A is correct.
If, however, the measurements were performed
on each wafer then Analyst B is correct. The results are
highly significant.
Most fab data is now collected at the wafer level, so in
most cases you should be able to work at the wafer level.
Be very wary when aggregated lot-level data is mixed
with wafer-level data. The average failure rate for
a lot, for example, has little meaning when examined at
the wafer level.
Note that there is a distinction between a lot parameter
which applies equally to each wafer in the batch (such as
which machine was used to process it), and a batch parameter
which is an aggregate (e.g. defect rate for entire lot).
The former can normally be considered independent and can
safely be used at the wafer level, whereas the latter are
only really of use when all the data involved is aggregated
at the lot level.
 Back
to FAQ
|
