Q-YIELD FAQ

Problems Involving Bins (2 of 2)

It might at first appear that by using these adjusted figures we can analyze the variability in test B without concerning ourselves with the result of test A. At least we are now considering populations of equal sizes when looking at the variability of test B. However, this forgets the contents of the remaining columns of data. If these represent total or average data across all the die in a wafer, or properties of the wafer as a whole(which is the normal case), then these have still not been corrected for the bias introduced by test A.

Let's consider a simple example.

Suppose we have two pieces of equipment X and Y. Die from wafers processed by X have a 60% chance of failing Test A. Die from wafers processed by Y have a 40% chance of failing test A. All die have a 10% chance of failing Test B.

If we now look at the population failing Test B and compare it with the general population, we find:

Processed by X Processed by Y

General Population
50%

50%

Failure rate at Test B
4%

6%

Failure Rate of
die takingTest B
8%

12%

On the basis of Test B it appears that equipment X produces superior results! It's 1.5 times more likely that a die processed by X will pass Test B. But obviously this is not the case. The population bias introduced by Test A is manifesting itself in Test B, and in fact Test B is not affected by the choice of equipment.

This was a simple example where the effects of bias were obvious. In a more realistic example the effects can be more subtle.

In summary, when the testing of a die is conditional on the results of previous test, we typically introduce some form of bias into summary data. A test has to be viewed in the light of any biases introduced due to previous tests, and reasons for failure of a particular test will generally be mixed with reasons for the success of any previous tests.

Back to FAQ Previous page