A customer recently wanted to analyze the parametric variation of a key parameter measured at final test at the end of manufacturing. How does that parameter vary across the lots made in 2019? Their preferred chart is a normal probability plot. How hard can it be?
Actually, it’s very easy with a good software system and the plot is really very satisfying. But what’s going on behind the scenes?
In this case, 12 Lots being compared means 109 separate data files are available relating to these lots. The 109 files are a mixture of set up, first-time production test, production retest, and QA buys off sampling. Those files total about 500 megabytes of data! Oh also, the data is created in Asia and the engineer reviewing it is in the USA and the support engineer (me) is in Europe!
So the first step is correctly interpreting or “parsing” the data and collating it in a database. Then comes one of the really clever steps that is very tedious and time-consuming to do manually. At yieldHUB, we call it “consolidation”. This is where the data for rejects at first time test is replaced with the test results from retest and a consolidated or “cons” file is built. This means that each lot ends up having one summarised accurate file for the final data for each device that was tested (between 1,000 and 6,500 per lot in this case but other customers have up to 100,000 device results in a lot).
Now the analysis is a case of selecting (no need to download) these lots (or cons files) and choosing to analyze with a per lot ID grouping. The system organizes the data and gives a summary per test (over 300 tests in this case, but customers range from <10 to >20,000 tests!). Simply select the tests that you’re interested in and the plot(s) that you want and they are presented as required. Using colors and different symbols to represent the data points allows the lot to lot variation to be seen and the outliers from the tens of thousands of data points can be clearly identified.
The final step is to click share – write a description for the customer (in this case or colleague in normal cases) and share the exact analysis instantly to any authorized user anywhere in the world. It is very satisfying and very effective to see such clear analysis and it took far longer to write this blog than it did to do the analysis.