Analyzing STDF data? Learn how test numbers improve productivity
The Standard Test Datalog Format or STDF was developed by Teradyne, Inc. and has become some sort of a de-facto standard data format for test systems. Most semiconductor yield management and analysis systems support the STDF format. The data structure was designed for efficient datalogging while testing. It is very flexible and can actually be a very efficient in terms of storage if used properly.
One of the important data fields on the STDF is the test number (TEST_NUM for the PTR, FTR and MPR records) and unfortunately it is oftentimes taken for granted by test developers. The test number is an ideal index for data analysis tools because it is stored as an integer and can be more efficient for the software when used as an index versus using the test name. For people also, looking for a test number can be faster vs. reading lengthy test names.
Test developers should make sure that test numbers are unique for a single device under test (DUT) so that data analysis software and humans can use it as a reliable reference. Also, the test number assigned to the test should not change even if some tests are skipped or added dynamically for a DUT. Even if the user simply exports the STDF data into Excel, having unique and consistent test numbers would be helpful in organizing the data for analysis.
Yield management, analysis and information systems would also greatly benefit from unique test numbers because these will have a small field (a few bytes for integer) to be used as an index. On database systems like MySQL, integer indices are much smaller vs. string indices. So if these systems can use the test number as an index, then queries can be faster and storage more efficient. The user is therefore more productive as he or she can get things done faster.
If yield analysis software cannot reliably determine a unique identifier for a test, data from two or more tests can get mixed up and can affect analysis results.
Sure, there are ways around this on the parsing side but the best approach from the get-go is to make sure the test program generates unique test numbers. yieldHUB already incorporates several methods of generating unique test numbers in case the STDF or any other datalog file doesn’t have them.