Data cleaning seems easy, but there are other areas as well. Data sources were often not designed for easy post-collection engineering analysis. There are data translation errors, freeform entry and/or human errors, as well. Classification of fails are critical. Errors can be made in root cause. There is need to reviewing the historical data and re-classifying.

There are stress tests and leakage measurements as well. Data can be cleaned by construction. There is post-wafer probe outlier analysis, or test data for specific tests analyzed and the outlier die are downgraded. There are also proactive checks on data integrity.

There can be ML-based data trigerring. Human and machine resource allocations can be done based on test utilization data. Data is hard to gather due to lack of test naming standards.

Data democratization is next. We need to make data and tools available globally. Much data is open access. Some data is restricted by business unit or customer. Small portion of the data is available on a need-to-know basis. Customizable tools can be supplied to facilitate reporting, engineering analysis, etc. There is need to make the data available, wherever possible, to give people the right tools to analyze it.

There is need for mining the manufacturing data for downstream impact. Most fab equipment is monitored continuously. Signal-to-noise ratio can be a challenge, but strides are being made. Aging, excursion events can impact the product quality.

There is specification compliance matrix coverage tracking. There is analog test content, largely spec-based. Product plans, execution are captured in the spec management system. TI is working to audit the test plans vs. test production. There is a Bosch/Teradyne specification to test management system. The vision is tighter integration from design to test. The other vendors have similar goals, including TI.

We need excellent coverage and quality that requires collaboration across all the disciplines. We also need advanced analytical methods.