Detecting Sudden Process Time Variations in a Semicon Fab using Non-standard Control Charts (2026-EU-PO-2840)

This poster is about tracking how long wafers – the thin slices of material used to build chips – spend inside a piece of equipment during a specific manufacturing step. Essentially, we're monitoring the equipment's process time.

Sometimes, due to maintenance or other production events, these times can suddenly change or "drift."

Our main goal was to create a statistical method that could automatically spot these process time drifts. This task is complicated because each piece of equipment can run different programs (or recipes), these programs don't all take the same amount of time, and the programs can run one after another.

After cleaning up the data and normalizing the process times based on the specific program/recipe, we tested a few different detection techniques. The method that performed best was one that compares the difference in medians using sliding windows. It proved to be effective because it's robust to extreme values (like random spikes) and can reliably detect those sudden shifts in the process time.

The JMP platform proved most suitable for validating this method and approach before its transition to a full industrialization.