Bachelor thesis

Automated Process Family Identification and Anomaly Detection in Semiconductor Manufacturing

Bachelor thesis with the Big Data Engineering (DAMS) group at TU Berlin, in cooperation with the Ferdinand-Braun-Institut (FBH). Currently under supervisor review.

Semiconductor wafers move through long, ordered sequences of process steps. In low-volume research production these routes vary widely, which makes later analysis hard: comparing two wafers is only meaningful if they went through comparable routes in the first place.

I built an end-to-end pipeline that reconstructs each wafer's route from row-level event data, represents routes by short contiguous fragments, clusters them, and then challenges every resulting group against a stricter whole-route edit-distance test. Groups that survive are treated as candidate route families; wafers that sit near a boundary are surfaced as cases for expert review rather than silently reclassified.

The emphasis throughout is interpretability and auditability: every acceptance decision is traceable, the validation is adversarial by design, and results are reported in a form that does not expose the underlying confidential data.

Techniques: Python · SQLite · sequence mining · clustering · edit-distance validation