<div>In semiconductor manufacturing, metrology is generally</div><div>a high cost, non-value added operation that impacts</div><div>significantly on cycle time. As such, reducing wafer</div><div>metrology continues to be a major target in semiconductor</div><div>manufacturing efficiency initiatives. A novel</div><div>data-driven spatial dynamic sampling methodology is</div><div>presented that minimises the number of sites that need</div><div>to be measured across a wafer surface while maintaining</div><div>an acceptable level of wafer profile reconstruction</div><div>accuracy. The methodology is based on analysing historical</div><div>metrology data using Forward Selection Component</div><div>Analysis (FSCA) to determine, from a set of candidate wafer sites, the minimum set of sites that</div><div>need to be monitored in order to reconstruct the full</div><div>wafer profile using statistical regression techniques.</div><div>Dynamic sampling is then implemented by clustering</div><div>unmeasured sites in accordance with their similarity</div><div>to the FSCA selected sites, and temporally selecting a</div><div>different sample from each cluster. In this way, the risk</div><div>of not detecting previously unseen process behaviour</div><div>is mitigated. We demonstrate the efficacy of the proposed</div><div>methodology using both simulation studies and</div><div>metrology data from a semiconductor manufacturing</div><div>process.</div>