Abstract: Acoustic emission samples for a chemical mechanical polishing process are acquired and analyzed using a Fourier transform to detect wafer vibrations characteristic of scratching. When excess noise levels are detected at frequencies or within frequency bands being monitored, the polishing process is halted and an alarm is generated for the operator. Such in-situ detection minimizes damage to the wafer being polished and limits the damage to a single wafer rather than a run of wafers. Polish endpoint detection may be integrated within the scratch detection mechanism.