Detection of process endpoint through monitoring fluctuation of output data

Abstract

Progress of a semiconductor fabrication process is monitored by detecting data output by the process, and then correlating a specific process event to fluctuations in the output data over a time period of 10 milliseconds or less. In one embodiment, endpoint of a plasma chamber cleaning process may be identified by calculating standard deviation of intensity of optical chamber emissions based upon a local time period. The time at which standard deviation of optical emissions attains a steady state indicates endpoint of the cleaning process. Another approach to characterizing fluctuation is to perform a Fast Fourier Transform (FFT) on the output emissions data, and then to plot over time the total power of the emissions over a relevant frequency range. The time at which total power attains a steady state also reveals endpoint of the process. Other techniques for characterizing fluctuation for process monitoring include calculation of the root-mean-square or entropy of an output signal.