A method of online predicting maintenance of an apparatus is disclosed. Using an
optical emission spectroscopy (OES) positioned on the apparatus and the change of
emission spectrum intensity detected by the OES in the process, according to the detected results, measuring the parameter in the process, the function relation between the process parameter and spectrum intensity is acquired. A control threshold is decided by the
processing requirement to the apparatus. When the parameter exceeds the control threshold, maintenance to the
etching apparatus is engaged in order to avoid
processing error caused by frequent shutdown or deficient maintenance which is estimated by experience, and hence decreasing the cost and increasing
processing efficiency of substrates (such as
silicon wafers) without changing apparatus and adding other online sensor, and improving
production rate by avoiding waste substrates caused by
error processing results. The method is suitable for
semiconductor substrate
etching maintenance of the apparatus and also other maintenance of the apparatus.