Characterization and control of a chemical vapour deposition reactor.

Much work has been done on developing mechanistic models of Chemical Vapour Deposition (CVD) reactors, however the use of these models is often cumbersome and computer intensive. Equipment specific empirical models are simpler to use and develop yet provide good representation of the process behaviour. In this work, empirical models were developed to characterize film thickness of an industrial CVD reactor used in the semiconductor industry to deposit polycrystalline silicon films on silicon wafers. The goals were to reduce variability in film thickness over time and space by appropriately choosing operating conditions and implementing more effective process monitoring and control. Spacial variability in film thickness is measured across a wafer (within-wafer) and along the length of the reactor (cross-load). From these measurements, film thickness variance is calculated. The predictor variables for the model were the controllable reactor inputs: centre temperature (T$\sb{\rm c}$), temperature difference between the front and centre of the reactor (R$\sb1$), temperature difference between end and centre of the reactor (R$\sb2$), pressure (P), silane flow rate (F$\sb{\rm SiH\sb4})$, and deposition time (t$\sb{\rm dep}).$ A central composite experimental design was used to generate the data for the models. Transformation of the response variables was required for the cross-load variance and within-wafer variance models to overcome model inadequacy. Optimization of the predictor variables to keep film thickness on target and reduce within-wafer and cross-load variances was achieved using a multi-objective optimization routine. Batch-to-batch variability was reduced by more than 50% through a proposed change in the operation of the reactor. Statistical Process Control (SPC) charts for mean thickness of a batch, within-wafer variance and cross-load variance were developed to monitor the process