Run-by-run process control: performance benchmarks

The run-by-run (RbR) controller provides a framework for controlling the manufacturing process which is subject to disturbances such as shifts and drifts as a normal part of its operation. The RbR controller combines the advantages of statistical process control and feedback control to provide accuracy and flexibility of control that cannot be obtained by using either method alone. The RbR controller was applied to the control of thickness mean and uniformity of a previously optimized epitaxial silicon deposition process in a radiantly heated barrel reactor in a production setting. The results were benchmarked against historical data obtained with standard process control methods from the same reactor. The gradual mode of the RbR controller reduced the variation from target (as measured by standard deviation) of the thickness mean by a factor of 2.7 as compared with the historical data. Similar results were obtained for uniformity control. The rapid mode controlled the switch of the thickness mean among three different specs of eighteen runs successfully; only one run deviates from the target spec by a magnitude barely greater than the background noise. The performance of the RbR controller with imperfect process models and process noise is also examined