Abstract
Semiconductor industry and its products change fast with time. There are about 300 to 500 processes in semiconductor manufacturing, and there are many factors affecting each process. Therefore, it is hard to estimate the input-output model for one process, and also hard to estimate the process disturbance distribution or model. However, in most of fabs in semiconductor industry, they use EWMA controller to control the process, which cannot adjust the control parameter with process disturbance. This research considers the situation that it is hard to estimate the process disturbance model, and propose a framework for the Dynamic Adjusted Proportional-Integral (DAPI) controller. By adjusting the control parameters lot by lot, we can effectively solve the problems mentioned above. And we will illustrate an empirical study of critical dimension control and overlay errors reduction of photolithography, both consider the time series disturbance model and real process disturbance. We will compare the output result with traditional EWMA (Exponentially Weighted Moving Average) controller, and show that DAPI controller can make the process closer to target and reduce the process output variance than EWMA controller. In addition, in real semiconductor manufacturing situation, there exists process sudden Shift and metrology Delay problems, which make the controller performs worse. Since DAPI controller adjusts input value slowly, it will not be affect by process sudden Shift. We will show that DAPI controller can solve those problems, and can be used to a real semiconductor manufacturing situation. Finally, in the real process disturbance situation, for critical dimension control, DAPI can improve 10% on Process Capability Ratio and improve 11% on process standard deviation. Also, for overlay errors reduction, DAPI can improve 16% on Range and improve 11% on root Mean squared error. For all scenarios and two real problems, DAPI can perform better than EWMA controller.