In semiconductor manufacturing, the implementation of advanced process control systems has become essential for cost effective manufacturing at high product quality. In addition to established process control methods, new control techniques such as virtual metrology, where post-process quality parameters are predicted from process and wafer state information need to be developed and implemented for critical process steps. This requires a fab-wide approach due to the objectives of VM, which are to supplement or replace stand-alone and in-line metrology operations, to support fault detection and classification, run-torun control, or other new control entities such as predictive maintenance. Virtual metrology is typically based on statistical learning methods, and a large variety of potentially applicable algorithms are available. A key challenge of the virtual metrology application is proving its capability to produce precise predictions even in complex semiconductor manufacturing processes. In addition, virtual metrology applications need to be implementable into the specific automation and control environment already present in the respective fab. In this paper, the approach and results of a feasibility study toward the development and implementation of virtual metrology applications in a logic fab are presented. The feasibility study was performed for the prediction of trench depth after a complex dry etch process as the specific use case studied.Index Terms-Advanced process control, manufacturing data processing, stochastic gradient boosting, virtual metrology. Georg Roeder received the Diploma in material science and the Ph.D. degree in electrical engineering

In semiconductor manufacturing, advanced process control systems have become essential for cost effective manufacturing at high quality. Algorithms for new control methods such as virtual metrology where post process quality parameters are predicted from process and wafer state information need to be developed and implemented for critical process steps. The objectives of virtual metrology application are to support or replace stand-alone and in-line metrology operations, to support fault detection and classification, run-to-run control, and other new control entities such as predictive maintenance. As virtual metrology is typically based on statistical learning methods, a large variety of potential algorithms are available. The challenge of virtual metrology application is the capability to obtain precise predictions even in complex semiconductor manufacturing processes. In this paper, the approach and results towards the development of a virtual metrology algorithm for the prediction of trench depth after a complex dry-etch process are presented