| The modern semiconductor industry is currently advancing into the world of sub 0.1-micron technology. This thesis aims at developing advanced methodologies for efficient and accurate diagnostics of process and equipment variations.;A furnace system is a typical dynamic system whose physical model can be constructed based on our understanding of the thermal systems. In this thesis we describe how to choose a linear dynamic model to approximate the real system, how to estimate model parameters when system information is only partially available, and how to detect variations of system parameters using a statistical model-classification approach.;Compared to furnace systems, photolithography processes are much more complex. While complicated physical-mathematical models of lithography are available in certain commercial packages, the prohibitively large computation required makes them infeasible for real-time applications. In this thesis, we explore replacing the complex first principle models with simpler empirical models. Due to the high dimensionality and high non-linearity of the problem, a simple mapping usually does not exist between the Critical Dimension (CD) profiles and input recipe parameters, which is desired for diagnosing input parameter variations. Two different approaches are proposed to solve this problem and they may be used complementarily in different situations. The first approach is to construct an explicit inverse model using statistical modeling techniques. The second is to build a library of input-output data pairs and, during diagnostics, search for a candidate-solution set whose statistics are used to calculate the final solution. The two approaches are evaluated and compared on computer simulation results. Finally, time series models are considered to enhance the diagnostics of input parameter variations. The key contribution of this work is that it provides a computationally tractable modeling and diagnostic framework for lithography processes. (Abstract shortened by UMI.). |
