A deep level reasoning tool for semiconductor process simulation

Summary form only given. An approach is proposed which attempts to combine features found in traditional process simulators and expert systems. The result is a system which generates process-specific knowledge from physical models by the sequential execution of a process recipe. This deep-level knowledge can be used in conjunction with techniques borrowed from the artificial intelligence community to reason about a variety of domain problems. Dependency trees are generated by the sequential execution of a process recipe. These trees can be used for identifying relationships between process steps and structures, which is useful in a problem-solving domain. The system design has been optimized for speed by using both analytical and empirical models. Typical recipes can be simulated in approximately one minute. Accuracy is achieved by providing a tuning mechanism which allows the user to input actual process data as a goal for the system output. The program described offers a logical evolution over traditional simulators and should find future applications in process control and failure analysis