Application of analytic models to computer-aided integrated circuit device diagnosis and parametric test selection

Electrical measurements made at the end of an integrated-circuit fabrication process are used for a number of purposes, including wafer qualification and, if a problem is found to exist, as a starting point for process diagnosis. Because complex relationships between electrical measurements and process conditions can make determination of a correct diagnosis very difficult, knowledge-based computer tools are sometimes applied to assist the engineer in this task.; This dissertation describes a computer-aided diagnosis system called MERLIN (MEasurement ReLational INterpreter). The system aids in the "device diagnosis" portion of the diagnosis task, which involves determination of device structure anomalies from electrical measurements. Distinguishing MERLIN from other diagnosis systems is its knowledge representation. Unlike most systems, which represent their knowledge in the form of "expert rules," MERLIN's knowledge is centered around a symbolic representation of analytic models. These models represent relationships among measurement and device structure variables in terms of well-known equations commonly employed by device design engineers. The models are compact, well-defined, and easy to acquire, making their entry and maintenance much less tedious and error-prone than expert rules.; MERLIN aids the diagnosis through algorithms developed to reason with the models and return a diagnosis, and a graphical display which describes the relationships between the measurements and the device structure. Of particular importance is MERLIN's ability to reason with the common situations where sufficient information is not available to make an unambiguous diagnosis. In such a case, MERLIN presents the most likely candidates in returning a diagnosis, and through the graphical user interface, aids in selection of additional measurements necessary to resolve the ambiguity.; Operation of MERLIN is described, including the model representation and the algorithms for reasoning with the models and returning a diagnosis. Results of some specific diagnoses are also described to illustrate the operation of the system.