A developmental expert system for test structure ata evaluation

This paper describes a developmental expert system, rule generat~on techniques, a test chip, data handling methods, and statistical data reduction techniques for characterizing the performance of a 1-pm lithography process. Examples of test re sults and an expert system diagnosis are given. Introduction Comprehensive electrical testing of complex. test chips often results in h g e quantities of data which contain detailed i n f o r ~ t ~ o n about a process. Very often, the volume of these results places them beyond the ability of the user to d i l y tively interpret. When accurate conclusions cannot be made in a relatively short time period, the data from these tests are often ignored and potential information provided by the testing is lost. This paper describes recent work in developing an expert system that can interpret large quantities of test results from test chips and p d d e the user with an English-hguage explanation of various process problems. At this time, the work is limited to developing, ~ m ~ r ~ ~ g , and testing a test chip and an expert system needed to characterize the performance of Ipm l ~ t h o ~ a p h y processes (0.5 pm 4 pm). A major emphasis of the work is the development of techniques for determining rules for the expert system knowledge base from test results. Expert Systems Expert systems emulate the problem-solving processes of human experts [I]. The information used in the emulation is stored in the form of several E-THEN rules which together constitute a knowledge base. The rules imply that IF a certain condition exists, TIHEN a certain specified action occurs. Rules may be assigned a confidence %&or which represents the total certainty that the rule is valid. The order in which the rules axe executed is controlled by the inference engine in order to reach the expert system’s goal state. A goal is the result the system is attempting to obtain, given values for parameters used by the rules. For this work, an expert system shell known as Tersond Consultant Plus” [Z] based on the EMYCEN inference engine 131 was used. The main function of an expert system is to allow a user to obtain information or guidance in an area in which the user is not an expert. For this work, the major function of the expert system is to provide the user with a diagnosis based on results from sufficiently large quantities of test structure data that would make meaningful or timely analysis otherwise impossible. The output of the system is an English-language set of statements identifying diagnosed process problems at the time the test structures were fabricated. An illustration of an expert system is seen in Figure 1. -wwi+t---! Inference Engi e j‘ Knowledge 1 Knowledge Engineer Explanation Expert Generator Figure 1. Diagram of an expert system.