Analysis of intra-level isolation test structure data by multiple regression facilitate rule identification for diagnostic expert systems

It is shown that multiple regression analysis of data from test structures designed to evaluate VLSI fabrication processes for multilevel interconnects can produce a reduced information set that can be operated on by the ID3 algorithm to produce candidate rules for a diagnostic expert system. Supplementary candidate rules that relate specific facets of the reduced information set to process conditions can be identified in a second level of statistical regression, but the authors experience has been that this is probably less effective as a stand-alone approach. The value of the expert system in the interpolation of test structure data for VLSI manufacturing support is the relative rapidity with which it can deliver diagnoses, its easily maintained knowledge base, its essentially infinite memory capacity, and the useful technical discipline it asks of those who wish to exploit these advantages.