Test structure data classification using a directed graph approach

Abstract

Directed graph techniques are introduced, serving as an expert system rule generator by classifying selections of tested wafers into groups based on similarities of the spatial distributions of their parametric test structure measurements. A self-normalizing equivalent vector inner product is devised to accommodate the ternary nature of the DC parametric test. It provides for test results that do not definitively pass a self-validation test. An algorithmic feature for avoiding special cases of nonoptimum search termination is conceived and implemented. The rules can be used to supplement those derived by other means of diagnostic process analysis, work-in-process wafer screening, and yield and reliability management.<>