Design of robust test criteria in analog testing

Abstract

Test design of analog circuits based on statistical methods for decision making is a topic of growing interest. The major problem of such statistical approaches with respect to industrial applicability concerns the confidence with which the determined test criteria can be applied in production testing. This mainly refers to the consideration of measurement noise, to the selected measurements, as well as to the required training and validation samples. These crucial topics are addressed in this paper. On exploiting experience from the statistical design of analog circuits and from pattern recognition methods, efficient solutions to these problems are provided. A very robust test design is achieved by systematically considering measurement noise, by selecting most significant measurements, and by using most meaningful samples. Moreover, parametric as well as catastrophic faults are covered on application of digital testing methods.