A framework for the testing and fault diagnosis of analog and mixed signal circuits using artificial neural networks

This work presents a framework for linear and non-linear analog fault diagnosis. The primary focus of the work is to provide a mechanism for robust diagnosis testing of analog subsystems is used to provide a reasonable approach to testing in mixed signal systems. This allows the partitioning of the analog from digital components.; Artificial neural networks are used to provide the diagnosis mechanism networks of reasonable dimension are shown to be capable of robust diagnosis of analog circuits including effects due to tolerances. This is coupled with a methodology for selecting fault models and test stimulus. A unique use of broad band, white noise, test stimulus moves the diagnosis of linear circuits one step closer to Built-In-Self-Test. An arbitrary level of diagnosis is achieved through proper test point selection. The approach is shown to extend to the diagnosis of non-linear circuits.; The work concludes with a look at the issues surrounding the implementation and deployment of the diagnosis mechanism. Primarily the method by which the test stimulus and circuit responses will be achieved. Two approaches emerge: analog test bus and Built-In-Self-Test. Design-for-Test (DFT) issues arise in both approaches. The additional circuit cost is balanced against test scheduling since the cost of testing can be directly related to test time. This work explores some of the DFT issues regulating the balance of circuitry versus concurrent test using a new cost model.