Neural network based system for testing and diagnostics of analogue integrated circuits

The paper describes a new implementation of testing-algorithm model for analogue and mixed-signal circuits with neural network evaluation. The simulation system is programmable (different fault models, input stimuli, and different analyses), it covers a wide spectrum of test methods and desired post-process techniques. It is based on the possibilities of HSPICE to manage whole test simulation including the simulations of faulty or fault free circuits on one go as a multi-sweep and file driven simulation. The simulation results (for example: current and voltage data from DC or transient analyses, parameters obtained by the measure command) are read by mathematical package, MATLAB, where these are filtered and then post-processed by means of a neural-network classifier. The approach takes account to tolerances deviations on parameters. Aim of this project is to model test techniques in order to classify their effectiveness in the sense of fault coverage (or degree of detectability). Moreover, the diagnostic results can be formulated. The system was verified on practical test task.