A Neural-Based Analog IC Parametric Fault Diagnostic System with Discriminant Simulation Sampling

Abstract

Abstract This article presents experimental results that show feedforward neural networks are well suited for analog IC parametric fault diagnosis. It describes a discriminant sampling method for analog IC parametric faults simulation for a neural-based diagnostic system to reduce the computational overhead of the neural network training phase. Analog circuit fault simulation has not achieved the same degree of success as its digital counterpart owing to the difficulty in modeling the more complex analog behavior. Analog circuit fault simulation does rely on time-consuming circuit simulations on a large number of sampled points subject to various process deviations to emulate the actual faulty conditions. The proposed algorithm uses an effective importance sampling strategy to capture the parametric acceptance boundaries of the circuit under test. It makes use of circuit simulations effectively to relate each circuit performance to the process disturbances and to establish response functions that are accurate at the transition boundaries of the acceptance region. With the proposed approach, simulation samples can be generated for fault simulations mostly at the boundaries of the acceptance region to institute the discriminant functions for parametric faults testing using a neural network approach. The proposed algorithm is more computationally effective when compared to the Monte Carlo method.