Soft Fault Diagnosis in Embedded Switched-Capacitor Filters

This paper presents a scheme to diagnose soft faults in switched-capacitor (SC) filters embedded in the PSoC1 processor from Infineon. The work addresses faults that cause reductions in the values of the filter capacitors due to degradations produced by electrical stress. The diagnosis scheme employs the step response of the pass band output of the filter under test. After simple signal processing steps, the test signal is delivered to a nearest neighbor (1NN) classifier that uses a similarity measure (dynamic time warping) to compare the incoming waveform with patterns stored in a dictionary. The signals in the dictionary are obtained from the filter's step response (using its transfer function) under fault-free and faulty conditions. The diagnosis characterization procedure consists of evaluations at different abstraction levels, including transfer function level simulations in MatLab (over a wide range of faulty conditions), SPICE level simulations, experimental fault injection, and on-chip signal measurements using the internal resources of the processor. All the simulations consider non-ideal effects like noise, jitter, thermal drift, capacitors mismatch, and inter-chip variation in offset voltages at a reasonable computational cost. The evaluations, performed at different abstraction levels, show an excellent performance of the method for diagnosing the addressed faults.