HABIST: histogram-based analog built in self test

TLDR: The HABIST method and optional variations in its implementation, algorithms for processing histograms to obtain signatures and other compressed form of data, including waveform parameters, examples of the difference histograms that result from applying the algorithm, and methods and circuits for histogram generation are described.

Abstract: This histogram based method of test collects a statistical representation of the activity at a node and processes that representation using a template histogram as a reference. In most cases, no special stimulus is required-data is collected in-situ, while the circuit under test is functioning. (Alternatively, analog stimulus, e.g. using a pseudo random sequence generator or stored digital vectors with a D to A converter, may be provided). The result of processing the data against the template histogram is a compressed human readable signature that defines gain, offset, noise, and distortion errors. These errors can then be used heuristically to determine causation. This paper describes the HABIST method and optional variations in its implementation, algorithms for processing histograms to obtain signatures and other compressed form of data, including waveform parameters, examples of the difference histograms that result from applying the algorithm, and methods and circuits for histogram generation.