Digital Spectral Analysis for Mixed-Signal System-on-a-Chip Testing and Verification

Rutgers University New Brunswick, New Brunswick NJ

Investigators

Abstract

We investigate a new application of the theory of signals and systems to test and verification of mixed-signal systems-on-a-chip. Both digital and analog input signals are considered as a set of time-varying waveforms, characterized by a correlation matrix. The matrix elements are the auto-correlation and cross-correlation coefficients. Auto-correlation indicates how much a signal resembles its prior values in time. Cross-correlation indicates how similar or dissimilar two signals are. The matrix is determined from circuit inputs having "good" fault detection properties. Auto-correlation was used on tests for a sequential digital circuit. Random vectors were generated, and only those detecting faults were retained during vector compaction. The correlation matrix (generated from the vectors) was used to create additional test vectors. The results were spectacular -- more faults were detected, using significantly fewer vectors, and with less computation, than by any other known method. The shorter vector length significantly reduces testing costs, which are typically one third of integrated circuit costs. We apply this idea to analog and mixed-signal circuits. We compose the Hadamard matrix (which describes the time history of prior vectors) for a digital circuit with an analog circuit transfer function. The matrix expresses successful digital test waveforms in terms of their digital spectrum. Analog circuits are tested using spectral analysis, but this is the first time that digital circuits have been tested spectrally. In a mixed digital/analog system, we arrange digital test waveforms to configure the digital part as a "programmable tone generator" to test the analog part. Likewise, the analog part is configured to provide the appropriate signal spectrum to test the digital part. The possible benefits would be: (1) Removal of isolation test hardware between digital and analog circuits, which are now tested together; (2) Elimination of delays and distortion due to test hardware; (3) Lower cost; and (4) Simpler test generation algorithms. We also are applying spectral testing ideas to formal hardware verification, which determines whether a circuit, as implemented, is consistent with its specification.