AMS-Miner: Mining AMS Assertions Using Interval Arithmetic

We present AMS-Miner, a methodology for generating Analog Mixed-Signal (AMS) Assertions automatically. The proposed methodology uses predicates over real variables (PORVs) to map real-valued signals obtained from timestamped traces of the AMS design to Boolean signals. Dense time for AMS designs is dealt as intervals of time. This mapping yields a set of time intervals wherein each PORV is true. Interval arithmetic coupled with decision tree learning are used to operate on time-intervals in the AMS-data space to extract assertions in the form of cause-effect patterns, as timed sequences of PORVs. This article, for the first time explores assertion mining for AMS and builds the theory for interval based assertion mining for AMS traces.