Cone of influence analysis at the electronic system level using machine learning

Cone of influence analysis, i.e. determining the parts of the circuit which are relevant to a considered circuit signal, is an established methodology applied in several design tasks. In abstractions like the Register Transfer Level (RTL) or the gate level, cone of influence analysis is simple. However, the introduction of higher levels of abstractions, particularly the Electronic System Level (ESL), made it significantly harder to reliably extract a cone of influence. In this paper, we propose a methodology that enables cone of influence analysis at the ESL. Instead of a structural analysis, a behavioral scheme is proposed, i.e. stimuli representing different system executions are analyzed. To this end, machine learning techniques are exploited. This enables a very good approximation of the desired cone of influence which is non-invasive, does not rely on the availability of the source code, and performs fast. Case studies confirm the applicability of the proposed approach.