Software Framework For Runtime Application Monitoring Of Fail-Safe Multi-Processor Adas Socs

In Advanced Driver Assistance (ADAS) systems, it is very critical to ensure a system malfunction does not cause harm to humans (also referred to as functional safety). Malfunctions could be due to incorrect software (SW) execution. Many techniques exist to catch systematic SW faults during development. However without run-time monitoring, it is impossible to guarantee a fail-safe system. In this paper, we describe a framework to monitor application statistics in heterogeneous multiprocessor SoCs targeted for fail-safe ADAS systems. With < 0.5% CPU load overhead per monitor, the proposed framework was found to accurately report statistics for a front camera analytics application.