Characterization and handling of low-cost micro-architectural signatures in MPSoCs

In recent years the wide spread introduction of small embedded systems into every corner of everyday life lead to the strong need for highly reliable and secure computing machines. These machines now affect the safety of humans as well as the security of personal data and consequently money transactions. To ensure the integrity of these systems' operating state, several fault detection mechanisms have been developed to safely correct or stop unforeseen execution behavior. Because of the rise of battery or even field-supplied systems these mechanisms often heavily decrease available power budgets or lead to significantly increased production costs. Therefore, this paper introduces novel micro-architectural execution signature characterization and handling techniques for system-on-chip designs providing power estimation hardware. Existing power sensor infrastructure is reused to enable efficient system-state monitoring using micro-architectural hashes to cover a wide range of implemented system functionality. Reduced hashing implementations are characterized for their fault detection efficiency. This hardware-based approach provides a completely transparent solution to counteract faults resulting from emerging wearout defects or intentional attacks on the execution integrity.