SecX: A Framework for Collecting Runtime Statistics for SoCs with Multiple Accelerators

We are moving into an era where large SoCs will have a portfolio of different kinds of cores and accelerators. Many of these computational elements might be designed by third parties. In this setting, it is beneficial to collect accurate runtime information such that we can diagnose performance problems, verify and report correctness issues, and collect usage scenarios of third party hardware. This problem is non-trivial if we consider the possibility of defective or malicious elements in the chip. We design an architecture, SecX, which helps us collect various metrics of potential interest in a fair, reliable and secure fashion. These logs can subsequently be made available to users, IP vendors, and the system integrator through a trusted third party called the auditor. The performance (0.03%), power (1.04W) and area (0.32%) overheads of our scheme are minimal.