Resumen de Non-functional considerations of time-randomized processor architectures
Critical Real-Time Embedded Systems (CRTES) are the subset of embedded systems with timing constraints whose miss behavior can endanger human lives or expensive equipment. To provide evidence of correctness, CRTES are designed, implemented and deployed in adherence to safety standards and certification regulations. To that end, CRTES follow strict Validation & Verification (V&V) procedures of their functional and non-functional properties. One of the most important non-functional properties is timing, which builds on computing the worst-case execution time of tasks and a schedule of tasks so that the overall system timing behavior is correct. However, the use of more complex hardware and software to satisfy CRTES unprecedented performance requirements, heavily increase the cost of V&V. For timing V&V, statistical techniques, like Measurement-Based Probabilistic Timing Analysis (MBPTA) help to address the complexity of hardware and software in CRTES. To that end, they benefit from randomization of temporal behavior at the hardware level. In this line, Time-Randomized Processors (TRP) contain timing V&V costs by breaking systematic pathological behaviors and enabling MBPTA applicability. In the context of TRP, this thesis shows that hardware and software designs incorporating randomization can not only successfully tackle the existing timing analysis problem, but also provide helpful properties to other emerging non-functional metrics key in CRTES like reliability, security and energy. For reliability, we show that TRP are naturally resilient against hardware aging effects and voltage noise and we add up to such resilience by improving its design. Also, TRP hinders security threats and intrusions by breaking and mangling the deterministic association between memory mapping and access time and we develop a framework for secure automotive operation. Finally for energy, we introduce a taxonomy to guide the future challenges for worst-case energy estimation and make the first steps towards the use of MBPTA-like methodology to address worst-case energy estimation under the presence of process variation. Moreover this thesis also shows that together with the application of MBPTA-like methodology, TRP also naturally expose and break pathological energy consumption patterns and help in validating and accounting instantaneous peak power demands. In summary, this thesis pioneers several aspects of the use of TRP to address the emerging challenges that CRTES face in the reliability, security and energy domains.
